Merge remote-tracking branch 'origin/main' into alexcheema/speculative-decoding

Add draft_model and num_draft_tokens fields to PlaceInstance command
PlaceInstance was missing these fields that are accessed in placement.py when creating MlxJaccl and MlxRing instances with speculative decoding support, causing type errors after merging main. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-18 23:06:23 -05:00 · 2026-02-13 09:39:11 -08:00 · 2026-02-13 05:51:11 -08:00 · 2026-02-13 05:46:52 -08:00 · 2026-02-05 06:09:50 -08:00 · 2026-02-05 06:07:51 -08:00
99 changed files with 1776 additions and 2530 deletions
--- a/.github/workflows/pipeline.yml
+++ b/.github/workflows/pipeline.yml
@@ -8,6 +8,33 @@ on:
      - main

 jobs:
+  typecheck:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          lfs: false
+
+      - uses: cachix/install-nix-action@v31
+        with:
+          nix_path: nixpkgs=channel:nixos-unstable
+
+      - uses: cachix/cachix-action@v14
+        name: Configure Cachix
+        with:
+          name: exo
+          authToken: "${{ secrets.CACHIX_AUTH_TOKEN }}"
+
+      - name: Load nix develop environment
+        run: nix run github:nicknovitski/nix-develop/v1
+
+      - name: Sync dependencies
+        run: uv sync --all-packages
+
+      - name: Run type checker
+        run: uv run basedpyright --project pyproject.toml
+
  nix:
    name: Build and check (${{ matrix.system }})
    runs-on: ${{ matrix.runner }}
--- a/.mlx_typings/mlx_lm/models/glm_moe_dsa.pyi
+++ b/.mlx_typings/mlx_lm/models/glm_moe_dsa.pyi
@@ -1,46 +0,0 @@
-"""Type stubs for mlx_lm.models.glm_moe_dsa"""
-
-from dataclasses import dataclass
-from typing import Any, Dict, Optional
-
-from .base import BaseModelArgs
-from .deepseek_v32 import Model as DSV32Model
-
-@dataclass
-class ModelArgs(BaseModelArgs):
-    model_type: str
-    vocab_size: int
-    hidden_size: int
-    index_head_dim: int
-    index_n_heads: int
-    index_topk: int
-    intermediate_size: int
-    moe_intermediate_size: int
-    num_hidden_layers: int
-    num_attention_heads: int
-    num_key_value_heads: int
-    n_shared_experts: Optional[int]
-    n_routed_experts: Optional[int]
-    routed_scaling_factor: float
-    kv_lora_rank: int
-    q_lora_rank: int
-    qk_rope_head_dim: int
-    v_head_dim: int
-    qk_nope_head_dim: int
-    topk_method: str
-    scoring_func: str
-    norm_topk_prob: bool
-    n_group: int
-    topk_group: int
-    num_experts_per_tok: int
-    moe_layer_freq: int
-    first_k_dense_replace: int
-    max_position_embeddings: int
-    rms_norm_eps: float
-    rope_parameters: Dict[str, Any]
-    attention_bias: bool
-    rope_scaling: Dict[str, Any] | None
-    rope_theta: float | None
-
-class Model(DSV32Model):
-    def __init__(self, config: ModelArgs) -> None: ...
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -141,6 +141,12 @@ version = "0.3.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "76a2e8124351fda1ef8aaaa3bbd7ebbcb486bbcd4225aca0aa0d84bb2db8fecb"

+[[package]]
+name = "arrayvec"
+version = "0.7.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7c02d123df017efcdfbd739ef81735b36c5ba83ec3c59c80a9d7ecc718f92e50"
+
 [[package]]
 name = "asn1-rs"
 version = "0.7.1"
@@ -298,6 +304,19 @@ version = "1.8.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "55248b47b0caf0546f7988906588779981c43bb1bc9d0c44087278f80cdb44ba"

+[[package]]
+name = "bigdecimal"
+version = "0.4.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "560f42649de9fa436b73517378a147ec21f6c997a546581df4b4b31677828934"
+dependencies = [
+ "autocfg",
+ "libm",
+ "num-bigint",
+ "num-integer",
+ "num-traits",
+]
+
 [[package]]
 name = "bimap"
 version = "0.6.3"
@@ -497,6 +516,15 @@ version = "0.4.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "2f421161cb492475f1661ddc9815a745a1c894592070661180fdec3d4872e9c3"

+[[package]]
+name = "convert_case"
+version = "0.10.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "633458d4ef8c78b72454de2d54fd6ab2e60f9e02be22f3c6104cdc8a4e0fceb9"
+dependencies = [
+ "unicode-segmentation",
+]
+
 [[package]]
 name = "core-foundation"
 version = "0.9.4"
@@ -718,6 +746,29 @@ dependencies = [
 "powerfmt",
 ]

+[[package]]
+name = "derive_more"
+version = "2.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "10b768e943bed7bf2cab53df09f4bc34bfd217cdb57d971e769874c9a6710618"
+dependencies = [
+ "derive_more-impl",
+]
+
+[[package]]
+name = "derive_more-impl"
+version = "2.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6d286bfdaf75e988b4a78e013ecd79c581e06399ab53fbacd2d916c2f904f30b"
+dependencies = [
+ "convert_case",
+ "proc-macro2",
+ "quote",
+ "rustc_version",
+ "syn 2.0.111",
+ "unicode-xid",
+]
+
 [[package]]
 name = "digest"
 version = "0.10.7"
@@ -888,17 +939,22 @@ name = "exo_pyo3_bindings"
 version = "0.0.1"
 dependencies = [
 "delegate",
+ "derive_more",
 "env_logger",
 "extend",
 "futures",
+ "impl-trait-for-tuples",
 "libp2p",
 "log",
 "networking",
+ "once_cell",
 "pin-project",
 "pyo3",
 "pyo3-async-runtimes",
 "pyo3-log",
 "pyo3-stub-gen",
+ "thiserror 2.0.17",
+ "thread_local",
 "tokio",
 "util",
 ]
@@ -1584,6 +1640,17 @@ dependencies = [
 "xmltree",
 ]

+[[package]]
+name = "impl-trait-for-tuples"
+version = "0.2.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a0eb5a3343abf848c0984fe4604b2b105da9539376e24fc0a3b0007411ae4fd9"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.111",
+]
+
 [[package]]
 name = "indexmap"
 version = "2.12.1"
@@ -1762,6 +1829,12 @@ version = "0.2.178"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "37c93d8daa9d8a012fd8ab92f088405fb202ea0b6ab73ee2482ae66af4f42091"

+[[package]]
+name = "libm"
+version = "0.2.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f9fbbcab51052fe104eb5e5d351cf728d30a5be1fe14d9be8a3b097481fb97de"
+
 [[package]]
 name = "libp2p"
 version = "0.56.0"
@@ -2751,13 +2824,16 @@ name = "networking"
 version = "0.0.1"
 dependencies = [
 "delegate",
+ "derive_more",
 "either",
 "extend",
 "futures",
 "futures-timer",
+ "impl-trait-for-tuples",
 "keccak-const",
 "libp2p",
 "log",
+ "thiserror 2.0.17",
 "tokio",
 "tracing-subscriber",
 "util",
@@ -2842,6 +2918,17 @@ dependencies = [
 "num-traits",
 ]

+[[package]]
+name = "num-rational"
+version = "0.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f83d14da390562dca69fc84082e73e548e1ad308d24accdedd2720017cb37824"
+dependencies = [
+ "num-bigint",
+ "num-integer",
+ "num-traits",
+]
+
 [[package]]
 name = "num-traits"
 version = "0.2.19"
@@ -3192,14 +3279,28 @@ version = "0.27.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "ab53c047fcd1a1d2a8820fe84f05d6be69e9526be40cb03b73f86b6b03e6d87d"
 dependencies = [
+ "bigdecimal",
+ "either",
+ "hashbrown 0.16.1",
+ "indexmap",
 "indoc",
+ "inventory",
 "libc",
+ "lock_api",
 "memoffset",
+ "num-bigint",
+ "num-complex",
+ "num-rational",
+ "num-traits",
 "once_cell",
+ "ordered-float",
+ "parking_lot",
 "portable-atomic",
 "pyo3-build-config",
 "pyo3-ffi",
 "pyo3-macros",
+ "rust_decimal",
+ "smallvec",
 "unindent",
 ]

@@ -3640,6 +3741,16 @@ dependencies = [
 "tokio",
 ]

+[[package]]
+name = "rust_decimal"
+version = "1.39.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "35affe401787a9bd846712274d97654355d21b2a2c092a3139aabe31e9022282"
+dependencies = [
+ "arrayvec",
+ "num-traits",
+]
+
 [[package]]
 name = "rustc-hash"
 version = "1.1.0"
@@ -4504,12 +4615,24 @@ version = "1.0.22"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9312f7c4f6ff9069b165498234ce8be658059c6728633667c526e27dc2cf1df5"

+[[package]]
+name = "unicode-segmentation"
+version = "1.12.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f6ccf251212114b54433ec949fd6a7841275f9ada20dddd2f29e9ceea4501493"
+
 [[package]]
 name = "unicode-width"
 version = "0.2.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b4ac048d71ede7ee76d585517add45da530660ef4390e49b098733c6e897f254"

+[[package]]
+name = "unicode-xid"
+version = "0.2.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ebc1c04c71510c7f702b52b7c350734c9ff1295c464a03335b00bb84fc54f853"
+
 [[package]]
 name = "unicode_names2"
 version = "1.3.0"
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -26,21 +26,49 @@ opt-level = 3
 networking = { path = "rust/networking" }
 util = { path = "rust/util" }

+# Proc-macro authoring tools
+syn = "2.0"
+quote = "1.0"
+proc-macro2 = "1.0"
+darling = "0.20"
+
 # Macro dependecies
 extend = "1.2"
 delegate = "0.13"
+impl-trait-for-tuples = "0.2"
+clap = "4.5"
+derive_more = { version = "2.0.1", features = ["display"] }
 pin-project = "1"

 # Utility dependencies
+itertools = "0.14"
+thiserror = "2"
+internment = "0.8"
+recursion = "0.5"
+regex = "1.11"
+once_cell = "1.21"
+thread_local = "1.1"
+bon = "3.4"
+generativity = "1.1"
+anyhow = "1.0"
 keccak-const = "0.2"

+# Functional generics/lenses frameworks
+frunk_core = "0.4"
+frunk = "0.4"
+frunk_utils = "0.2"
+frunk-enum-core = "0.3"
+
 # Async dependencies
 tokio = "1.46"
 futures = "0.3"
+futures-util = "0.3"
 futures-timer = "3.0"

 # Data structures
 either = "1.15"
+ordered-float = "5.0"
+ahash = "0.8"

 # Tracing/logging
 log = "0.4"
--- a/MISSED_THINGS.md
+++ b/MISSED_THINGS.md
@@ -5,21 +5,21 @@
 [X] Fetching download status of all models on start
 [X] Deduplication of tasks in plan_step.
 [X] resolve_allow_patterns should just be wildcard now.
-[X] no mx_barrier in genreate.py mlx_generate at the end.
+[] no mx_barrier in genreate.py mlx_generate at the end.
 [] cache assertion not needed in auto_parallel.py PipelineLastLayer.
-[X] GPTOSS support dropped in auto_parallel.py.
-[X] sharding changed "all-to-sharded" became _all_to_sharded in auto_parallel.py.
-[X] same as above with "sharded-to-all" became _sharded_to_all in auto_parallel.py.
-[X] Dropped support for Ministral3Model, DeepseekV32Model, Glm4MoeModel, Qwen3NextModel, GptOssMode in auto_parallel.py.
+[] GPTOSS support dropped in auto_parallel.py.
+[] sharding changed "all-to-sharded" became _all_to_sharded in auto_parallel.py.
+[] same as above with "sharded-to-all" became _sharded_to_all in auto_parallel.py.
+[] Dropped support for Ministral3Model, DeepseekV32Model, Glm4MoeModel, Qwen3NextModel, GptOssMode in auto_parallel.py.
 [] Dropped prefill/decode code in auto_parallel.py and utils_mlx.py.
 [X] KV_CACHE_BITS should be None to disable quantized KV cache.
-[X] Dropped _set_nofile_limit in utils_mlx.py.
-[X] We have group optional in load_mlx_items in utils_mlx.py.
-[X] Dropped add_missing_chat_templates for GptOss in load_mlx_items in utils_mlx.py.
-[X] Dropped model.make_cache in make_kv_cache in utils_mlx.py.
+[] Dropped _set_nofile_limit in utils_mlx.py.
+[] We have group optional in load_mlx_items in utils_mlx.py.
+[] Dropped add_missing_chat_templates for GptOss in load_mlx_items in utils_mlx.py.
+[] Dropped model.make_cache in make_kv_cache in utils_mlx.py.
 [X] We put cache limit back in utils_mlx.py.
-[X] topology.py remove_node removes the connections after checking if node is is in self._node_id_to_rx_id_map. on beta_1 it checks after, so would remove stale connections I guess?
-[X] Missing Glm 4.7 model cards (this isn't ready yet but should be picked up, probably create an issue... the blocker is transforemrs version doesn't support the tokenizer for Glm 4.7. rc-1 does but we can't upgrade as it breaks other things.)
+[] topology.py remove_node removes the connections after checking if node is is in self._node_id_to_rx_id_map. on beta_1 it checks after, so would remove stale connections I guess?
+[] Missing Glm 4.7 model cards (this isn't ready yet but should be picked up, probably create an issue... the blocker is transforemrs version doesn't support the tokenizer for Glm 4.7. rc-1 does but we can't upgrade as it breaks other things.)
 [] try-except in _command_processor only excepts ValueError. This was silently failing leading to un-debuggable errors (we had a KeyError that was happening ). Changed this to catch Exception instead of ValueError. See exo-v2 89ae38405e0052e3c22405daf094b065878aa873 and fb99fea69b5a39017efc90c5dad0072e677455f0.
 [X] In placement.py, place_instance no longer looks at model_meta.supports_tensor and check if this tensor parallel number of nodes is supported by the model's tensor dimensions.
 [X] In placement.py, place_instanec, we no longer have the special case to exclude DeepSeek v3.1 pipeline parallel (it doesn't work).
--- a/README.md
+++ b/README.md
@@ -72,23 +72,16 @@ There are two ways to run exo:

 ### Run from Source (macOS)

-If you have [Nix](https://nixos.org/) installed, you can skip most of the steps below and run exo directly (after accepting the Cachix cache):
-
-```bash
-nix run .#exo
-```
-
 **Prerequisites:**
- [Xcode](https://developer.apple.com/xcode/) (provides the Metal ToolChain required for MLX compilation)
 - [brew](https://github.com/Homebrew/brew) (for simple package management on macOS)
-
+  
  ```bash
  /bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
  ```
 - [uv](https://github.com/astral-sh/uv) (for Python dependency management)
 - [macmon](https://github.com/vladkens/macmon) (for hardware monitoring on Apple Silicon)
 - [node](https://github.com/nodejs/node) (for building the dashboard)
-
+  
  ```bash
  brew install uv macmon node
  ```
--- a/app/EXO/EXO/ExoProcessController.swift
+++ b/app/EXO/EXO/ExoProcessController.swift
@@ -126,37 +126,11 @@ final class ExoProcessController: ObservableObject {
            return
        }
        process.terminationHandler = nil
-        status = .stopped
-
-        guard process.isRunning else {
-            self.process = nil
-            return
+        if process.isRunning {
+            process.terminate()
        }
-
-        let proc = process
        self.process = nil
-
-        Task.detached {
-            proc.interrupt()
-
-            for _ in 0..<50 {
-                if !proc.isRunning { return }
-                try? await Task.sleep(nanoseconds: 100_000_000)
-            }
-
-            if proc.isRunning {
-                proc.terminate()
-            }
-
-            for _ in 0..<30 {
-                if !proc.isRunning { return }
-                try? await Task.sleep(nanoseconds: 100_000_000)
-            }
-
-            if proc.isRunning {
-                kill(proc.processIdentifier, SIGKILL)
-            }
-        }
+        status = .stopped
    }

    func restart() {
--- a/bench/bench.toml
+++ b/bench/bench.toml
@@ -1,7 +0,0 @@
-# Canary benchmark manifest
-#
-# Lists the suite files to include. Each file defines benchmarks
-# with shared constraints, topology, and default args.
-include = [
-    "single-m3-ultra.toml",
-]
--- a/bench/exo_bench.py
+++ b/bench/exo_bench.py
@@ -288,151 +288,6 @@ def resolve_model_short_id(client: ExoClient, model_arg: str) -> tuple[str, str]
    raise ValueError(f"Model not found in /models: {model_arg}")


-def run_planning_phase(
-    client: ExoClient,
-    full_model_id: str,
-    preview: dict[str, Any],
-    danger_delete: bool,
-    timeout: float,
-    settle_deadline: float | None,
-) -> None:
-    """Check disk space and ensure model is downloaded before benchmarking."""
-    # Get model size from /models
-    models = client.request_json("GET", "/models") or {}
-    model_bytes = 0
-    for m in models.get("data", []):
-        if m.get("hugging_face_id") == full_model_id:
-            model_bytes = m.get("storage_size_megabytes", 0) * 1024 * 1024
-            break
-
-    if not model_bytes:
-        logger.warning(
-            f"Could not determine size for {full_model_id}, skipping disk check"
-        )
-        return
-
-    # Get nodes from preview
-    inner = unwrap_instance(preview["instance"])
-    node_ids = list(inner["shardAssignments"]["nodeToRunner"].keys())
-    runner_to_shard = inner["shardAssignments"]["runnerToShard"]
-
-    state = client.request_json("GET", "/state")
-    downloads = state.get("downloads", {})
-    node_disk = state.get("nodeDisk", {})
-
-    for node_id in node_ids:
-        node_downloads = downloads.get(node_id, [])
-
-        # Check if model already downloaded on this node
-        already_downloaded = any(
-            "DownloadCompleted" in p
-            and unwrap_instance(p["DownloadCompleted"]["shardMetadata"])["modelCard"][
-                "modelId"
-            ]
-            == full_model_id
-            for p in node_downloads
-        )
-        if already_downloaded:
-            continue
-
-        # Wait for disk info if settle_deadline is set
-        disk_info = node_disk.get(node_id, {})
-        backoff = _SETTLE_INITIAL_BACKOFF_S
-        while not disk_info and settle_deadline and time.monotonic() < settle_deadline:
-            remaining = settle_deadline - time.monotonic()
-            logger.info(
-                f"Waiting for disk info on {node_id} ({remaining:.0f}s remaining)..."
-            )
-            time.sleep(min(backoff, remaining))
-            backoff = min(backoff * _SETTLE_BACKOFF_MULTIPLIER, _SETTLE_MAX_BACKOFF_S)
-            state = client.request_json("GET", "/state")
-            node_disk = state.get("nodeDisk", {})
-            disk_info = node_disk.get(node_id, {})
-
-        if not disk_info:
-            logger.warning(f"No disk info for {node_id}, skipping space check")
-            continue
-
-        avail = disk_info.get("available", {}).get("inBytes", 0)
-        if avail >= model_bytes:
-            continue
-
-        if not danger_delete:
-            raise RuntimeError(
-                f"Insufficient disk on {node_id}: need {model_bytes // (1024**3)}GB, "
-                f"have {avail // (1024**3)}GB. Use --danger-delete-downloads to free space."
-            )
-
-        # Delete from smallest to largest
-        completed = [
-            (
-                unwrap_instance(p["DownloadCompleted"]["shardMetadata"])["modelCard"][
-                    "modelId"
-                ],
-                p["DownloadCompleted"]["totalBytes"]["inBytes"],
-            )
-            for p in node_downloads
-            if "DownloadCompleted" in p
-        ]
-        for del_model, size in sorted(completed, key=lambda x: x[1]):
-            logger.info(f"Deleting {del_model} from {node_id} ({size // (1024**2)}MB)")
-            client.request_json("DELETE", f"/download/{node_id}/{del_model}")
-            avail += size
-            if avail >= model_bytes:
-                break
-
-        if avail < model_bytes:
-            raise RuntimeError(f"Could not free enough space on {node_id}")
-
-    # Start downloads (idempotent)
-    for node_id in node_ids:
-        runner_id = inner["shardAssignments"]["nodeToRunner"][node_id]
-        shard = runner_to_shard[runner_id]
-        client.request_json(
-            "POST",
-            "/download/start",
-            body={
-                "targetNodeId": node_id,
-                "shardMetadata": shard,
-            },
-        )
-        logger.info(f"Started download on {node_id}")
-
-    # Wait for downloads
-    start = time.time()
-    while time.time() - start < timeout:
-        state = client.request_json("GET", "/state")
-        downloads = state.get("downloads", {})
-        all_done = True
-        for node_id in node_ids:
-            done = any(
-                "DownloadCompleted" in p
-                and unwrap_instance(p["DownloadCompleted"]["shardMetadata"])[
-                    "modelCard"
-                ]["modelId"]
-                == full_model_id
-                for p in downloads.get(node_id, [])
-            )
-            failed = [
-                p["DownloadFailed"]["errorMessage"]
-                for p in downloads.get(node_id, [])
-                if "DownloadFailed" in p
-                and unwrap_instance(p["DownloadFailed"]["shardMetadata"])["modelCard"][
-                    "modelId"
-                ]
-                == full_model_id
-            ]
-            if failed:
-                raise RuntimeError(f"Download failed on {node_id}: {failed[0]}")
-            if not done:
-                all_done = False
-        if all_done:
-            return
-        time.sleep(1)
-
-    raise TimeoutError("Downloads did not complete in time")
-
-
 def placement_filter(instance_meta: str, wanted: str) -> bool:
    s = (instance_meta or "").lower()
    if wanted == "both":
@@ -680,11 +535,6 @@ def main() -> int:
        default=0,
        help="Max seconds to wait for the cluster to produce valid placements (0 = try once).",
    )
-    ap.add_argument(
-        "--danger-delete-downloads",
-        action="store_true",
-        help="Delete existing models from smallest to largest to make room for benchmark model.",
-    )
    args = ap.parse_args()

    pp_list = parse_int_list(args.pp)
@@ -719,16 +569,13 @@ def main() -> int:
        logger.error("[exo-bench] tokenizer usable but prompt sizing failed")
        raise

-    settle_deadline = (
-        time.monotonic() + args.settle_timeout if args.settle_timeout > 0 else None
-    )
-
    selected = fetch_and_filter_placements(client, full_model_id, args)

-    if not selected and settle_deadline:
+    if not selected and args.settle_timeout > 0:
        backoff = _SETTLE_INITIAL_BACKOFF_S
-        while not selected and time.monotonic() < settle_deadline:
-            remaining = settle_deadline - time.monotonic()
+        deadline = time.monotonic() + args.settle_timeout
+        while not selected and time.monotonic() < deadline:
+            remaining = deadline - time.monotonic()
            logger.warning(
                f"No valid placements yet (cluster may still be settling). "
                f"Retrying in {backoff:.1f}s ({remaining:.0f}s remaining)..."
@@ -760,16 +607,6 @@ def main() -> int:
    if args.dry_run:
        return 0

-    logger.info("Planning phase: checking downloads...")
-    run_planning_phase(
-        client,
-        full_model_id,
-        selected[0],
-        args.danger_delete_downloads,
-        args.timeout,
-        settle_deadline,
-    )
-
    all_rows: list[dict[str, Any]] = []

    for preview in selected:
--- a/bench/single-m3-ultra.toml
+++ b/bench/single-m3-ultra.toml
@@ -1,189 +0,0 @@
-# Single-node M3 Ultra benchmarks
-#
-# Shared constraints applied to ALL benchmarks in this file.
-constraints = [
-    "All(MacOsBuild(=25D125))",
-    "Hosts(=1)",
-    "All(Chip(m3_ultra))",
-    "All(GpuCores(=80))",
-]
-
-[topology]
-type = "none"
-
-# Default args merged into each benchmark's args (benchmark-level args win).
-[defaults]
-pp = [512, 2048, 8192, 16384]
-tg = 128
-
-[[benchmark]]
-model = "mlx-community/Meta-Llama-3.1-70B-Instruct-4bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/gpt-oss-120b-MXFP4-Q8"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/GLM-4.7-Flash-8bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-Coder-Next-6bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-30B-A3B-8bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-0.6B-4bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-0.6B-8bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Llama-3.2-1B-Instruct-4bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Llama-3.2-3B-Instruct-4bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Llama-3.2-3B-Instruct-8bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Meta-Llama-3.1-8B-Instruct-4bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Meta-Llama-3.1-8B-Instruct-8bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Meta-Llama-3.1-8B-Instruct-bf16"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/gpt-oss-20b-MXFP4-Q8"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-30B-A3B-4bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/GLM-4.7-Flash-4bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/GLM-4.7-Flash-5bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/GLM-4.7-Flash-6bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Llama-3.3-70B-Instruct-4bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-Coder-Next-4bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-Coder-Next-5bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-Coder-Next-8bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-Next-80B-A3B-Instruct-4bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-Next-80B-A3B-Instruct-8bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-Next-80B-A3B-Thinking-4bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-Next-80B-A3B-Thinking-8bit"
-extra_constraints = ["All(Memory(>=96GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Llama-3.3-70B-Instruct-8bit"
-extra_constraints = ["All(Memory(>=256GiB))"]
-
-[[benchmark]]
-model = "mlx-community/llama-3.3-70b-instruct-fp16"
-extra_constraints = ["All(Memory(>=256GiB))"]
-
-[[benchmark]]
-model = "mlx-community/GLM-4.5-Air-8bit"
-extra_constraints = ["All(Memory(>=256GiB))"]
-
-[[benchmark]]
-model = "mlx-community/GLM-4.5-Air-bf16"
-extra_constraints = ["All(Memory(>=256GiB))"]
-
-[[benchmark]]
-model = "mlx-community/GLM-4.7-4bit"
-extra_constraints = ["All(Memory(>=256GiB))"]
-
-[[benchmark]]
-model = "mlx-community/MiniMax-M2.1-3bit"
-extra_constraints = ["All(Memory(>=256GiB))"]
-
-[[benchmark]]
-model = "mlx-community/MiniMax-M2.1-8bit"
-extra_constraints = ["All(Memory(>=256GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-235B-A22B-Instruct-2507-4bit"
-extra_constraints = ["All(Memory(>=256GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-Coder-Next-bf16"
-extra_constraints = ["All(Memory(>=256GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Step-3.5-Flash-4bit"
-extra_constraints = ["All(Memory(>=256GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Step-3.5-Flash-6bit"
-extra_constraints = ["All(Memory(>=256GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Step-3.5-Flash-8Bit"
-extra_constraints = ["All(Memory(>=256GiB))"]
-
-[[benchmark]]
-model = "mlx-community/DeepSeek-V3.1-4bit"
-extra_constraints = ["All(Memory(>=512GiB))"]
-
-[[benchmark]]
-model = "mlx-community/GLM-4.7-6bit"
-extra_constraints = ["All(Memory(>=512GiB))"]
-
-[[benchmark]]
-model = "mlx-community/GLM-4.7-8bit-gs32"
-extra_constraints = ["All(Memory(>=512GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-235B-A22B-Instruct-2507-8bit"
-extra_constraints = ["All(Memory(>=512GiB))"]
-
-[[benchmark]]
-model = "mlx-community/Qwen3-Coder-480B-A35B-Instruct-4bit"
-extra_constraints = ["All(Memory(>=512GiB))"]
--- a/dashboard/src/app.css
+++ b/dashboard/src/app.css
@@ -16,10 +16,9 @@
 	/* Gotham-inspired accent colors */
 	--exo-grid: oklch(0.25 0 0);
 	--exo-scanline: oklch(0.15 0 0);
-	--exo-glow-yellow: oklch(0.85 0.18 85 / 0.3);
-	--exo-glow-yellow-strong: oklch(0.85 0.18 85 / 0.5);
-	--exo-bg-hover: oklch(0.18 0 0);
-
+	--exo-glow-yellow: 0 0 20px oklch(0.85 0.18 85 / 0.3);
+	--exo-glow-yellow-strong: 0 0 40px oklch(0.85 0.18 85 / 0.5);
+	
 	/* Theme Variables */
 	--radius: 0.375rem;
 	--background: var(--exo-black);
@@ -42,237 +41,6 @@
 	--ring: var(--exo-yellow);
 }

-/* ============================================================
-   LIGHT THEME — "Mission Control, Dawn Shift"
-   Warm parchment + deep amber. Applied when <html> has .light class.
-   ============================================================ */
-html.light {
-	/* EXO brand palette — warm amber shift */
-	--exo-black: oklch(0.97 0.015 80);
-	--exo-dark-gray: oklch(0.92 0.012 80);
-	--exo-medium-gray: oklch(0.83 0.009 78);
-	--exo-light-gray: oklch(0.50 0.018 75);
-	--exo-yellow: oklch(0.50 0.14 65);
-	--exo-yellow-darker: oklch(0.40 0.13 65);
-	--exo-yellow-glow: oklch(0.60 0.14 65);
-
-	--exo-grid: oklch(0.88 0.009 80);
-	--exo-scanline: oklch(0.93 0.010 80);
-	--exo-glow-yellow: oklch(0.50 0.14 65 / 0.12);
-	--exo-glow-yellow-strong: oklch(0.50 0.14 65 / 0.22);
-	--exo-bg-hover: oklch(0.89 0.010 80);
-
-	/* Semantic tokens */
-	--background: oklch(0.97 0.015 80);
-	--foreground: oklch(0.13 0.015 75);
-	--card: oklch(0.92 0.012 80);
-	--card-foreground: oklch(0.13 0.015 75);
-	--popover: oklch(0.95 0.012 80);
-	--popover-foreground: oklch(0.13 0.015 75);
-	--primary: oklch(0.50 0.14 65);
-	--primary-foreground: oklch(0.97 0.015 80);
-	--secondary: oklch(0.88 0.008 80);
-	--secondary-foreground: oklch(0.15 0.012 75);
-	--muted: oklch(0.90 0.009 80);
-	--muted-foreground: oklch(0.50 0.018 75);
-	--accent: oklch(0.88 0.008 80);
-	--accent-foreground: oklch(0.15 0.012 75);
-	--destructive: oklch(0.52 0.22 25);
-	--border: oklch(0.84 0.007 78);
-	--input: oklch(0.87 0.008 80);
-	--ring: oklch(0.50 0.14 65);
-}
-
-/* ============================================================
-   LIGHT MODE UTILITY OVERRIDES
-   ============================================================ */
-html.light {
-	& .text-white,
-	& .text-white\/90,
-	& .text-white\/80,
-	& .text-white\/70 {
-		color: var(--foreground) !important;
-	}
-	& .text-white\/60,
-	& .text-white\/50 {
-		color: color-mix(in oklch, var(--foreground) 60%, transparent) !important;
-	}
-	& .text-white\/40,
-	& .text-white\/30 {
-		color: color-mix(in oklch, var(--foreground) 38%, transparent) !important;
-	}
-
-	& .bg-black\/80,
-	& .bg-black\/60,
-	& .bg-black\/50,
-	& .bg-black\/40 {
-		background-color: oklch(0.90 0.010 80 / 0.7) !important;
-	}
-	& [class*="bg-exo-black/"] {
-		background-color: oklch(0.90 0.010 80 / 0.6) !important;
-	}
-	& [class*="shadow-black"] {
-		--tw-shadow-color: oklch(0.30 0.010 75 / 0.10) !important;
-	}
-
-	& ::-webkit-scrollbar-track {
-		background: oklch(0.93 0.010 80) !important;
-	}
-	& ::-webkit-scrollbar-thumb {
-		background: oklch(0.76 0.010 78) !important;
-	}
-	& ::-webkit-scrollbar-thumb:hover {
-		background: oklch(0.50 0.14 65 / 0.6) !important;
-	}
-
-	& .command-panel {
-		background: linear-gradient(
-			180deg,
-			oklch(0.94 0.012 80 / 0.96) 0%,
-			oklch(0.91 0.010 80 / 0.98) 100%
-		) !important;
-		border-color: oklch(0.82 0.008 78) !important;
-		box-shadow:
-			inset 0 1px 0 oklch(1 0 0 / 0.6),
-			0 4px 20px oklch(0.30 0.010 75 / 0.08) !important;
-	}
-
-	& .glow-text {
-		text-shadow:
-			0 0 12px oklch(0.50 0.14 65 / 0.20),
-			0 1px 3px oklch(0.30 0.010 75 / 0.12) !important;
-	}
-
-	& .grid-bg {
-		background-image:
-			linear-gradient(oklch(0.75 0.008 78 / 0.25) 1px, transparent 1px),
-			linear-gradient(90deg, oklch(0.75 0.008 78 / 0.25) 1px, transparent 1px) !important;
-	}
-
-	& .scanlines::before {
-		background: repeating-linear-gradient(
-			0deg,
-			transparent,
-			transparent 2px,
-			oklch(0.50 0.010 78 / 0.018) 2px,
-			oklch(0.50 0.010 78 / 0.018) 4px
-		) !important;
-	}
-
-	& .crt-screen {
-		background: radial-gradient(
-			ellipse at center,
-			oklch(0.95 0.012 80) 0%,
-			oklch(0.92 0.010 80) 50%,
-			oklch(0.89 0.009 80) 100%
-		) !important;
-		box-shadow:
-			inset 0 0 60px oklch(0.30 0.010 75 / 0.04),
-			0 0 30px oklch(0.50 0.14 65 / 0.04) !important;
-	}
-
-	& .graph-link {
-		stroke: oklch(0.50 0.018 75 / 0.45) !important;
-		filter: none !important;
-	}
-	& .graph-link-active {
-		stroke: oklch(0.50 0.14 65 / 0.75) !important;
-		filter: none !important;
-	}
-
-	& .shooting-stars {
-		display: none !important;
-	}
-
-	& img[alt="EXO"] {
-		filter: brightness(0) drop-shadow(0 0 6px oklch(0.30 0.010 75 / 0.10)) !important;
-	}
-
-	& .text-red-400 { color: oklch(0.52 0.22 25) !important; }
-	& .text-green-400 { color: oklch(0.48 0.17 155) !important; }
-	& .text-blue-200,
-	& .text-blue-300,
-	& .text-blue-400 { color: oklch(0.48 0.17 250) !important; }
-
-	& .bg-red-500\/10 { background-color: oklch(0.52 0.22 25 / 0.07) !important; }
-	& .bg-red-500\/20 { background-color: oklch(0.52 0.22 25 / 0.11) !important; }
-	& .bg-red-500\/30 { background-color: oklch(0.52 0.22 25 / 0.14) !important; }
-
-	& textarea,
-	& input[type="text"] { color: var(--foreground) !important; }
-	& textarea::placeholder,
-	& input::placeholder { color: oklch(0.50 0.012 78 / 0.55) !important; }
-
-	& .code-block-wrapper,
-	& .math-display-wrapper {
-		background: oklch(0.95 0.010 80) !important;
-		border-color: oklch(0.83 0.007 78) !important;
-	}
-	& .code-block-header,
-	& .math-display-header {
-		background: oklch(0.91 0.009 80) !important;
-		border-color: oklch(0.85 0.007 78) !important;
-	}
-	& .inline-code {
-		background: oklch(0.89 0.009 80) !important;
-		color: oklch(0.20 0.012 75) !important;
-	}
-
-	& blockquote { background: oklch(0.93 0.010 80) !important; }
-	& th {
-		background: oklch(0.90 0.009 80) !important;
-		border-color: oklch(0.80 0.007 78) !important;
-	}
-	& td { border-color: oklch(0.84 0.007 78) !important; }
-	& hr { border-color: oklch(0.84 0.007 78) !important; }
-
-	& .hljs { color: oklch(0.22 0.012 75) !important; }
-	& .hljs-keyword, & .hljs-selector-tag, & .hljs-literal, & .hljs-section, & .hljs-link {
-		color: oklch(0.45 0.18 300) !important;
-	}
-	& .hljs-string, & .hljs-title, & .hljs-name, & .hljs-type,
-	& .hljs-attribute, & .hljs-symbol, & .hljs-bullet, & .hljs-addition,
-	& .hljs-variable, & .hljs-template-tag, & .hljs-template-variable {
-		color: oklch(0.45 0.14 65) !important;
-	}
-	& .hljs-comment, & .hljs-quote, & .hljs-deletion, & .hljs-meta {
-		color: oklch(0.55 0.010 78) !important;
-	}
-	& .hljs-number, & .hljs-regexp, & .hljs-built_in {
-		color: oklch(0.45 0.15 160) !important;
-	}
-	& .hljs-function, & .hljs-class .hljs-title {
-		color: oklch(0.42 0.17 240) !important;
-	}
-
-	& .katex, & .katex .mord, & .katex .minner, & .katex .mop,
-	& .katex .mbin, & .katex .mrel, & .katex .mpunct {
-		color: oklch(0.15 0.012 75) !important;
-	}
-	& .katex .frac-line, & .katex .overline-line, & .katex .underline-line,
-	& .katex .hline, & .katex .rule {
-		border-color: oklch(0.25 0.012 75) !important;
-		background: oklch(0.25 0.012 75) !important;
-	}
-	& .katex svg { fill: oklch(0.25 0.012 75) !important; stroke: oklch(0.25 0.012 75) !important; }
-	& .katex svg path { stroke: oklch(0.25 0.012 75) !important; }
-	& .katex .mopen, & .katex .mclose,
-	& .katex .delimsizing, & [class^="katex .delim-size"] {
-		color: oklch(0.35 0.012 75) !important;
-	}
-
-	& .latex-proof { background: oklch(0.96 0.010 80) !important; border-left-color: oklch(0.72 0.010 78) !important; }
-	& .latex-proof-header { color: oklch(0.22 0.012 75) !important; }
-	& .latex-proof-content { color: oklch(0.15 0.012 75) !important; }
-	& .latex-proof-content::after { color: oklch(0.48 0.012 75) !important; }
-	& .latex-theorem { background: oklch(0.94 0.010 80) !important; border-color: oklch(0.80 0.008 78) !important; }
-	& .latex-diagram-placeholder {
-		background: oklch(0.96 0.010 80) !important;
-		border-color: oklch(0.80 0.008 78) !important;
-		color: oklch(0.38 0.012 75) !important;
-	}
-}
-
@theme inline {
 	--radius-sm: calc(var(--radius) - 2px);
 	--radius-md: var(--radius);
--- a/dashboard/src/app.html
+++ b/dashboard/src/app.html
@@ -1,15 +1,7 @@
 <!doctype html>
-<html lang="en" class="dark">
+<html lang="en">
 	<head>
 		<meta charset="utf-8" />
-		<script>
-			try {
-				if (localStorage.getItem('exo-theme') === 'light') {
-					document.documentElement.classList.remove('dark');
-					document.documentElement.classList.add('light');
-				}
-			} catch (_) {}
-		</script>
 		<link rel="icon" href="%sveltekit.assets%/favicon.ico" />
 		<meta name="viewport" content="width=device-width, initial-scale=1" />
 		<title>EXO</title>
--- a/dashboard/src/lib/components/ChatForm.svelte
+++ b/dashboard/src/lib/components/ChatForm.svelte
@@ -103,7 +103,7 @@
  const modelSupportsThinking = $derived(() => {
    if (!currentModel) return false;
    const caps = modelCapabilities[currentModel] || [];
-    return caps.includes("thinking_toggle") && caps.includes("text");
+    return caps.includes("thinking") && caps.includes("text");
  });

  const isEditOnlyWithoutImage = $derived(
@@ -265,7 +265,6 @@

  function handleSubmit() {
    if ((!message.trim() && uploadedFiles.length === 0) || loading) return;
-    if (isEditOnlyWithoutImage) return;

    const content = message.trim();
    const files = [...uploadedFiles];
@@ -290,11 +289,7 @@
      if (imageFile.preview) {
        editImage(content, imageFile.preview);
      }
-    } else if (
-      currentModel &&
-      modelSupportsTextToImage(currentModel) &&
-      content
-    ) {
+    } else if (isImageModel() && content) {
      // Use image generation for text-to-image models
      generateImage(content);
    } else {
--- a/dashboard/src/lib/components/ChatMessages.svelte
+++ b/dashboard/src/lib/components/ChatMessages.svelte
@@ -225,7 +225,6 @@
  }

  function handleDeleteClick(messageId: string) {
-    if (loading) return;
    deleteConfirmId = messageId;
  }

@@ -256,7 +255,7 @@
 </script>

 <div class="flex flex-col gap-4 sm:gap-6 {className}">
-  {#each messageList as message, i (message.id)}
+  {#each messageList as message (message.id)}
    <div
      class="group flex {message.role === 'user'
        ? 'justify-end'
@@ -318,11 +317,9 @@
          <!-- Delete confirmation -->
          <div class="bg-red-500/10 border border-red-500/30 rounded-lg p-3">
            <p class="text-xs text-red-400 mb-3">
-              {#if i === messageList.length - 1}
-                Delete this message?
-              {:else}
-                Delete this message and all messages after it?
-              {/if}
+              Delete this message{message.role === "user"
+                ? " and all responses after it"
+                : ""}?
            </p>
            <div class="flex gap-2 justify-end">
              <button
@@ -754,13 +751,8 @@
            <!-- Delete button -->
            <button
              onclick={() => handleDeleteClick(message.id)}
-              disabled={loading}
-              class="p-1.5 transition-colors rounded {loading
-                ? 'text-exo-light-gray/30 cursor-not-allowed'
-                : 'text-exo-light-gray hover:text-red-400 hover:bg-red-500/10 cursor-pointer'}"
-              title={loading
-                ? "Cannot delete while generating"
-                : "Delete message"}
+              class="p-1.5 text-exo-light-gray hover:text-red-400 transition-colors rounded hover:bg-red-500/10 cursor-pointer"
+              title="Delete message"
            >
              <svg
                class="w-3.5 h-3.5"
--- a/dashboard/src/lib/components/HeaderNav.svelte
+++ b/dashboard/src/lib/components/HeaderNav.svelte
@@ -1,6 +1,5 @@
 <script lang="ts">
  import { browser } from "$app/environment";
-  import { theme } from "$lib/stores/theme.svelte";

  export let showHome = true;
  export let onHome: (() => void) | null = null;
@@ -80,48 +79,10 @@
    />
  </button>

-  <!-- Right: Theme toggle + Home + Downloads -->
+  <!-- Right: Home + Downloads -->
  <div
    class="absolute right-6 top-1/2 -translate-y-1/2 flex items-center gap-4"
  >
-    <button
-      onclick={() => theme.toggle()}
-      class="p-2 rounded border border-exo-medium-gray/40 hover:border-exo-yellow/50 transition-colors cursor-pointer"
-      title={theme.isLight ? "Switch to dark mode" : "Switch to light mode"}
-      aria-label={theme.isLight
-        ? "Switch to dark mode"
-        : "Switch to light mode"}
-    >
-      {#if theme.isLight}
-        <svg
-          class="w-4 h-4 text-exo-light-gray"
-          fill="none"
-          viewBox="0 0 24 24"
-          stroke="currentColor"
-          stroke-width="2"
-        >
-          <path
-            stroke-linecap="round"
-            stroke-linejoin="round"
-            d="M21 12.79A9 9 0 1111.21 3a7 7 0 009.79 9.79z"
-          />
-        </svg>
-      {:else}
-        <svg
-          class="w-4 h-4 text-exo-light-gray"
-          fill="none"
-          viewBox="0 0 24 24"
-          stroke="currentColor"
-          stroke-width="2"
-        >
-          <circle cx="12" cy="12" r="5" />
-          <path
-            stroke-linecap="round"
-            d="M12 1v2m0 18v2M4.22 4.22l1.42 1.42m12.72 12.72l1.42 1.42M1 12h2m18 0h2M4.22 19.78l1.42-1.42M18.36 5.64l1.42-1.42"
-          />
-        </svg>
-      {/if}
-    </button>
    {#if showHome}
      <button
        onclick={handleHome}
--- a/dashboard/src/lib/components/ImageParamsPanel.svelte
+++ b/dashboard/src/lib/components/ImageParamsPanel.svelte
@@ -59,14 +59,13 @@
  }

  const sizeOptions: ImageGenerationParams["size"][] = [
-    "auto",
    "512x512",
    "768x768",
    "1024x1024",
    "1024x768",
    "768x1024",
-    "1024x1536",
-    "1536x1024",
+    "1024x1365",
+    "1365x1024",
  ];

  const qualityOptions: ImageGenerationParams["quality"][] = [
@@ -177,90 +176,92 @@
 <div class="border-b border-exo-medium-gray/30 px-3 py-2">
  <!-- Basic params row -->
  <div class="flex items-center gap-3 flex-wrap">
-    <!-- Size -->
-    <div class="flex items-center gap-1.5">
-      <span class="text-xs text-exo-light-gray uppercase tracking-wider"
-        >SIZE:</span
-      >
-      <div class="relative">
-        <button
-          bind:this={sizeButtonRef}
-          type="button"
-          onclick={() => (isSizeDropdownOpen = !isSizeDropdownOpen)}
-          class="bg-exo-medium-gray/50 border border-exo-yellow/30 rounded pl-2 pr-6 py-1 text-xs font-mono text-exo-yellow cursor-pointer transition-all duration-200 hover:border-exo-yellow/50 focus:outline-none focus:border-exo-yellow/70 {isSizeDropdownOpen
-            ? 'border-exo-yellow/70'
-            : ''}"
+    <!-- Size (hidden in edit mode - output size comes from input image) -->
+    {#if !isEditMode}
+      <div class="flex items-center gap-1.5">
+        <span class="text-xs text-exo-light-gray uppercase tracking-wider"
+          >SIZE:</span
        >
-          {params.size.toUpperCase()}
-        </button>
-        <div
-          class="absolute right-1.5 top-1/2 -translate-y-1/2 pointer-events-none transition-transform duration-200 {isSizeDropdownOpen
-            ? 'rotate-180'
-            : ''}"
-        >
-          <svg
-            class="w-3 h-3 text-exo-yellow/60"
-            fill="none"
-            viewBox="0 0 24 24"
-            stroke="currentColor"
+        <div class="relative">
+          <button
+            bind:this={sizeButtonRef}
+            type="button"
+            onclick={() => (isSizeDropdownOpen = !isSizeDropdownOpen)}
+            class="bg-exo-medium-gray/50 border border-exo-yellow/30 rounded pl-2 pr-6 py-1 text-xs font-mono text-exo-yellow cursor-pointer transition-all duration-200 hover:border-exo-yellow/50 focus:outline-none focus:border-exo-yellow/70 {isSizeDropdownOpen
+              ? 'border-exo-yellow/70'
+              : ''}"
          >
-            <path
-              stroke-linecap="round"
-              stroke-linejoin="round"
-              stroke-width="2"
-              d="M19 9l-7 7-7-7"
-            />
-          </svg>
-        </div>
-      </div>
-
-      {#if isSizeDropdownOpen}
-        <!-- Backdrop to close dropdown -->
-        <button
-          type="button"
-          class="fixed inset-0 z-[9998] cursor-default"
-          onclick={() => (isSizeDropdownOpen = false)}
-          aria-label="Close dropdown"
-        ></button>
-
-        <!-- Dropdown Panel - fixed positioning to escape overflow:hidden -->
-        <div
-          class="fixed bg-exo-dark-gray border border-exo-yellow/30 rounded shadow-lg shadow-black/50 z-[9999] max-h-48 overflow-y-auto overflow-x-hidden min-w-max"
-          style="bottom: calc(100vh - {sizeDropdownPosition()
-            .top}px + 4px); left: {sizeDropdownPosition().left}px;"
-        >
-          <div class="py-1">
-            {#each sizeOptions as size}
-              <button
-                type="button"
-                onclick={() => selectSize(size)}
-                class="w-full px-3 py-1.5 text-left text-xs font-mono tracking-wide transition-colors duration-100 flex items-center gap-2 {params.size ===
-                size
-                  ? 'bg-transparent text-exo-yellow'
-                  : 'text-exo-light-gray hover:text-exo-yellow'}"
-              >
-                {#if params.size === size}
-                  <svg
-                    class="w-3 h-3 flex-shrink-0"
-                    fill="currentColor"
-                    viewBox="0 0 20 20"
-                  >
-                    <path
-                      fill-rule="evenodd"
-                      d="M16.707 5.293a1 1 0 010 1.414l-8 8a1 1 0 01-1.414 0l-4-4a1 1 0 011.414-1.414L8 12.586l7.293-7.293a1 1 0 011.414 0z"
-                      clip-rule="evenodd"
-                    />
-                  </svg>
-                {:else}
-                  <span class="w-3"></span>
-                {/if}
-                <span>{size.toUpperCase()}</span>
-              </button>
-            {/each}
+            {params.size}
+          </button>
+          <div
+            class="absolute right-1.5 top-1/2 -translate-y-1/2 pointer-events-none transition-transform duration-200 {isSizeDropdownOpen
+              ? 'rotate-180'
+              : ''}"
+          >
+            <svg
+              class="w-3 h-3 text-exo-yellow/60"
+              fill="none"
+              viewBox="0 0 24 24"
+              stroke="currentColor"
+            >
+              <path
+                stroke-linecap="round"
+                stroke-linejoin="round"
+                stroke-width="2"
+                d="M19 9l-7 7-7-7"
+              />
+            </svg>
          </div>
        </div>
-      {/if}
-    </div>
+
+        {#if isSizeDropdownOpen}
+          <!-- Backdrop to close dropdown -->
+          <button
+            type="button"
+            class="fixed inset-0 z-[9998] cursor-default"
+            onclick={() => (isSizeDropdownOpen = false)}
+            aria-label="Close dropdown"
+          ></button>
+
+          <!-- Dropdown Panel - fixed positioning to escape overflow:hidden -->
+          <div
+            class="fixed bg-exo-dark-gray border border-exo-yellow/30 rounded shadow-lg shadow-black/50 z-[9999] max-h-48 overflow-y-auto min-w-max"
+            style="bottom: calc(100vh - {sizeDropdownPosition()
+              .top}px + 4px); left: {sizeDropdownPosition().left}px;"
+          >
+            <div class="py-1">
+              {#each sizeOptions as size}
+                <button
+                  type="button"
+                  onclick={() => selectSize(size)}
+                  class="w-full px-3 py-1.5 text-left text-xs font-mono tracking-wide transition-colors duration-100 flex items-center gap-2 {params.size ===
+                  size
+                    ? 'bg-transparent text-exo-yellow'
+                    : 'text-exo-light-gray hover:text-exo-yellow'}"
+                >
+                  {#if params.size === size}
+                    <svg
+                      class="w-3 h-3 flex-shrink-0"
+                      fill="currentColor"
+                      viewBox="0 0 20 20"
+                    >
+                      <path
+                        fill-rule="evenodd"
+                        d="M16.707 5.293a1 1 0 010 1.414l-8 8a1 1 0 01-1.414 0l-4-4a1 1 0 011.414-1.414L8 12.586l7.293-7.293a1 1 0 011.414 0z"
+                        clip-rule="evenodd"
+                      />
+                    </svg>
+                  {:else}
+                    <span class="w-3"></span>
+                  {/if}
+                  <span>{size}</span>
+                </button>
+              {/each}
+            </div>
+          </div>
+        {/if}
+      </div>
+    {/if}

    <!-- Quality -->
    <div class="flex items-center gap-1.5">
@@ -310,7 +311,7 @@

        <!-- Dropdown Panel - fixed positioning to escape overflow:hidden -->
        <div
-          class="fixed bg-exo-dark-gray border border-exo-yellow/30 rounded shadow-lg shadow-black/50 z-[9999] max-h-48 overflow-y-auto overflow-x-hidden min-w-max"
+          class="fixed bg-exo-dark-gray border border-exo-yellow/30 rounded shadow-lg shadow-black/50 z-[9999] max-h-48 overflow-y-auto min-w-max"
          style="bottom: calc(100vh - {qualityDropdownPosition()
            .top}px + 4px); left: {qualityDropdownPosition().left}px;"
        >
--- a/dashboard/src/lib/stores/app.svelte.ts
+++ b/dashboard/src/lib/stores/app.svelte.ts
@@ -72,6 +72,8 @@ export interface Instance {
    runnerToShard?: Record<string, unknown>;
    nodeToRunner?: Record<string, string>;
  };
+  draftModel?: string;
+  numDraftTokens?: number;
 }

 // Granular node state types from the new state structure
@@ -306,14 +308,13 @@ const IMAGE_PARAMS_STORAGE_KEY = "exo-image-generation-params";
 export interface ImageGenerationParams {
  // Basic params
  size:
-    | "auto"
    | "512x512"
    | "768x768"
    | "1024x1024"
    | "1024x768"
    | "768x1024"
-    | "1024x1536"
-    | "1536x1024";
+    | "1024x1365"
+    | "1365x1024";
  quality: "low" | "medium" | "high";
  outputFormat: "png" | "jpeg";
  numImages: number;
@@ -337,7 +338,7 @@ export interface EditingImage {
 }

 const DEFAULT_IMAGE_PARAMS: ImageGenerationParams = {
-  size: "auto",
+  size: "1024x1024",
  quality: "medium",
  outputFormat: "png",
  numImages: 1,
--- a/dashboard/src/lib/stores/theme.svelte.ts
+++ b/dashboard/src/lib/stores/theme.svelte.ts
@@ -1,28 +0,0 @@
-import { browser } from "$app/environment";
-
-let _isLight = $state(false);
-
-export const theme = {
-  get isLight() {
-    return _isLight;
-  },
-
-  init() {
-    if (!browser) return;
-    _isLight = document.documentElement.classList.contains("light");
-  },
-
-  toggle() {
-    if (!browser) return;
-    _isLight = !_isLight;
-    if (_isLight) {
-      document.documentElement.classList.remove("dark");
-      document.documentElement.classList.add("light");
-      localStorage.setItem("exo-theme", "light");
-    } else {
-      document.documentElement.classList.remove("light");
-      document.documentElement.classList.add("dark");
-      localStorage.setItem("exo-theme", "dark");
-    }
-  },
-};
--- a/dashboard/src/routes/+layout.svelte
+++ b/dashboard/src/routes/+layout.svelte
@@ -1,13 +1,7 @@
 <script lang="ts">
  import "../app.css";
-  import { onMount } from "svelte";
-  import { theme } from "$lib/stores/theme.svelte";

  let { children } = $props();
-
-  onMount(() => {
-    theme.init();
-  });
 </script>

 <svelte:head>
--- a/dashboard/src/routes/downloads/+page.svelte
+++ b/dashboard/src/routes/downloads/+page.svelte
--- a/flake.nix
+++ b/flake.nix
@@ -115,7 +115,7 @@
          packages = lib.optionalAttrs pkgs.stdenv.hostPlatform.isDarwin (
            let
              uvLock = builtins.fromTOML (builtins.readFile ./uv.lock);
-              mlxPackage = builtins.head (builtins.filter (p: p.name == "mlx" && p.source ? git) uvLock.package);
+              mlxPackage = builtins.head (builtins.filter (p: p.name == "mlx") uvLock.package);
              uvLockMlxVersion = mlxPackage.version;
            in
            {
--- a/nix/mlx.nix
+++ b/nix/mlx.nix
@@ -41,16 +41,16 @@ let

  mlx = stdenv.mkDerivation rec {
    pname = "mlx";
-    version = let v = "0.30.7.dev20260218+14841977"; in
+    version = let v = "0.30.6"; in
      assert v == uvLockMlxVersion || throw "MLX version mismatch: nix/mlx.nix has ${v} but uv.lock has ${uvLockMlxVersion}. Update both the version and hash in nix/mlx.nix.";
      v;
    pyproject = true;

    src = fetchFromGitHub {
-      owner = "rltakashige";
-      repo = "mlx-jaccl-fix-small-recv";
-      rev = "1484197707f35186ad3bd614357c7c47fdf86ebc";
-      hash = "sha256-FupCMoK/SF/ldfKuvMSAKECcOP8c+ANgkQlPZttDsLk=";
+      owner = "ml-explore";
+      repo = "mlx";
+      tag = "v${version}";
+      hash = "sha256-avD5EGhwgmPdXLAyQSqTO6AXk/W3ziH+f6AetjK3Sdo=";
    };

    patches = [
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -17,9 +17,9 @@ dependencies = [
    "loguru>=0.7.3",
    "exo_pyo3_bindings", # rust bindings
    "anyio==4.11.0",
-    "mlx; sys_platform == 'darwin'",
+    "mlx==0.30.6; sys_platform == 'darwin'",
    "mlx[cpu]==0.30.6; sys_platform == 'linux'",
-    "mlx-lm==0.30.7",
+    "mlx-lm==0.30.6",
    "tiktoken>=0.12.0", # required for kimi k2 tokenizer
    "hypercorn>=0.18.0",
    "openai-harmony>=0.0.8",
@@ -64,7 +64,6 @@ members = [

 [tool.uv.sources]
 exo_pyo3_bindings = { workspace = true }
-mlx = { git = "https://github.com/rltakashige/mlx-jaccl-fix-small-recv.git", branch = "address-rdma-gpu-locks", marker = "sys_platform == 'darwin'" }
 #mlx-lm = { git = "https://github.com/davidmcc73/mlx-lm", branch = "stable" }
 # Uncomment to use local mlx/mlx-lm development versions:
 # mlx = { path = "/Users/Shared/mlx", editable=true }
@@ -133,7 +132,7 @@ markers = [
 env = [
  "EXO_TESTS=1"
 ]
-addopts = "-m 'not slow' --ignore=tests/start_distributed_test.py"
+addopts = "-m 'not slow'"
 filterwarnings = [
    "ignore:builtin type Swig:DeprecationWarning",
 ]
--- a/python/parts.nix
+++ b/python/parts.nix
@@ -14,9 +14,7 @@

      # Override overlay to inject Nix-built components
      exoOverlay = final: prev: {
-        # Replace workspace exo_pyo3_bindings with Nix-built wheel.
-        # Preserve passthru so mkVirtualEnv can resolve dependency groups.
-        # Copy .pyi stub + py.typed marker so basedpyright can find the types.
+        # Replace workspace exo_pyo3_bindings with Nix-built wheel
        exo-pyo3-bindings = pkgs.stdenv.mkDerivation {
          pname = "exo-pyo3-bindings";
          version = "0.1.0";
@@ -24,12 +22,6 @@
          # Install from pre-built wheel
          nativeBuildInputs = [ final.pyprojectWheelHook ];
          dontStrip = true;
-          passthru = prev.exo-pyo3-bindings.passthru or { };
-          postInstall = ''
-            local siteDir=$out/${final.python.sitePackages}/exo_pyo3_bindings
-            cp ${inputs.self}/rust/exo_pyo3_bindings/exo_pyo3_bindings.pyi $siteDir/
-            touch $siteDir/py.typed
-          '';
        };
      };

@@ -37,47 +29,17 @@

      # Overlay to provide build systems and custom packages
      buildSystemsOverlay = final: prev: {
+        # Use our pure Nix-built MLX with Metal support
+        mlx = self'.packages.mlx;
+
        # mlx-lm is a git dependency that needs setuptools
        mlx-lm = prev.mlx-lm.overrideAttrs (old: {
          nativeBuildInputs = (old.nativeBuildInputs or [ ]) ++ [
            final.setuptools
          ];
        });
-      } // lib.optionalAttrs pkgs.stdenv.hostPlatform.isDarwin {
-        # Use our pure Nix-built MLX with Metal support (macOS only)
-        mlx = self'.packages.mlx;
      };

-      # Additional overlay for Linux-specific fixes (type checking env).
-      # Native wheels have shared lib dependencies we don't need at type-check time.
-      linuxOverlay = final: prev:
-        let
-          ignoreMissing = drv: drv.overrideAttrs { autoPatchelfIgnoreMissingDeps = [ "*" ]; };
-          nvidiaPackages = lib.filterAttrs (name: _: lib.hasPrefix "nvidia-" name) prev;
-        in
-        lib.optionalAttrs pkgs.stdenv.hostPlatform.isLinux (
-          (lib.mapAttrs (_: ignoreMissing) nvidiaPackages) // {
-            mlx = ignoreMissing prev.mlx;
-            mlx-cuda-13 = prev.mlx-cuda-13.overrideAttrs (old: {
-              buildInputs = (old.buildInputs or [ ]) ++ [
-                final.nvidia-cublas
-                final.nvidia-cuda-nvrtc
-                final.nvidia-cudnn-cu13
-                final.nvidia-nccl-cu13
-              ];
-              preFixup = ''
-                addAutoPatchelfSearchPath ${final.nvidia-cublas}
-                addAutoPatchelfSearchPath ${final.nvidia-cuda-nvrtc}
-                addAutoPatchelfSearchPath ${final.nvidia-cudnn-cu13}
-                addAutoPatchelfSearchPath ${final.nvidia-nccl-cu13}
-              '';
-              autoPatchelfIgnoreMissingDeps = [ "libcuda.so.1" ];
-            });
-            torch = ignoreMissing prev.torch;
-            triton = ignoreMissing prev.triton;
-          }
-        );
-
      pythonSet = (pkgs.callPackage inputs.pyproject-nix.build.packages {
        inherit python;
      }).overrideScope (
@@ -86,28 +48,16 @@
          overlay
          exoOverlay
          buildSystemsOverlay
-          linuxOverlay
        ]
      );
-      # mlx-cpu and mlx-cuda-13 both ship mlx/ site-packages files; keep first.
-      # mlx-cpu/mlx-cuda-13 and nvidia-cudnn-cu12/cu13 ship overlapping files.
-      venvCollisionPaths = lib.optionals pkgs.stdenv.hostPlatform.isLinux [
-        "lib/python3.13/site-packages/mlx*"
-        "lib/python3.13/site-packages/nvidia*"
-      ];
-
-      exoVenv = (pythonSet.mkVirtualEnv "exo-env" workspace.deps.default).overrideAttrs {
-        venvIgnoreCollisions = venvCollisionPaths;
-      };
+      exoVenv = pythonSet.mkVirtualEnv "exo-env" workspace.deps.default;

      # Virtual environment with dev dependencies for testing
-      testVenv = (pythonSet.mkVirtualEnv "exo-test-env" (
+      testVenv = pythonSet.mkVirtualEnv "exo-test-env" (
        workspace.deps.default // {
          exo = [ "dev" ]; # Include pytest, pytest-asyncio, pytest-env
        }
-      )).overrideAttrs {
-        venvIgnoreCollisions = venvCollisionPaths;
-      };
+      );

      mkPythonScript = name: path: pkgs.writeShellApplication {
        inherit name;
@@ -168,21 +118,6 @@
          ${pkgs.ruff}/bin/ruff check ${inputs.self}
          touch $out
        '';
-
-        # Hermetic basedpyright type checking
-        typecheck = pkgs.runCommand "typecheck"
-          {
-            nativeBuildInputs = [
-              testVenv
-              pkgs.basedpyright
-            ];
-          }
-          ''
-            cd ${inputs.self}
-            export HOME=$TMPDIR
-            basedpyright --pythonpath ${testVenv}/bin/python
-            touch $out
-          '';
      };
    };
 }
--- a/resources/inference_model_cards/mlx-community--DeepSeek-V3.1-4bit.toml
+++ b/resources/inference_model_cards/mlx-community--DeepSeek-V3.1-4bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "deepseek"
 quantization = "4bit"
 base_model = "DeepSeek V3.1"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 405874409472
--- a/resources/inference_model_cards/mlx-community--DeepSeek-V3.1-8bit.toml
+++ b/resources/inference_model_cards/mlx-community--DeepSeek-V3.1-8bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "deepseek"
 quantization = "8bit"
 base_model = "DeepSeek V3.1"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 765577920512
--- a/resources/inference_model_cards/mlx-community--GLM-4.5-Air-8bit.toml
+++ b/resources/inference_model_cards/mlx-community--GLM-4.5-Air-8bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "glm"
 quantization = "8bit"
 base_model = "GLM 4.5 Air"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 122406567936
--- a/resources/inference_model_cards/mlx-community--GLM-4.5-Air-bf16.toml
+++ b/resources/inference_model_cards/mlx-community--GLM-4.5-Air-bf16.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "glm"
 quantization = "bf16"
 base_model = "GLM 4.5 Air"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 229780750336
--- a/resources/inference_model_cards/mlx-community--GLM-4.7-4bit.toml
+++ b/resources/inference_model_cards/mlx-community--GLM-4.7-4bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "glm"
 quantization = "4bit"
 base_model = "GLM 4.7"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 198556925568
--- a/resources/inference_model_cards/mlx-community--GLM-4.7-6bit.toml
+++ b/resources/inference_model_cards/mlx-community--GLM-4.7-6bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "glm"
 quantization = "6bit"
 base_model = "GLM 4.7"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 286737579648
--- a/resources/inference_model_cards/mlx-community--GLM-4.7-8bit-gs32.toml
+++ b/resources/inference_model_cards/mlx-community--GLM-4.7-8bit-gs32.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "glm"
 quantization = "8bit"
 base_model = "GLM 4.7"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 396963397248
--- a/resources/inference_model_cards/mlx-community--GLM-4.7-Flash-4bit.toml
+++ b/resources/inference_model_cards/mlx-community--GLM-4.7-Flash-4bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "glm"
 quantization = "4bit"
 base_model = "GLM 4.7 Flash"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 19327352832
--- a/resources/inference_model_cards/mlx-community--GLM-4.7-Flash-5bit.toml
+++ b/resources/inference_model_cards/mlx-community--GLM-4.7-Flash-5bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "glm"
 quantization = "5bit"
 base_model = "GLM 4.7 Flash"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 22548578304
--- a/resources/inference_model_cards/mlx-community--GLM-4.7-Flash-6bit.toml
+++ b/resources/inference_model_cards/mlx-community--GLM-4.7-Flash-6bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "glm"
 quantization = "6bit"
 base_model = "GLM 4.7 Flash"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 26843545600
--- a/resources/inference_model_cards/mlx-community--GLM-4.7-Flash-8bit.toml
+++ b/resources/inference_model_cards/mlx-community--GLM-4.7-Flash-8bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "glm"
 quantization = "8bit"
 base_model = "GLM 4.7 Flash"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 34359738368
--- a/resources/inference_model_cards/mlx-community--GLM-5-8bit.toml
+++ b/resources/inference_model_cards/mlx-community--GLM-5-8bit.toml
@@ -1,12 +0,0 @@
-model_id = "mlx-community/GLM-5-8bit-MXFP8"
-n_layers = 78
-hidden_size = 6144
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "glm"
-quantization = "8bit"
-base_model = "GLM-5"
-capabilities = ["text", "thinking"]
-
-[storage_size]
-in_bytes = 790517400864
--- a/resources/inference_model_cards/mlx-community--GLM-5-MXFP4-Q8.toml
+++ b/resources/inference_model_cards/mlx-community--GLM-5-MXFP4-Q8.toml
@@ -1,12 +0,0 @@
-model_id = "mlx-community/GLM-5-MXFP4-Q8"
-n_layers = 78
-hidden_size = 6144
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "glm"
-quantization = "MXFP4-Q8"
-base_model = "GLM-5"
-capabilities = ["text", "thinking"]
-
-[storage_size]
-in_bytes = 405478939008
--- a/resources/inference_model_cards/mlx-community--GLM-5-bf16.toml
+++ b/resources/inference_model_cards/mlx-community--GLM-5-bf16.toml
@@ -1,12 +0,0 @@
-model_id = "mlx-community/GLM-5"
-n_layers = 78
-hidden_size = 6144
-supports_tensor = true
-tasks = ["TextGeneration"]
-family = "glm"
-quantization = "bf16"
-base_model = "GLM-5"
-capabilities = ["text", "thinking"]
-
-[storage_size]
-in_bytes = 1487822475264
--- a/resources/inference_model_cards/mlx-community--Kimi-K2-Thinking.toml
+++ b/resources/inference_model_cards/mlx-community--Kimi-K2-Thinking.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "kimi"
 quantization = ""
 base_model = "Kimi K2"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 706522120192
--- a/resources/inference_model_cards/mlx-community--Kimi-K2.5.toml
+++ b/resources/inference_model_cards/mlx-community--Kimi-K2.5.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "kimi"
 quantization = ""
 base_model = "Kimi K2.5"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 662498705408
--- a/resources/inference_model_cards/mlx-community--MiniMax-M2.1-3bit.toml
+++ b/resources/inference_model_cards/mlx-community--MiniMax-M2.1-3bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "minimax"
 quantization = "3bit"
 base_model = "MiniMax M2.1"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 100086644736
--- a/resources/inference_model_cards/mlx-community--MiniMax-M2.1-8bit.toml
+++ b/resources/inference_model_cards/mlx-community--MiniMax-M2.1-8bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "minimax"
 quantization = "8bit"
 base_model = "MiniMax M2.1"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 242986745856
--- a/resources/inference_model_cards/mlx-community--Qwen3-0.6B-4bit.toml
+++ b/resources/inference_model_cards/mlx-community--Qwen3-0.6B-4bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "qwen"
 quantization = "4bit"
 base_model = "Qwen3 0.6B"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 342884352
--- a/resources/inference_model_cards/mlx-community--Qwen3-0.6B-8bit.toml
+++ b/resources/inference_model_cards/mlx-community--Qwen3-0.6B-8bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "qwen"
 quantization = "8bit"
 base_model = "Qwen3 0.6B"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 698351616
--- a/resources/inference_model_cards/mlx-community--Qwen3-235B-A22B-Instruct-2507-4bit.toml
+++ b/resources/inference_model_cards/mlx-community--Qwen3-235B-A22B-Instruct-2507-4bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "qwen"
 quantization = "4bit"
 base_model = "Qwen3 235B"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 141733920768
--- a/resources/inference_model_cards/mlx-community--Qwen3-235B-A22B-Instruct-2507-8bit.toml
+++ b/resources/inference_model_cards/mlx-community--Qwen3-235B-A22B-Instruct-2507-8bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "qwen"
 quantization = "8bit"
 base_model = "Qwen3 235B"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 268435456000
--- a/resources/inference_model_cards/mlx-community--Qwen3-30B-A3B-4bit.toml
+++ b/resources/inference_model_cards/mlx-community--Qwen3-30B-A3B-4bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "qwen"
 quantization = "4bit"
 base_model = "Qwen3 30B"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 17612931072
--- a/resources/inference_model_cards/mlx-community--Qwen3-30B-A3B-8bit.toml
+++ b/resources/inference_model_cards/mlx-community--Qwen3-30B-A3B-8bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "qwen"
 quantization = "8bit"
 base_model = "Qwen3 30B"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 33279705088
--- a/resources/inference_model_cards/mlx-community--Qwen3-Next-80B-A3B-Thinking-4bit.toml
+++ b/resources/inference_model_cards/mlx-community--Qwen3-Next-80B-A3B-Thinking-4bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "qwen"
 quantization = "4bit"
 base_model = "Qwen3 Next 80B"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 47080074240
--- a/resources/inference_model_cards/mlx-community--Qwen3-Next-80B-A3B-Thinking-8bit.toml
+++ b/resources/inference_model_cards/mlx-community--Qwen3-Next-80B-A3B-Thinking-8bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "qwen"
 quantization = "8bit"
 base_model = "Qwen3 Next 80B"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 88814387200
--- a/resources/inference_model_cards/mlx-community--Step-3.5-Flash-4bit.toml
+++ b/resources/inference_model_cards/mlx-community--Step-3.5-Flash-4bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "step"
 quantization = "4bit"
 base_model = "Step 3.5 Flash"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 114572190076
--- a/resources/inference_model_cards/mlx-community--Step-3.5-Flash-6bit.toml
+++ b/resources/inference_model_cards/mlx-community--Step-3.5-Flash-6bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "step"
 quantization = "6bit"
 base_model = "Step 3.5 Flash"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 159039627774
--- a/resources/inference_model_cards/mlx-community--Step-3.5-Flash-8Bit.toml
+++ b/resources/inference_model_cards/mlx-community--Step-3.5-Flash-8Bit.toml
@@ -6,7 +6,7 @@ tasks = ["TextGeneration"]
 family = "step"
 quantization = "8bit"
 base_model = "Step 3.5 Flash"
-capabilities = ["text", "thinking", "thinking_toggle"]
+capabilities = ["text", "thinking"]

 [storage_size]
 in_bytes = 209082699847
--- a/rust/exo_pyo3_bindings/Cargo.toml
+++ b/rust/exo_pyo3_bindings/Cargo.toml
@@ -25,17 +25,17 @@ workspace = true
 networking = { workspace = true }

 # interop
-pyo3 = { version = "0.27.2", features = [
-    # "abi3-py313", # tells pyo3 (and maturin) to build using the stable ABI with minimum Python version 3.13
+pyo3 = { version = "0.27.1", features = [
+    # "abi3-py311", # tells pyo3 (and maturin) to build using the stable ABI with minimum Python version 3.11
    "nightly", # enables better-supported GIL integration
    "experimental-async", # async support in #[pyfunction] & #[pymethods]
    #"experimental-inspect", # inspection of generated binary => easier to automate type-hint generation
    #"py-clone", # adding Clone-ing of `Py<T>` without GIL (may cause panics - remove if panics happen)
-    # "multiple-pymethods", # allows multiple #[pymethods] sections per class
+    "multiple-pymethods", # allows multiple #[pymethods] sections per class

    # integrations with other libraries
-    # "arc_lock", "bigdecimal", "either", "hashbrown", "indexmap", "num-bigint", "num-complex", "num-rational",
-    # "ordered-float", "rust_decimal", "smallvec",
+    "arc_lock", "bigdecimal", "either", "hashbrown", "indexmap", "num-bigint", "num-complex", "num-rational",
+    "ordered-float", "rust_decimal", "smallvec",
    # "anyhow", "chrono", "chrono-local", "chrono-tz", "eyre", "jiff-02", "lock_api", "parking-lot", "time",  "serde",
 ] }
 pyo3-stub-gen = { version = "0.17.2" }
@@ -45,6 +45,8 @@ pyo3-log = "0.13.2"
 # macro dependencies
 extend = { workspace = true }
 delegate = { workspace = true }
+impl-trait-for-tuples = { workspace = true }
+derive_more = { workspace = true }
 pin-project = { workspace = true }

 # async runtime
@@ -52,11 +54,24 @@ tokio = { workspace = true, features = ["full", "tracing"] }
 futures = { workspace = true }

 # utility dependencies
+once_cell = "1.21.3"
+thread_local = "1.1.9"
 util = { workspace = true }
+thiserror = { workspace = true }
+#internment = { workspace = true }
+#recursion = { workspace = true }
+#generativity = { workspace = true }
+#itertools = { workspace = true }
+

 # Tracing
+#tracing = "0.1"
+#tracing-subscriber = "0.3"
+#console-subscriber = "0.1.5"
+#tracing-log = "0.2.0"
 log = { workspace = true }
 env_logger = "0.11"

+
 # Networking
 libp2p = { workspace = true, features = ["full"] }
--- a/rust/exo_pyo3_bindings/src/allow_threading.rs
+++ b/rust/exo_pyo3_bindings/src/allow_threading.rs
@@ -6,7 +6,7 @@ use pyo3::marker::Ungil;
 use pyo3::prelude::*;
 use std::{
    future::Future,
-    pin::Pin,
+    pin::{Pin, pin},
    task::{Context, Poll},
 };

@@ -33,6 +33,8 @@ where

    fn poll(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Self::Output> {
        let waker = cx.waker();
-        Python::attach(|py| py.detach(|| self.project().0.poll(&mut Context::from_waker(waker))))
+        Python::with_gil(|py| {
+            py.allow_threads(|| self.project().0.poll(&mut Context::from_waker(waker)))
+        })
    }
 }
--- a/rust/exo_pyo3_bindings/src/examples/mod.rs
+++ b/rust/exo_pyo3_bindings/src/examples/mod.rs
@@ -0,0 +1,240 @@
+//! This module exists to hold examples of some pyo3 patterns that may be too complex to
+//! re-create from scratch, but too inhomogenous to create an abstraction/wrapper around.
+//!
+//! Pattern examples include:
+//!  - Async task handles: with GC-integrated cleanup
+//!  - Sync/async callbacks from python: with propper eventloop handling
+//!
+//! Mutability pattern: https://pyo3.rs/v0.26.0/async-await.html#send--static-constraint
+//!  - Store mutable fields in tokio's `Mutex<T>`
+//!  - For async code: take `&self` and `.lock().await`
+//!  - For sync code: take `&mut self` and `.get_mut()`
+
+use crate::ext::{PyResultExt as _, ResultExt as _, TokioRuntimeExt as _};
+use futures::FutureExt as _;
+use futures::future::BoxFuture;
+use pyo3::exceptions::PyRuntimeError;
+use pyo3::prelude::{PyModule, PyModuleMethods as _};
+use pyo3::{
+    Bound, Py, PyAny, PyErr, PyResult, PyTraverseError, PyVisit, Python, pyclass, pymethods,
+};
+use std::time::Duration;
+use tokio::sync::mpsc;
+use tokio::sync::mpsc::error::TryRecvError;
+
+fn needs_tokio_runtime() {
+    tokio::runtime::Handle::current();
+}
+
+type SyncCallback = Box<dyn Fn() + Send + Sync>;
+type AsyncCallback = Box<dyn Fn() -> BoxFuture<'static, ()> + Send + Sync>;
+
+enum AsyncTaskMessage {
+    SyncCallback(SyncCallback),
+    AsyncCallback(AsyncCallback),
+}
+
+async fn async_task(
+    sender: mpsc::UnboundedSender<()>,
+    mut receiver: mpsc::UnboundedReceiver<AsyncTaskMessage>,
+) {
+    log::info!("RUST: async task started");
+
+    // task state
+    let mut interval = tokio::time::interval(Duration::from_secs(1));
+
+    let mut sync_cbs: Vec<SyncCallback> = vec![];
+    let mut async_cbs: Vec<AsyncCallback> = vec![];
+
+    loop {
+        tokio::select! {
+            // handle incoming messages from task-handle
+            message = receiver.recv() => {
+                // handle closed channel by exiting
+                let Some(message) = message else {
+                    log::info!("RUST: channel closed");
+                    break;
+                };
+
+                // dispatch incoming event
+                match message {
+                    AsyncTaskMessage::SyncCallback(cb) => {
+                        sync_cbs.push(cb);
+                    }
+                    AsyncTaskMessage::AsyncCallback(cb) => {
+                        async_cbs.push(cb);
+                    }
+                }
+            }
+
+            // handle all other events
+            _ = interval.tick() => {
+                log::info!("RUST: async task tick");
+
+                // call back all sync callbacks
+                for cb in &sync_cbs {
+                    cb();
+                }
+
+                // call back all async callbacks
+                for cb in &async_cbs {
+                    cb().await;
+                }
+
+                // send event on unbounded channel
+                sender.send(()).expect("handle receiver cannot be closed/dropped");
+            }
+        }
+    }
+
+    log::info!("RUST: async task stopped");
+}
+
+// #[gen_stub_pyclass]
+#[pyclass(name = "AsyncTaskHandle")]
+#[derive(Debug)]
+struct PyAsyncTaskHandle {
+    sender: Option<mpsc::UnboundedSender<AsyncTaskMessage>>,
+    receiver: mpsc::UnboundedReceiver<()>,
+}
+
+#[allow(clippy::expect_used)]
+impl PyAsyncTaskHandle {
+    const fn sender(&self) -> &mpsc::UnboundedSender<AsyncTaskMessage> {
+        self.sender
+            .as_ref()
+            .expect("The sender should only be None after de-initialization.")
+    }
+
+    const fn sender_mut(&mut self) -> &mpsc::UnboundedSender<AsyncTaskMessage> {
+        self.sender
+            .as_mut()
+            .expect("The sender should only be None after de-initialization.")
+    }
+
+    const fn new(
+        sender: mpsc::UnboundedSender<AsyncTaskMessage>,
+        receiver: mpsc::UnboundedReceiver<()>,
+    ) -> Self {
+        Self {
+            sender: Some(sender),
+            receiver,
+        }
+    }
+}
+
+// #[gen_stub_pymethods]
+#[pymethods]
+impl PyAsyncTaskHandle {
+    #[new]
+    fn py_new(py: Python<'_>) -> PyResult<Self> {
+        use pyo3_async_runtimes::tokio::get_runtime;
+
+        // create communication channel TOWARDS our task
+        let (h_sender, t_receiver) = mpsc::unbounded_channel::<AsyncTaskMessage>();
+
+        // create communication channel FROM our task
+        let (t_sender, h_receiver) = mpsc::unbounded_channel::<()>();
+
+        // perform necessary setup within tokio context - or it crashes
+        let () = get_runtime().block_on(async { needs_tokio_runtime() });
+
+        // spawn tokio task with this thread's task-locals - without this, async callbacks on the new threads will not work!!
+        _ = get_runtime().spawn_with_scope(py, async move {
+            async_task(t_sender, t_receiver).await;
+        });
+        Ok(Self::new(h_sender, h_receiver))
+    }
+
+    /// NOTE: exceptions in callbacks are silently ignored until end of execution
+    fn add_sync_callback(
+        &self,
+        // #[gen_stub(override_type(
+        //     type_repr="collections.abc.Callable[[], None]",
+        //     imports=("collections.abc")
+        // ))]
+        callback: Py<PyAny>,
+    ) -> PyResult<()> {
+        // blocking call to async method -> can do non-blocking if needed
+        self.sender()
+            .send(AsyncTaskMessage::SyncCallback(Box::new(move || {
+                _ = Python::with_gil(|py| callback.call0(py).write_unraisable_with(py));
+            })))
+            .pyerr()?;
+        Ok(())
+    }
+
+    /// NOTE: exceptions in callbacks are silently ignored until end of execution
+    fn add_async_callback(
+        &self,
+        // #[gen_stub(override_type(
+        //     type_repr="collections.abc.Callable[[], collections.abc.Awaitable[None]]",
+        //     imports=("collections.abc")
+        // ))]
+        callback: Py<PyAny>,
+    ) -> PyResult<()> {
+        // blocking call to async method -> can do non-blocking if needed
+        self.sender()
+            .send(AsyncTaskMessage::AsyncCallback(Box::new(move || {
+                let c = Python::with_gil(|py| callback.clone_ref(py));
+                async move {
+                    if let Some(f) = Python::with_gil(|py| {
+                        let coroutine = c.call0(py).write_unraisable_with(py)?;
+                        pyo3_async_runtimes::tokio::into_future(coroutine.into_bound(py))
+                            .write_unraisable_with(py)
+                    }) {
+                        _ = f.await.write_unraisable();
+                    }
+                }
+                .boxed()
+            })))
+            .pyerr()?;
+        Ok(())
+    }
+
+    async fn receive_unit(&mut self) -> PyResult<()> {
+        self.receiver
+            .recv()
+            .await
+            .ok_or(PyErr::new::<PyRuntimeError, _>(
+                "cannot receive unit on closed channel",
+            ))
+    }
+
+    fn drain_units(&mut self) -> PyResult<i32> {
+        let mut cnt = 0;
+        loop {
+            match self.receiver.try_recv() {
+                Err(TryRecvError::Disconnected) => {
+                    return Err(PyErr::new::<PyRuntimeError, _>(
+                        "cannot receive unit on closed channel",
+                    ));
+                }
+                Err(TryRecvError::Empty) => return Ok(cnt),
+                Ok(()) => {
+                    cnt += 1;
+                    continue;
+                }
+            }
+        }
+    }
+
+    // #[gen_stub(skip)]
+    const fn __traverse__(&self, _visit: PyVisit<'_>) -> Result<(), PyTraverseError> {
+        Ok(()) // This is needed purely so `__clear__` can work
+    }
+
+    // #[gen_stub(skip)]
+    fn __clear__(&mut self) {
+        // TODO: may or may not need to await a "kill-signal" oneshot channel message,
+        //       to ensure that the networking task is done BEFORE exiting the clear function...
+        //       but this may require GIL?? and it may not be safe to call GIL here??
+        self.sender = None; // Using Option<T> as a trick to force `sender` channel to be dropped
+    }
+}
+
+pub fn examples_submodule(m: &Bound<'_, PyModule>) -> PyResult<()> {
+    m.add_class::<PyAsyncTaskHandle>()?;
+
+    Ok(())
+}
--- a/rust/exo_pyo3_bindings/src/lib.rs
+++ b/rust/exo_pyo3_bindings/src/lib.rs
@@ -17,6 +17,7 @@

 extern crate core;
 mod allow_threading;
+mod examples;
 pub(crate) mod networking;
 pub(crate) mod pylibp2p;

@@ -24,6 +25,7 @@ use crate::networking::networking_submodule;
 use crate::pylibp2p::ident::ident_submodule;
 use crate::pylibp2p::multiaddr::multiaddr_submodule;
 use pyo3::prelude::PyModule;
+use pyo3::prelude::*;
 use pyo3::{Bound, PyResult, pyclass, pymodule};
 use pyo3_stub_gen::define_stub_info_gatherer;

@@ -34,10 +36,14 @@ pub(crate) mod r#const {

 /// Namespace for all the type/trait aliases used by this crate.
 pub(crate) mod alias {
+    use std::error::Error;
    use std::marker::Tuple;

    pub trait SendFn<Args: Tuple + Send + 'static, Output> =
        Fn<Args, Output = Output> + Send + 'static;
+
+    pub type AnyError = Box<dyn Error + Send + Sync + 'static>;
+    pub type AnyResult<T> = Result<T, AnyError>;
 }

 /// Namespace for crate-wide extension traits/methods
@@ -45,6 +51,7 @@ pub(crate) mod ext {
    use crate::allow_threading::AllowThreads;
    use extend::ext;
    use pyo3::exceptions::{PyConnectionError, PyRuntimeError};
+    use pyo3::marker::Ungil;
    use pyo3::types::PyBytes;
    use pyo3::{Py, PyErr, PyResult, Python};
    use tokio::runtime::Runtime;
@@ -55,7 +62,7 @@ pub(crate) mod ext {
    #[ext(pub, name = ByteArrayExt)]
    impl [u8] {
        fn pybytes(&self) -> Py<PyBytes> {
-            Python::attach(|py| PyBytes::new(py, self).unbind())
+            Python::with_gil(|py| PyBytes::new(py, self).unbind())
        }
    }

@@ -91,7 +98,7 @@ pub(crate) mod ext {
    #[ext(pub, name = PyResultExt)]
    impl<T> PyResult<T> {
        fn write_unraisable(self) -> Option<T> {
-            Python::attach(|py| self.write_unraisable_with(py))
+            Python::with_gil(|py| self.write_unraisable_with(py))
        }

        fn write_unraisable_with(self, py: Python<'_>) -> Option<T> {
@@ -168,6 +175,24 @@ pub(crate) mod ext {
    }
 }

+pub(crate) mod private {
+    use std::marker::Sized;
+
+    /// Sealed traits support
+    pub trait Sealed {}
+    impl<T: ?Sized> Sealed for T {}
+}
+
+/// A wrapper around [`Py`] that implements [`Clone`] using [`Python::with_gil`].
+#[repr(transparent)]
+pub(crate) struct ClonePy<T>(pub Py<T>);
+
+impl<T> Clone for ClonePy<T> {
+    fn clone(&self) -> Self {
+        Python::with_gil(|py| Self(self.0.clone_ref(py)))
+    }
+}
+
 /// A Python module implemented in Rust. The name of this function must match
 /// the `lib.name` setting in the `Cargo.toml`, else Python will not be able to
 /// import the module.
--- a/rust/exo_pyo3_bindings/src/networking.rs
+++ b/rust/exo_pyo3_bindings/src/networking.rs
@@ -11,9 +11,9 @@ use crate::ext::{ResultExt as _, TokioMpscReceiverExt as _, TokioMpscSenderExt a
 use crate::pyclass;
 use crate::pylibp2p::ident::{PyKeypair, PyPeerId};
 use libp2p::futures::StreamExt as _;
-use libp2p::gossipsub;
 use libp2p::gossipsub::{IdentTopic, Message, MessageId, PublishError};
 use libp2p::swarm::SwarmEvent;
+use libp2p::{gossipsub, mdns};
 use networking::discovery;
 use networking::swarm::create_swarm;
 use pyo3::prelude::{PyModule, PyModuleMethods as _};
@@ -25,7 +25,7 @@ use tokio::sync::{Mutex, mpsc, oneshot};

 mod exception {
    use pyo3::types::PyTuple;
-    use pyo3::{exceptions::PyException, prelude::*};
+    use pyo3::{PyErrArguments, exceptions::PyException, prelude::*};
    use pyo3_stub_gen::derive::*;

    #[gen_stub_pyclass]
@@ -155,6 +155,7 @@ async fn networking_task(
 ) {
    use SwarmEvent::*;
    use ToTask::*;
+    use mdns::Event::*;
    use networking::swarm::BehaviourEvent::*;

    log::info!("RUST: networking task started");
@@ -484,7 +485,7 @@ impl PyNetworkingHandle {
        let (tx, rx) = oneshot::channel();

        // send off request to subscribe
-        let data = Python::attach(|py| Vec::from(data.as_bytes(py)));
+        let data = Python::with_gil(|py| Vec::from(data.as_bytes(py)));
        self.to_task_tx()
            .send_py(ToTask::GossipsubPublish {
                topic,
--- a/rust/networking/Cargo.toml
+++ b/rust/networking/Cargo.toml
@@ -19,6 +19,8 @@ either = { workspace = true }
 # macro dependencies
 extend = { workspace = true }
 delegate = { workspace = true }
+impl-trait-for-tuples = { workspace = true }
+derive_more = { workspace = true }

 # async
 tokio = { workspace = true, features = ["full"] }
@@ -27,6 +29,11 @@ futures-timer = { workspace = true }

 # utility dependencies
 util = { workspace = true }
+thiserror = { workspace = true }
+#internment = { workspace = true }
+#recursion = { workspace = true }
+#generativity = { workspace = true }
+#itertools = { workspace = true }
 tracing-subscriber = { version = "0.3.19", features = ["default", "env-filter"] }
 keccak-const = { workspace = true }

@@ -34,4 +41,4 @@ keccak-const = { workspace = true }
 log = { workspace = true }

 # networking
-libp2p = { workspace = true, features = ["full"] }
+libp2p = { workspace = true, features = ["full"] }
--- a/rust/networking/examples/chatroom_manual.rs
+++ b/rust/networking/examples/chatroom_manual.rs
@@ -24,8 +24,8 @@ use libp2p::{
    swarm::{NetworkBehaviour, SwarmEvent},
    tcp, yamux,
 };
-use std::error::Error;
 use std::time::Duration;
+use std::{error::Error, hash::Hash};
 use tokio::{io, io::AsyncBufReadExt, select};
 use tracing_subscriber::EnvFilter;

--- a/rust/networking/src/discovery.rs
+++ b/rust/networking/src/discovery.rs
@@ -1,4 +1,5 @@
 use crate::ext::MultiaddrExt;
+use crate::keep_alive;
 use delegate::delegate;
 use either::Either;
 use futures::FutureExt;
--- a/rust/networking/src/keep_alive.rs
+++ b/rust/networking/src/keep_alive.rs
@@ -0,0 +1,44 @@
+use delegate::delegate;
+use libp2p::swarm::handler::ConnectionEvent;
+use libp2p::swarm::{ConnectionHandlerEvent, SubstreamProtocol, dummy, handler};
+use std::task::{Context, Poll};
+
+/// An implementation of [`ConnectionHandler`] that doesn't handle any protocols, but it keeps
+/// the connection alive.
+#[derive(Clone)]
+#[repr(transparent)]
+pub struct ConnectionHandler(dummy::ConnectionHandler);
+
+impl ConnectionHandler {
+    pub fn new() -> Self {
+        ConnectionHandler(dummy::ConnectionHandler)
+    }
+}
+
+impl handler::ConnectionHandler for ConnectionHandler {
+    // delegate types and implementation mostly to dummy handler
+    type FromBehaviour = <dummy::ConnectionHandler as handler::ConnectionHandler>::FromBehaviour;
+    type ToBehaviour = <dummy::ConnectionHandler as handler::ConnectionHandler>::ToBehaviour;
+    type InboundProtocol =
+        <dummy::ConnectionHandler as handler::ConnectionHandler>::InboundProtocol;
+    type OutboundProtocol =
+        <dummy::ConnectionHandler as handler::ConnectionHandler>::OutboundProtocol;
+    type InboundOpenInfo =
+        <dummy::ConnectionHandler as handler::ConnectionHandler>::InboundOpenInfo;
+    type OutboundOpenInfo =
+        <dummy::ConnectionHandler as handler::ConnectionHandler>::OutboundOpenInfo;
+
+    delegate! {
+        to self.0 {
+            fn listen_protocol(&self) -> SubstreamProtocol<Self::InboundProtocol, Self::InboundOpenInfo>;
+            fn poll(&mut self, cx: &mut Context<'_>) -> Poll<ConnectionHandlerEvent<Self::OutboundProtocol, Self::OutboundOpenInfo, Self::ToBehaviour>>;
+            fn on_behaviour_event(&mut self, event: Self::FromBehaviour);
+            fn on_connection_event(&mut self, event: ConnectionEvent<Self::InboundProtocol, Self::OutboundProtocol, Self::InboundOpenInfo, Self::OutboundOpenInfo>);
+        }
+    }
+
+    // specifically override this to force connection to stay alive
+    fn connection_keep_alive(&self) -> bool {
+        true
+    }
+}
--- a/rust/networking/src/lib.rs
+++ b/rust/networking/src/lib.rs
@@ -3,7 +3,19 @@
 //! this is here as a placeholder documentation
 //!
 //!
+
+// enable Rust-unstable features for convenience
+#![feature(trait_alias)]
+// #![feature(stmt_expr_attributes)]
+// #![feature(unboxed_closures)]
+// #![feature(assert_matches)]
+// #![feature(async_fn_in_dyn_trait)]
+// #![feature(async_for_loop)]
+// #![feature(auto_traits)]
+// #![feature(negative_impls)]
+
 pub mod discovery;
+pub mod keep_alive;
 pub mod swarm;

 /// Namespace for all the type/trait aliases used by this crate.
@@ -42,3 +54,11 @@ pub(crate) mod ext {
        }
    }
 }
+
+pub(crate) mod private {
+    #![allow(dead_code)]
+
+    /// Sealed traits support
+    pub trait Sealed {}
+    impl<T: ?Sized> Sealed for T {}
+}
--- a/src/exo/download/coordinator.py
+++ b/src/exo/download/coordinator.py
@@ -14,7 +14,6 @@ from exo.download.download_utils import (
    map_repo_download_progress_to_download_progress_data,
 )
 from exo.download.shard_downloader import ShardDownloader
-from exo.shared.constants import EXO_MODELS_DIR
 from exo.shared.models.model_cards import ModelId
 from exo.shared.types.commands import (
    CancelDownload,
@@ -47,7 +46,6 @@ class DownloadCoordinator:
    download_command_receiver: Receiver[ForwarderDownloadCommand]
    local_event_sender: Sender[ForwarderEvent]
    event_index_counter: Iterator[int]
-    offline: bool = False

    # Local state
    download_status: dict[ModelId, DownloadProgress] = field(default_factory=dict)
@@ -63,13 +61,8 @@ class DownloadCoordinator:

    def __post_init__(self) -> None:
        self.event_sender, self.event_receiver = channel[Event]()
-        if self.offline:
-            self.shard_downloader.set_internet_connection(False)
        self.shard_downloader.on_progress(self._download_progress_callback)

-    def _model_dir(self, model_id: ModelId) -> str:
-        return str(EXO_MODELS_DIR / model_id.normalize())
-
    async def _download_progress_callback(
        self, callback_shard: ShardMetadata, progress: RepoDownloadProgress
    ) -> None:
@@ -81,7 +74,6 @@ class DownloadCoordinator:
                shard_metadata=callback_shard,
                node_id=self.node_id,
                total_bytes=progress.total_bytes,
-                model_directory=self._model_dir(model_id),
            )
            self.download_status[model_id] = completed
            await self.event_sender.send(
@@ -101,7 +93,6 @@ class DownloadCoordinator:
                download_progress=map_repo_download_progress_to_download_progress_data(
                    progress
                ),
-                model_directory=self._model_dir(model_id),
            )
            self.download_status[model_id] = ongoing
            await self.event_sender.send(
@@ -110,17 +101,13 @@ class DownloadCoordinator:
            self._last_progress_time[model_id] = current_time()

    async def run(self) -> None:
-        logger.info(
-            f"Starting DownloadCoordinator{' (offline mode)' if self.offline else ''}"
-        )
-        if not self.offline:
-            self._test_internet_connection()
+        logger.info("Starting DownloadCoordinator")
+        self._test_internet_connection()
        async with self._tg as tg:
            tg.start_soon(self._command_processor)
            tg.start_soon(self._forward_events)
            tg.start_soon(self._emit_existing_download_progress)
-            if not self.offline:
-                tg.start_soon(self._check_internet_connection)
+            tg.start_soon(self._check_internet_connection)

    def _test_internet_connection(self) -> None:
        try:
@@ -183,11 +170,7 @@ class DownloadCoordinator:
                return

        # Emit pending status
-        progress = DownloadPending(
-            shard_metadata=shard,
-            node_id=self.node_id,
-            model_directory=self._model_dir(model_id),
-        )
+        progress = DownloadPending(shard_metadata=shard, node_id=self.node_id)
        self.download_status[model_id] = progress
        await self.event_sender.send(NodeDownloadProgress(download_progress=progress))

@@ -201,7 +184,6 @@ class DownloadCoordinator:
                shard_metadata=shard,
                node_id=self.node_id,
                total_bytes=initial_progress.total_bytes,
-                model_directory=self._model_dir(model_id),
            )
            self.download_status[model_id] = completed
            await self.event_sender.send(
@@ -209,20 +191,6 @@ class DownloadCoordinator:
            )
            return

-        if self.offline:
-            logger.warning(
-                f"Offline mode: model {model_id} is not fully available locally, cannot download"
-            )
-            failed = DownloadFailed(
-                shard_metadata=shard,
-                node_id=self.node_id,
-                error_message=f"Model files not found locally in offline mode: {model_id}",
-                model_directory=self._model_dir(model_id),
-            )
-            self.download_status[model_id] = failed
-            await self.event_sender.send(NodeDownloadProgress(download_progress=failed))
-            return
-
        # Start actual download
        self._start_download_task(shard, initial_progress)

@@ -238,7 +206,6 @@ class DownloadCoordinator:
            download_progress=map_repo_download_progress_to_download_progress_data(
                initial_progress
            ),
-            model_directory=self._model_dir(model_id),
        )
        self.download_status[model_id] = status
        self.event_sender.send_nowait(NodeDownloadProgress(download_progress=status))
@@ -252,7 +219,6 @@ class DownloadCoordinator:
                    shard_metadata=shard,
                    node_id=self.node_id,
                    error_message=str(e),
-                    model_directory=self._model_dir(model_id),
                )
                self.download_status[model_id] = failed
                await self.event_sender.send(
@@ -287,7 +253,6 @@ class DownloadCoordinator:
            pending = DownloadPending(
                shard_metadata=current_status.shard_metadata,
                node_id=self.node_id,
-                model_directory=self._model_dir(model_id),
            )
            await self.event_sender.send(
                NodeDownloadProgress(download_progress=pending)
@@ -330,18 +295,11 @@ class DownloadCoordinator:
                            node_id=self.node_id,
                            shard_metadata=progress.shard,
                            total_bytes=progress.total_bytes,
-                            model_directory=self._model_dir(
-                                progress.shard.model_card.model_id
-                            ),
                        )
                    elif progress.status in ["in_progress", "not_started"]:
                        if progress.downloaded_bytes_this_session.in_bytes == 0:
                            status = DownloadPending(
-                                node_id=self.node_id,
-                                shard_metadata=progress.shard,
-                                model_directory=self._model_dir(
-                                    progress.shard.model_card.model_id
-                                ),
+                                node_id=self.node_id, shard_metadata=progress.shard
                            )
                        else:
                            status = DownloadOngoing(
@@ -350,9 +308,6 @@ class DownloadCoordinator:
                                download_progress=map_repo_download_progress_to_download_progress_data(
                                    progress
                                ),
-                                model_directory=self._model_dir(
-                                    progress.shard.model_card.model_id
-                                ),
                            )
                    else:
                        continue
--- a/src/exo/download/download_utils.py
+++ b/src/exo/download/download_utils.py
@@ -448,13 +448,12 @@ async def download_file_with_retry(
    target_dir: Path,
    on_progress: Callable[[int, int, bool], None] = lambda _, __, ___: None,
    on_connection_lost: Callable[[], None] = lambda: None,
-    skip_internet: bool = False,
 ) -> Path:
    n_attempts = 3
    for attempt in range(n_attempts):
        try:
            return await _download_file(
-                model_id, revision, path, target_dir, on_progress, skip_internet
+                model_id, revision, path, target_dir, on_progress
            )
        except HuggingFaceAuthenticationError:
            raise
@@ -488,14 +487,10 @@ async def _download_file(
    path: str,
    target_dir: Path,
    on_progress: Callable[[int, int, bool], None] = lambda _, __, ___: None,
-    skip_internet: bool = False,
 ) -> Path:
    target_path = target_dir / path

    if await aios.path.exists(target_path):
-        if skip_internet:
-            return target_path
-
        local_size = (await aios.stat(target_path)).st_size

        # Try to verify against remote, but allow offline operation
@@ -515,11 +510,6 @@ async def _download_file(
            )
            return target_path

-    if skip_internet:
-        raise FileNotFoundError(
-            f"File {path} not found locally and cannot download in offline mode"
-        )
-
    await aios.makedirs((target_dir / path).parent, exist_ok=True)
    length, etag = await file_meta(model_id, revision, path)
    remote_hash = etag[:-5] if etag.endswith("-gzip") else etag
@@ -824,7 +814,6 @@ async def download_shard(
                    file, curr_bytes, total_bytes, is_renamed
                ),
                on_connection_lost=on_connection_lost,
-                skip_internet=skip_internet,
            )

    if not skip_download:
--- a/src/exo/download/tests/test_offline_mode.py
+++ b/src/exo/download/tests/test_offline_mode.py
@@ -1,230 +0,0 @@
-"""Tests for offline/air-gapped mode."""
-
-from collections.abc import AsyncIterator
-from pathlib import Path
-from unittest.mock import AsyncMock, patch
-
-import aiofiles
-import aiofiles.os as aios
-import pytest
-
-from exo.download.download_utils import (
-    _download_file,  # pyright: ignore[reportPrivateUsage]
-    download_file_with_retry,
-    fetch_file_list_with_cache,
-)
-from exo.shared.types.common import ModelId
-from exo.shared.types.worker.downloads import FileListEntry
-
-
-@pytest.fixture
-def model_id() -> ModelId:
-    return ModelId("test-org/test-model")
-
-
-@pytest.fixture
-async def temp_models_dir(tmp_path: Path) -> AsyncIterator[Path]:
-    models_dir = tmp_path / "models"
-    await aios.makedirs(models_dir, exist_ok=True)
-    with patch("exo.download.download_utils.EXO_MODELS_DIR", models_dir):
-        yield models_dir
-
-
-class TestDownloadFileOffline:
-    """Tests for _download_file with skip_internet=True."""
-
-    async def test_returns_local_file_without_http_verification(
-        self, model_id: ModelId, tmp_path: Path
-    ) -> None:
-        """When skip_internet=True and file exists locally, return it immediately
-        without making any HTTP calls (no file_meta verification)."""
-        target_dir = tmp_path / "downloads"
-        await aios.makedirs(target_dir, exist_ok=True)
-
-        local_file = target_dir / "model.safetensors"
-        async with aiofiles.open(local_file, "wb") as f:
-            await f.write(b"model weights data")
-
-        with patch(
-            "exo.download.download_utils.file_meta",
-            new_callable=AsyncMock,
-        ) as mock_file_meta:
-            result = await _download_file(
-                model_id,
-                "main",
-                "model.safetensors",
-                target_dir,
-                skip_internet=True,
-            )
-
-            assert result == local_file
-            mock_file_meta.assert_not_called()
-
-    async def test_raises_file_not_found_for_missing_file(
-        self, model_id: ModelId, tmp_path: Path
-    ) -> None:
-        """When skip_internet=True and file does NOT exist locally,
-        raise FileNotFoundError instead of attempting download."""
-        target_dir = tmp_path / "downloads"
-        await aios.makedirs(target_dir, exist_ok=True)
-
-        with pytest.raises(FileNotFoundError, match="offline mode"):
-            await _download_file(
-                model_id,
-                "main",
-                "missing_model.safetensors",
-                target_dir,
-                skip_internet=True,
-            )
-
-    async def test_returns_local_file_in_subdirectory(
-        self, model_id: ModelId, tmp_path: Path
-    ) -> None:
-        """When skip_internet=True and file exists in a subdirectory,
-        return it without HTTP calls."""
-        target_dir = tmp_path / "downloads"
-        subdir = target_dir / "transformer"
-        await aios.makedirs(subdir, exist_ok=True)
-
-        local_file = subdir / "diffusion_pytorch_model.safetensors"
-        async with aiofiles.open(local_file, "wb") as f:
-            await f.write(b"weights")
-
-        with patch(
-            "exo.download.download_utils.file_meta",
-            new_callable=AsyncMock,
-        ) as mock_file_meta:
-            result = await _download_file(
-                model_id,
-                "main",
-                "transformer/diffusion_pytorch_model.safetensors",
-                target_dir,
-                skip_internet=True,
-            )
-
-            assert result == local_file
-            mock_file_meta.assert_not_called()
-
-
-class TestDownloadFileWithRetryOffline:
-    """Tests for download_file_with_retry with skip_internet=True."""
-
-    async def test_propagates_skip_internet_to_download_file(
-        self, model_id: ModelId, tmp_path: Path
-    ) -> None:
-        """Verify skip_internet is passed through to _download_file."""
-        target_dir = tmp_path / "downloads"
-        await aios.makedirs(target_dir, exist_ok=True)
-
-        local_file = target_dir / "config.json"
-        async with aiofiles.open(local_file, "wb") as f:
-            await f.write(b'{"model_type": "qwen2"}')
-
-        with patch(
-            "exo.download.download_utils.file_meta",
-            new_callable=AsyncMock,
-        ) as mock_file_meta:
-            result = await download_file_with_retry(
-                model_id,
-                "main",
-                "config.json",
-                target_dir,
-                skip_internet=True,
-            )
-
-            assert result == local_file
-            mock_file_meta.assert_not_called()
-
-    async def test_file_not_found_does_not_retry(
-        self, model_id: ModelId, tmp_path: Path
-    ) -> None:
-        """FileNotFoundError from offline mode should not trigger retries."""
-        target_dir = tmp_path / "downloads"
-        await aios.makedirs(target_dir, exist_ok=True)
-
-        with pytest.raises(FileNotFoundError):
-            await download_file_with_retry(
-                model_id,
-                "main",
-                "nonexistent.safetensors",
-                target_dir,
-                skip_internet=True,
-            )
-
-
-class TestFetchFileListOffline:
-    """Tests for fetch_file_list_with_cache with skip_internet=True."""
-
-    async def test_uses_cached_file_list(
-        self, model_id: ModelId, temp_models_dir: Path
-    ) -> None:
-        """When skip_internet=True and cache file exists, use it without network."""
-        from pydantic import TypeAdapter
-
-        cache_dir = temp_models_dir / "caches" / model_id.normalize()
-        await aios.makedirs(cache_dir, exist_ok=True)
-
-        cached_list = [
-            FileListEntry(type="file", path="model.safetensors", size=1000),
-            FileListEntry(type="file", path="config.json", size=200),
-        ]
-        cache_file = cache_dir / f"{model_id.normalize()}--main--file_list.json"
-        async with aiofiles.open(cache_file, "w") as f:
-            await f.write(
-                TypeAdapter(list[FileListEntry]).dump_json(cached_list).decode()
-            )
-
-        with patch(
-            "exo.download.download_utils.fetch_file_list_with_retry",
-            new_callable=AsyncMock,
-        ) as mock_fetch:
-            result = await fetch_file_list_with_cache(
-                model_id, "main", skip_internet=True
-            )
-
-            assert result == cached_list
-            mock_fetch.assert_not_called()
-
-    async def test_falls_back_to_local_directory_scan(
-        self, model_id: ModelId, temp_models_dir: Path
-    ) -> None:
-        """When skip_internet=True and no cache but local files exist,
-        build file list from local directory."""
-        import json
-
-        model_dir = temp_models_dir / model_id.normalize()
-        await aios.makedirs(model_dir, exist_ok=True)
-
-        async with aiofiles.open(model_dir / "config.json", "w") as f:
-            await f.write('{"model_type": "qwen2"}')
-
-        index_data = {
-            "metadata": {},
-            "weight_map": {"model.layers.0.weight": "model.safetensors"},
-        }
-        async with aiofiles.open(model_dir / "model.safetensors.index.json", "w") as f:
-            await f.write(json.dumps(index_data))
-
-        async with aiofiles.open(model_dir / "model.safetensors", "wb") as f:
-            await f.write(b"x" * 500)
-
-        with patch(
-            "exo.download.download_utils.fetch_file_list_with_retry",
-            new_callable=AsyncMock,
-        ) as mock_fetch:
-            result = await fetch_file_list_with_cache(
-                model_id, "main", skip_internet=True
-            )
-
-            mock_fetch.assert_not_called()
-            paths = {entry.path for entry in result}
-            assert "config.json" in paths
-            assert "model.safetensors" in paths
-
-    async def test_raises_when_no_cache_and_no_local_files(
-        self, model_id: ModelId, temp_models_dir: Path
-    ) -> None:
-        """When skip_internet=True and neither cache nor local files exist,
-        raise FileNotFoundError."""
-        with pytest.raises(FileNotFoundError, match="No internet"):
-            await fetch_file_list_with_cache(model_id, "main", skip_internet=True)
--- a/src/exo/main.py
+++ b/src/exo/main.py
@@ -39,7 +39,6 @@ class Node:

    node_id: NodeId
    event_index_counter: Iterator[int]
-    offline: bool
    _tg: TaskGroup = field(init=False, default_factory=anyio.create_task_group)

    @classmethod
@@ -69,7 +68,6 @@ class Node:
                download_command_receiver=router.receiver(topics.DOWNLOAD_COMMANDS),
                local_event_sender=router.sender(topics.LOCAL_EVENTS),
                event_index_counter=event_index_counter,
-                offline=args.offline,
            )
        else:
            download_coordinator = None
@@ -134,13 +132,10 @@ class Node:
            api,
            node_id,
            event_index_counter,
-            args.offline,
        )

    async def run(self):
        async with self._tg as tg:
-            signal.signal(signal.SIGINT, lambda _, __: self.shutdown())
-            signal.signal(signal.SIGTERM, lambda _, __: self.shutdown())
            tg.start_soon(self.router.run)
            tg.start_soon(self.election.run)
            if self.download_coordinator:
@@ -152,6 +147,8 @@ class Node:
            if self.api:
                tg.start_soon(self.api.run)
            tg.start_soon(self._elect_loop)
+            signal.signal(signal.SIGINT, lambda _, __: self.shutdown())
+            signal.signal(signal.SIGTERM, lambda _, __: self.shutdown())

    def shutdown(self):
        # if this is our second call to shutdown, just sys.exit
@@ -225,7 +222,6 @@ class Node:
                            ),
                            local_event_sender=self.router.sender(topics.LOCAL_EVENTS),
                            event_index_counter=self.event_index_counter,
-                            offline=self.offline,
                        )
                        self._tg.start_soon(self.download_coordinator.run)
                    if self.worker:
@@ -264,9 +260,6 @@ def main():
    logger.info("Starting EXO")
    logger.info(f"EXO_LIBP2P_NAMESPACE: {os.getenv('EXO_LIBP2P_NAMESPACE')}")

-    if args.offline:
-        logger.info("Running in OFFLINE mode — no internet checks, local models only")
-
    # Set FAST_SYNCH override env var for runner subprocesses
    if args.fast_synch is True:
        os.environ["EXO_FAST_SYNCH"] = "on"
@@ -289,7 +282,6 @@ class Args(CamelCaseModel):
    tb_only: bool = False
    no_worker: bool = False
    no_downloads: bool = False
-    offline: bool = False
    fast_synch: bool | None = None  # None = auto, True = force on, False = force off

    @classmethod
@@ -337,11 +329,6 @@ class Args(CamelCaseModel):
            action="store_true",
            help="Disable the download coordinator (node won't download models)",
        )
-        parser.add_argument(
-            "--offline",
-            action="store_true",
-            help="Run in offline/air-gapped mode: skip internet checks, use only pre-staged local models",
-        )
        fast_synch_group = parser.add_mutually_exclusive_group()
        fast_synch_group.add_argument(
            "--fast-synch",
--- a/src/exo/master/adapters/chat_completions.py
+++ b/src/exo/master/adapters/chat_completions.py
@@ -17,7 +17,6 @@ from exo.shared.types.api import (
    LogprobsContentItem,
    StreamingChoiceResponse,
    ToolCall,
-    Usage,
 )
 from exo.shared.types.chunks import ErrorChunk, TokenChunk, ToolCallChunk
 from exo.shared.types.common import CommandId
@@ -126,8 +125,6 @@ async def generate_chat_stream(
    chunk_stream: AsyncGenerator[ErrorChunk | ToolCallChunk | TokenChunk, None],
 ) -> AsyncGenerator[str, None]:
    """Generate Chat Completions API streaming events from chunks."""
-    last_usage: Usage | None = None
-
    async for chunk in chunk_stream:
        if isinstance(chunk, ErrorChunk):
            error_response = ErrorResponse(
@@ -141,8 +138,6 @@ async def generate_chat_stream(
            yield "data: [DONE]\n\n"
            return

-        last_usage = chunk.usage or last_usage
-
        if isinstance(chunk, ToolCallChunk):
            tool_call_deltas = [
                ToolCall(
@@ -166,15 +161,12 @@ async def generate_chat_stream(
                        finish_reason="tool_calls",
                    )
                ],
-                usage=last_usage,
            )
            yield f"data: {tool_response.model_dump_json()}\n\n"
            yield "data: [DONE]\n\n"
            return

        chunk_response = chunk_to_response(chunk, command_id)
-        if chunk.finish_reason is not None:
-            chunk_response = chunk_response.model_copy(update={"usage": last_usage})
        yield f"data: {chunk_response.model_dump_json()}\n\n"

        if chunk.finish_reason is not None:
@@ -184,9 +176,7 @@ async def generate_chat_stream(
 async def collect_chat_response(
    command_id: CommandId,
    chunk_stream: AsyncGenerator[ErrorChunk | ToolCallChunk | TokenChunk, None],
-) -> AsyncGenerator[str]:
-    # This is an AsyncGenerator[str] rather than returning a ChatCompletionReponse because
-    # FastAPI handles the cancellation better but wouldn't auto-serialize for some reason
+) -> ChatCompletionResponse:
    """Collect all token chunks and return a single ChatCompletionResponse."""
    text_parts: list[str] = []
    tool_calls: list[ToolCall] = []
@@ -194,7 +184,6 @@ async def collect_chat_response(
    model: str | None = None
    finish_reason: FinishReason | None = None
    error_message: str | None = None
-    last_usage: Usage | None = None

    async for chunk in chunk_stream:
        if isinstance(chunk, ErrorChunk):
@@ -204,8 +193,6 @@ async def collect_chat_response(
        if model is None:
            model = chunk.model

-        last_usage = chunk.usage or last_usage
-
        if isinstance(chunk, TokenChunk):
            text_parts.append(chunk.text)
            if chunk.logprob is not None:
@@ -236,7 +223,7 @@ async def collect_chat_response(
    combined_text = "".join(text_parts)
    assert model is not None

-    yield ChatCompletionResponse(
+    return ChatCompletionResponse(
        id=command_id,
        created=int(time.time()),
        model=model,
@@ -254,6 +241,4 @@ async def collect_chat_response(
                finish_reason=finish_reason,
            )
        ],
-        usage=last_usage,
-    ).model_dump_json()
-    return
+    )
--- a/src/exo/master/adapters/claude.py
+++ b/src/exo/master/adapters/claude.py
@@ -4,7 +4,7 @@ import json
 from collections.abc import AsyncGenerator
 from typing import Any

-from exo.shared.types.api import FinishReason, Usage
+from exo.shared.types.api import FinishReason
 from exo.shared.types.chunks import ErrorChunk, TokenChunk, ToolCallChunk
 from exo.shared.types.claude_api import (
    ClaudeContentBlock,
@@ -161,14 +161,12 @@ async def collect_claude_response(
    command_id: CommandId,
    model: str,
    chunk_stream: AsyncGenerator[ErrorChunk | ToolCallChunk | TokenChunk, None],
-) -> AsyncGenerator[str]:
-    # This is an AsyncGenerator[str] rather than returning a ChatCompletionReponse because
-    # FastAPI handles the cancellation better but wouldn't auto-serialize for some reason
+) -> ClaudeMessagesResponse:
    """Collect all token chunks and return a single ClaudeMessagesResponse."""
    text_parts: list[str] = []
    tool_use_blocks: list[ClaudeToolUseBlock] = []
    stop_reason: ClaudeStopReason | None = None
-    last_usage: Usage | None = None
+    last_stats = None
    error_message: str | None = None

    async for chunk in chunk_stream:
@@ -176,8 +174,6 @@ async def collect_claude_response(
            error_message = chunk.error_message or "Internal server error"
            break

-        last_usage = chunk.usage or last_usage
-
        if isinstance(chunk, ToolCallChunk):
            for tool in chunk.tool_calls:
                tool_use_blocks.append(
@@ -187,10 +183,12 @@ async def collect_claude_response(
                        input=json.loads(tool.arguments),  # pyright: ignore[reportAny]
                    )
                )
+            last_stats = chunk.stats or last_stats
            stop_reason = "tool_use"
            continue

        text_parts.append(chunk.text)
+        last_stats = chunk.stats or last_stats

        if chunk.finish_reason is not None:
            stop_reason = finish_reason_to_claude_stop_reason(chunk.finish_reason)
@@ -210,11 +208,11 @@ async def collect_claude_response(
    if not content:
        content.append(ClaudeTextBlock(text=""))

-    # Use actual usage data if available
-    input_tokens = last_usage.prompt_tokens if last_usage else 0
-    output_tokens = last_usage.completion_tokens if last_usage else 0
+    # Use actual usage data from stats if available
+    input_tokens = last_stats.prompt_tokens if last_stats else 0
+    output_tokens = last_stats.generation_tokens if last_stats else 0

-    yield ClaudeMessagesResponse(
+    return ClaudeMessagesResponse(
        id=f"msg_{command_id}",
        model=model,
        content=content,
@@ -223,8 +221,7 @@ async def collect_claude_response(
            input_tokens=input_tokens,
            output_tokens=output_tokens,
        ),
-    ).model_dump_json()
-    return
+    )


 async def generate_claude_stream(
@@ -252,7 +249,7 @@ async def generate_claude_stream(

    output_tokens = 0
    stop_reason: ClaudeStopReason | None = None
-    last_usage: Usage | None = None
+    last_stats = None
    next_block_index = 1  # text block is 0, tool blocks start at 1

    async for chunk in chunk_stream:
@@ -260,9 +257,8 @@ async def generate_claude_stream(
            # Close text block and bail
            break

-        last_usage = chunk.usage or last_usage
-
        if isinstance(chunk, ToolCallChunk):
+            last_stats = chunk.stats or last_stats
            stop_reason = "tool_use"

            # Emit tool_use content blocks
@@ -294,6 +290,7 @@ async def generate_claude_stream(
            continue

        output_tokens += 1  # Count each chunk as one token
+        last_stats = chunk.stats or last_stats

        # content_block_delta
        delta_event = ClaudeContentBlockDeltaEvent(
@@ -305,9 +302,9 @@ async def generate_claude_stream(
        if chunk.finish_reason is not None:
            stop_reason = finish_reason_to_claude_stop_reason(chunk.finish_reason)

-    # Use actual token count from usage if available
-    if last_usage is not None:
-        output_tokens = last_usage.completion_tokens
+    # Use actual token count from stats if available
+    if last_stats is not None:
+        output_tokens = last_stats.generation_tokens

    # content_block_stop for text block
    block_stop = ClaudeContentBlockStopEvent(index=0)
--- a/src/exo/master/adapters/responses.py
+++ b/src/exo/master/adapters/responses.py
@@ -4,7 +4,6 @@ from collections.abc import AsyncGenerator
 from itertools import count
 from typing import Any

-from exo.shared.types.api import Usage
 from exo.shared.types.chunks import ErrorChunk, TokenChunk, ToolCallChunk
 from exo.shared.types.common import CommandId
 from exo.shared.types.openai_responses import (
@@ -122,15 +121,13 @@ async def collect_responses_response(
    command_id: CommandId,
    model: str,
    chunk_stream: AsyncGenerator[ErrorChunk | ToolCallChunk | TokenChunk, None],
-) -> AsyncGenerator[str]:
-    # This is an AsyncGenerator[str] rather than returning a ChatCompletionReponse because
-    # FastAPI handles the cancellation better but wouldn't auto-serialize for some reason
+) -> ResponsesResponse:
    """Collect all token chunks and return a single ResponsesResponse."""
    response_id = f"resp_{command_id}"
    item_id = f"item_{command_id}"
    accumulated_text = ""
    function_call_items: list[ResponseFunctionCallItem] = []
-    last_usage: Usage | None = None
+    last_stats = None
    error_message: str | None = None

    async for chunk in chunk_stream:
@@ -138,32 +135,32 @@ async def collect_responses_response(
            error_message = chunk.error_message or "Internal server error"
            break

-        last_usage = chunk.usage or last_usage
-
        if isinstance(chunk, ToolCallChunk):
            for tool in chunk.tool_calls:
                function_call_items.append(
                    ResponseFunctionCallItem(
-                        id=tool.id,
-                        call_id=tool.id,
+                        id=f"fc_{tool.id}",
+                        call_id=f"call_{tool.id}",
                        name=tool.name,
                        arguments=tool.arguments,
                    )
                )
+            last_stats = chunk.stats or last_stats
            continue

        accumulated_text += chunk.text
+        last_stats = chunk.stats or last_stats

    if error_message is not None:
        raise ValueError(error_message)

-    # Create usage from usage data if available
+    # Create usage from stats if available
    usage = None
-    if last_usage is not None:
+    if last_stats is not None:
        usage = ResponseUsage(
-            input_tokens=last_usage.prompt_tokens,
-            output_tokens=last_usage.completion_tokens,
-            total_tokens=last_usage.total_tokens,
+            input_tokens=last_stats.prompt_tokens,
+            output_tokens=last_stats.generation_tokens,
+            total_tokens=last_stats.prompt_tokens + last_stats.generation_tokens,
        )

    output: list[ResponseItem] = [
@@ -175,15 +172,14 @@ async def collect_responses_response(
    ]
    output.extend(function_call_items)

-    yield ResponsesResponse(
+    return ResponsesResponse(
        id=response_id,
        model=model,
        status="completed",
        output=output,
        output_text=accumulated_text,
        usage=usage,
-    ).model_dump_json()
-    return
+    )


 async def generate_responses_stream(
@@ -239,16 +235,15 @@ async def generate_responses_stream(

    accumulated_text = ""
    function_call_items: list[ResponseFunctionCallItem] = []
-    last_usage: Usage | None = None
+    last_stats = None
    next_output_index = 1  # message item is at 0

    async for chunk in chunk_stream:
        if isinstance(chunk, ErrorChunk):
            break

-        last_usage = chunk.usage or last_usage
-
        if isinstance(chunk, ToolCallChunk):
+            last_stats = chunk.stats or last_stats
            for tool in chunk.tool_calls:
                fc_id = f"fc_{tool.id}"
                call_id = f"call_{tool.id}"
@@ -307,6 +302,7 @@ async def generate_responses_stream(
            continue

        accumulated_text += chunk.text
+        last_stats = chunk.stats or last_stats

        # response.output_text.delta
        delta_event = ResponseTextDeltaEvent(
@@ -350,13 +346,13 @@ async def generate_responses_stream(
    )
    yield f"event: response.output_item.done\ndata: {item_done.model_dump_json()}\n\n"

-    # Create usage from usage data if available
+    # Create usage from stats if available
    usage = None
-    if last_usage is not None:
+    if last_stats is not None:
        usage = ResponseUsage(
-            input_tokens=last_usage.prompt_tokens,
-            output_tokens=last_usage.completion_tokens,
-            total_tokens=last_usage.total_tokens,
+            input_tokens=last_stats.prompt_tokens,
+            output_tokens=last_stats.generation_tokens,
+            total_tokens=last_stats.prompt_tokens + last_stats.generation_tokens,
        )

    # response.completed
--- a/src/exo/master/api.py
+++ b/src/exo/master/api.py
@@ -85,7 +85,6 @@ from exo.shared.types.api import (
    ImageGenerationTaskParams,
    ImageListItem,
    ImageListResponse,
-    ImageSize,
    ModelList,
    ModelListModel,
    PlaceInstanceParams,
@@ -101,7 +100,6 @@ from exo.shared.types.api import (
    TraceRankStats,
    TraceResponse,
    TraceStatsResponse,
-    normalize_image_size,
 )
 from exo.shared.types.chunks import (
    ErrorChunk,
@@ -127,7 +125,6 @@ from exo.shared.types.commands import (
    PlaceInstance,
    SendInputChunk,
    StartDownload,
-    TaskCancelled,
    TaskFinished,
    TextGeneration,
 )
@@ -543,14 +540,16 @@ class API:
                        break

        except anyio.get_cancelled_exc_class():
-            command = TaskCancelled(cancelled_command_id=command_id)
-            with anyio.CancelScope(shield=True):
-                await self.command_sender.send(
-                    ForwarderCommand(origin=self.node_id, command=command)
-                )
+            # TODO: TaskCancelled
+            """
+            self.command_sender.send_nowait(
+                ForwarderCommand(origin=self.node_id, command=command)
+            )
+            """
            raise
        finally:
-            await self._send(TaskFinished(finished_command_id=command_id))
+            command = TaskFinished(finished_command_id=command_id)
+            await self._send(command)
            if command_id in self._text_generation_queues:
                del self._text_generation_queues[command_id]

@@ -645,14 +644,11 @@ class API:
                    "X-Accel-Buffering": "no",
                },
            )
-        else:
-            return StreamingResponse(
-                collect_chat_response(
-                    command.command_id,
-                    self._token_chunk_stream(command.command_id),
-                ),
-                media_type="application/json",
-            )
+
+        return await collect_chat_response(
+            command.command_id,
+            self._token_chunk_stream(command.command_id),
+        )

    async def bench_chat_completions(
        self, payload: BenchChatCompletionRequest
@@ -668,7 +664,8 @@ class API:
        command = TextGeneration(task_params=task_params)
        await self._send(command)

-        return await self._collect_text_generation_with_stats(command.command_id)
+        response = await self._collect_text_generation_with_stats(command.command_id)
+        return response

    async def _resolve_and_validate_text_model(self, model_id: ModelId) -> ModelId:
        """Validate a text model exists and return the resolved model ID.
@@ -753,11 +750,9 @@ class API:
        When stream=True and partial_images > 0, returns a StreamingResponse
        with SSE-formatted events for partial and final images.
        """
+        payload.model = await self._validate_image_model(ModelId(payload.model))
        payload = payload.model_copy(
-            update={
-                "model": await self._validate_image_model(ModelId(payload.model)),
-                "advanced_params": _ensure_seed(payload.advanced_params),
-            }
+            update={"advanced_params": _ensure_seed(payload.advanced_params)}
        )

        command = ImageGeneration(
@@ -888,11 +883,6 @@ class API:
                        del image_metadata[key]

        except anyio.get_cancelled_exc_class():
-            command = TaskCancelled(cancelled_command_id=command_id)
-            with anyio.CancelScope(shield=True):
-                await self.command_sender.send(
-                    ForwarderCommand(origin=self.node_id, command=command)
-                )
            raise
        finally:
            await self._send(TaskFinished(finished_command_id=command_id))
@@ -974,11 +964,6 @@ class API:

            return (images, stats if capture_stats else None)
        except anyio.get_cancelled_exc_class():
-            command = TaskCancelled(cancelled_command_id=command_id)
-            with anyio.CancelScope(shield=True):
-                await self.command_sender.send(
-                    ForwarderCommand(origin=self.node_id, command=command)
-                )
            raise
        finally:
            await self._send(TaskFinished(finished_command_id=command_id))
@@ -1013,13 +998,12 @@ class API:
    async def bench_image_generations(
        self, request: Request, payload: BenchImageGenerationTaskParams
    ) -> BenchImageGenerationResponse:
+        payload.model = await self._validate_image_model(ModelId(payload.model))
+
+        payload.stream = False
+        payload.partial_images = 0
        payload = payload.model_copy(
-            update={
-                "model": await self._validate_image_model(ModelId(payload.model)),
-                "stream": False,
-                "partial_images": 0,
-                "advanced_params": _ensure_seed(payload.advanced_params),
-            }
+            update={"advanced_params": _ensure_seed(payload.advanced_params)}
        )

        command = ImageGeneration(
@@ -1040,7 +1024,7 @@ class API:
        prompt: str,
        model: ModelId,
        n: int,
-        size: ImageSize,
+        size: str,
        response_format: Literal["url", "b64_json"],
        input_fidelity: Literal["low", "high"],
        stream: bool,
@@ -1110,7 +1094,7 @@ class API:
        prompt: str = Form(...),
        model: str = Form(...),
        n: int = Form(1),
-        size: str | None = Form(None),
+        size: str = Form("1024x1024"),
        response_format: Literal["url", "b64_json"] = Form("b64_json"),
        input_fidelity: Literal["low", "high"] = Form("low"),
        stream: str = Form("false"),
@@ -1136,7 +1120,7 @@ class API:
            prompt=prompt,
            model=ModelId(model),
            n=n,
-            size=normalize_image_size(size),
+            size=size,
            response_format=response_format,
            input_fidelity=input_fidelity,
            stream=stream_bool,
@@ -1172,7 +1156,7 @@ class API:
        prompt: str = Form(...),
        model: str = Form(...),
        n: int = Form(1),
-        size: str | None = Form(None),
+        size: str = Form("1024x1024"),
        response_format: Literal["url", "b64_json"] = Form("b64_json"),
        input_fidelity: Literal["low", "high"] = Form("low"),
        quality: Literal["high", "medium", "low"] = Form("medium"),
@@ -1192,7 +1176,7 @@ class API:
            prompt=prompt,
            model=ModelId(model),
            n=n,
-            size=normalize_image_size(size),
+            size=size,
            response_format=response_format,
            input_fidelity=input_fidelity,
            stream=False,
@@ -1237,15 +1221,12 @@ class API:
                    "X-Accel-Buffering": "no",
                },
            )
-        else:
-            return StreamingResponse(
-                collect_claude_response(
-                    command.command_id,
-                    payload.model,
-                    self._token_chunk_stream(command.command_id),
-                ),
-                media_type="application/json",
-            )
+
+        return await collect_claude_response(
+            command.command_id,
+            payload.model,
+            self._token_chunk_stream(command.command_id),
+        )

    async def openai_responses(
        self, payload: ResponsesRequest
@@ -1273,15 +1254,11 @@ class API:
                },
            )

-        else:
-            return StreamingResponse(
-                collect_responses_response(
-                    command.command_id,
-                    payload.model,
-                    self._token_chunk_stream(command.command_id),
-                ),
-                media_type="application/json",
-            )
+        return await collect_responses_response(
+            command.command_id,
+            payload.model,
+            self._token_chunk_stream(command.command_id),
+        )

    def _calculate_total_available_memory(self) -> Memory:
        """Calculate total available memory across all nodes in bytes."""
--- a/src/exo/master/main.py
+++ b/src/exo/master/main.py
@@ -24,7 +24,7 @@ from exo.shared.types.commands import (
    PlaceInstance,
    RequestEventLog,
    SendInputChunk,
-    TaskCancelled,
+    SetInstanceDraftModel,
    TaskFinished,
    TestCommand,
    TextGeneration,
@@ -36,11 +36,11 @@ from exo.shared.types.events import (
    IndexedEvent,
    InputChunkReceived,
    InstanceDeleted,
+    InstanceDraftModelUpdated,
    NodeGatheredInfo,
    NodeTimedOut,
    TaskCreated,
    TaskDeleted,
-    TaskStatusUpdated,
    TraceEventData,
    TracesCollected,
    TracesMerged,
@@ -281,7 +281,7 @@ class Master:
                        case DeleteInstance():
                            placement = delete_instance(command, self.state.instances)
                            transition_events = get_transition_events(
-                                self.state.instances, placement, self.state.tasks
+                                self.state.instances, placement
                            )
                            for cmd in cancel_unnecessary_downloads(
                                placement, self.state.downloads
@@ -301,7 +301,7 @@ class Master:
                                self.state.node_network,
                            )
                            transition_events = get_transition_events(
-                                self.state.instances, placement, self.state.tasks
+                                self.state.instances, placement
                            )
                            generated_events.extend(transition_events)
                        case CreateInstance():
@@ -311,7 +311,7 @@ class Master:
                                self.state.instances,
                            )
                            transition_events = get_transition_events(
-                                self.state.instances, placement, self.state.tasks
+                                self.state.instances, placement
                            )
                            generated_events.extend(transition_events)
                        case SendInputChunk(chunk=chunk):
@@ -321,18 +321,14 @@ class Master:
                                    chunk=chunk,
                                )
                            )
-                        case TaskCancelled():
-                            if (
-                                task_id := self.command_task_mapping.get(
-                                    command.cancelled_command_id
-                                )
-                            ) is not None:
-                                generated_events.append(
-                                    TaskStatusUpdated(
-                                        task_status=TaskStatus.Cancelled,
-                                        task_id=task_id,
-                                    )
+                        case SetInstanceDraftModel():
+                            generated_events.append(
+                                InstanceDraftModelUpdated(
+                                    instance_id=command.instance_id,
+                                    draft_model=command.draft_model,
+                                    num_draft_tokens=command.num_draft_tokens,
                                )
+                            )
                        case TaskFinished():
                            generated_events.append(
                                TaskDeleted(
@@ -341,9 +337,10 @@ class Master:
                                    ]
                                )
                            )
-                            self.command_task_mapping.pop(
-                                command.finished_command_id, None
-                            )
+                            if command.finished_command_id in self.command_task_mapping:
+                                del self.command_task_mapping[
+                                    command.finished_command_id
+                                ]
                        case RequestEventLog():
                            # We should just be able to send everything, since other buffers will ignore old messages
                            # rate limit to 1000 at a time
--- a/src/exo/master/placement.py
+++ b/src/exo/master/placement.py
@@ -22,15 +22,9 @@ from exo.shared.types.commands import (
    PlaceInstance,
 )
 from exo.shared.types.common import NodeId
-from exo.shared.types.events import (
-    Event,
-    InstanceCreated,
-    InstanceDeleted,
-    TaskStatusUpdated,
-)
+from exo.shared.types.events import Event, InstanceCreated, InstanceDeleted
 from exo.shared.types.memory import Memory
 from exo.shared.types.profiling import MemoryUsage, NodeNetworkInfo
-from exo.shared.types.tasks import Task, TaskId, TaskStatus
 from exo.shared.types.worker.downloads import (
    DownloadOngoing,
    DownloadProgress,
@@ -159,6 +153,8 @@ def place_instance(
                shard_assignments=shard_assignments,
                jaccl_devices=mlx_jaccl_devices,
                jaccl_coordinators=mlx_jaccl_coordinators,
+                draft_model=command.draft_model,
+                num_draft_tokens=command.num_draft_tokens,
            )
        case InstanceMeta.MlxRing:
            ephemeral_port = random_ephemeral_port()
@@ -173,6 +169,8 @@ def place_instance(
                shard_assignments=shard_assignments,
                hosts_by_node=hosts_by_node,
                ephemeral_port=ephemeral_port,
+                draft_model=command.draft_model,
+                num_draft_tokens=command.num_draft_tokens,
            )

    return target_instances
@@ -192,7 +190,6 @@ def delete_instance(
 def get_transition_events(
    current_instances: Mapping[InstanceId, Instance],
    target_instances: Mapping[InstanceId, Instance],
-    tasks: Mapping[TaskId, Task],
 ) -> Sequence[Event]:
    events: list[Event] = []

@@ -208,18 +205,6 @@ def get_transition_events(
    # find instances to delete
    for instance_id in current_instances:
        if instance_id not in target_instances:
-            for task in tasks.values():
-                if task.instance_id == instance_id and task.task_status in [
-                    TaskStatus.Pending,
-                    TaskStatus.Running,
-                ]:
-                    events.append(
-                        TaskStatusUpdated(
-                            task_status=TaskStatus.Cancelled,
-                            task_id=task.task_id,
-                        )
-                    )
-
            events.append(
                InstanceDeleted(
                    instance_id=instance_id,
--- a/src/exo/master/tests/test_claude_tool_use.py
+++ b/src/exo/master/tests/test_claude_tool_use.py
@@ -4,11 +4,7 @@ import json
 from collections.abc import AsyncGenerator
 from typing import Any, cast

-from exo.master.adapters.claude import (
-    ClaudeMessagesResponse,
-    collect_claude_response,
-    generate_claude_stream,
-)
+from exo.master.adapters.claude import collect_claude_response, generate_claude_stream
 from exo.shared.types.api import ToolCallItem
 from exo.shared.types.chunks import ErrorChunk, TokenChunk, ToolCallChunk
 from exo.shared.types.common import CommandId, ModelId
@@ -21,18 +17,6 @@ async def _chunks_to_stream(
        yield chunk


-async def _collect_response(
-    command_id: CommandId,
-    model: str,
-    chunk_stream: AsyncGenerator[ErrorChunk | ToolCallChunk | TokenChunk, None],
-) -> ClaudeMessagesResponse:
-    """Helper to consume the async generator and parse the JSON response."""
-    parts: list[str] = []
-    async for part in collect_claude_response(command_id, model, chunk_stream):
-        parts.append(part)
-    return ClaudeMessagesResponse.model_validate_json("".join(parts))
-
-
 MODEL = ModelId("test-model")
 COMMAND_ID = CommandId("cmd_test123")

@@ -63,7 +47,7 @@ class TestCollectClaudeResponseToolUse:
                ],
            ),
        ]
-        response = await _collect_response(
+        response = await collect_claude_response(
            COMMAND_ID, "test-model", _chunks_to_stream(chunks)
        )

@@ -93,7 +77,7 @@ class TestCollectClaudeResponseToolUse:
                ],
            ),
        ]
-        response = await _collect_response(
+        response = await collect_claude_response(
            COMMAND_ID, "test-model", _chunks_to_stream(chunks)
        )

@@ -118,7 +102,7 @@ class TestCollectClaudeResponseToolUse:
                ],
            ),
        ]
-        response = await _collect_response(
+        response = await collect_claude_response(
            COMMAND_ID, "test-model", _chunks_to_stream(chunks)
        )

@@ -132,7 +116,7 @@ class TestCollectClaudeResponseToolUse:

    async def test_no_content_produces_empty_text_block(self):
        chunks: list[ErrorChunk | ToolCallChunk | TokenChunk] = []
-        response = await _collect_response(
+        response = await collect_claude_response(
            COMMAND_ID, "test-model", _chunks_to_stream(chunks)
        )
        assert len(response.content) == 1
--- a/src/exo/master/tests/test_placement.py
+++ b/src/exo/master/tests/test_placement.py
@@ -239,7 +239,7 @@ def test_get_transition_events_no_change(instance: Instance):
    target_instances = {instance_id: instance}

    # act
-    events = get_transition_events(current_instances, target_instances, {})
+    events = get_transition_events(current_instances, target_instances)

    # assert
    assert len(events) == 0
@@ -252,7 +252,7 @@ def test_get_transition_events_create_instance(instance: Instance):
    target_instances: dict[InstanceId, Instance] = {instance_id: instance}

    # act
-    events = get_transition_events(current_instances, target_instances, {})
+    events = get_transition_events(current_instances, target_instances)

    # assert
    assert len(events) == 1
@@ -266,7 +266,7 @@ def test_get_transition_events_delete_instance(instance: Instance):
    target_instances: dict[InstanceId, Instance] = {}

    # act
-    events = get_transition_events(current_instances, target_instances, {})
+    events = get_transition_events(current_instances, target_instances)

    # assert
    assert len(events) == 1
--- a/src/exo/shared/apply.py
+++ b/src/exo/shared/apply.py
@@ -12,6 +12,7 @@ from exo.shared.types.events import (
    InputChunkReceived,
    InstanceCreated,
    InstanceDeleted,
+    InstanceDraftModelUpdated,
    NodeDownloadProgress,
    NodeGatheredInfo,
    NodeTimedOut,
@@ -72,6 +73,8 @@ def event_apply(event: Event, state: State) -> State:
            return apply_instance_created(event, state)
        case InstanceDeleted():
            return apply_instance_deleted(event, state)
+        case InstanceDraftModelUpdated():
+            return apply_instance_draft_model_updated(event, state)
        case NodeTimedOut():
            return apply_node_timed_out(event, state)
        case NodeDownloadProgress():
@@ -190,6 +193,25 @@ def apply_instance_deleted(event: InstanceDeleted, state: State) -> State:
    return state.model_copy(update={"instances": new_instances})


+def apply_instance_draft_model_updated(
+    event: InstanceDraftModelUpdated, state: State
+) -> State:
+    if event.instance_id not in state.instances:
+        return state
+    instance = state.instances[event.instance_id]
+    updated_instance = instance.model_copy(
+        update={
+            "draft_model": event.draft_model,
+            "num_draft_tokens": event.num_draft_tokens,
+        }
+    )
+    new_instances: Mapping[InstanceId, Instance] = {
+        **state.instances,
+        event.instance_id: updated_instance,
+    }
+    return state.model_copy(update={"instances": new_instances})
+
+
 def apply_runner_status_updated(event: RunnerStatusUpdated, state: State) -> State:
    new_runners: Mapping[RunnerId, RunnerStatus] = {
        **state.runners,
@@ -218,6 +240,11 @@ def apply_node_timed_out(event: NodeTimedOut, state: State) -> State:
        key: value for key, value in state.downloads.items() if key != event.node_id
    }
    # Clean up all granular node mappings
+    node_identities = {
+        key: value
+        for key, value in state.node_identities.items()
+        if key != event.node_id
+    }
    node_memory = {
        key: value for key, value in state.node_memory.items() if key != event.node_id
    }
@@ -258,6 +285,7 @@ def apply_node_timed_out(event: NodeTimedOut, state: State) -> State:
            "downloads": downloads,
            "topology": topology,
            "last_seen": last_seen,
+            "node_identities": node_identities,
            "node_memory": node_memory,
            "node_disk": node_disk,
            "node_system": node_system,
--- a/src/exo/shared/models/model_cards.py
+++ b/src/exo/shared/models/model_cards.py
@@ -44,8 +44,7 @@ async def _refresh_card_cache():
        async for toml_file in path.rglob("*.toml"):
            try:
                card = await ModelCard.load_from_path(toml_file)
-                if card.model_id not in _card_cache:
-                    _card_cache[card.model_id] = card
+                _card_cache[card.model_id] = card
            except (ValidationError, TOMLKitError):
                pass

@@ -183,7 +182,6 @@ class ConfigData(BaseModel):
    def supports_tensor(self) -> bool:
        return self.architectures in [
            ["Glm4MoeLiteForCausalLM"],
-            ["GlmMoeDsaForCausalLM"],
            ["DeepseekV32ForCausalLM"],
            ["DeepseekV3ForCausalLM"],
            ["Qwen3NextForCausalLM"],
--- a/src/exo/shared/types/api.py
+++ b/src/exo/shared/types/api.py
@@ -1,9 +1,10 @@
 import time
 from collections.abc import Generator
-from typing import Annotated, Any, Literal, get_args
+from typing import Annotated, Any, Literal
 from uuid import uuid4

 from pydantic import BaseModel, Field, field_validator
+from pydantic_core import PydanticUseDefault

 from exo.shared.models.model_cards import ModelCard, ModelId
 from exo.shared.types.common import CommandId, NodeId
@@ -227,6 +228,13 @@ class PlaceInstanceParams(BaseModel):
    instance_meta: InstanceMeta = InstanceMeta.MlxRing
    min_nodes: int = 1

+    @field_validator("sharding", "instance_meta", mode="plain")
+    @classmethod
+    def use_default(cls, v: object):
+        if not v or not isinstance(v, (Sharding, InstanceMeta)):
+            raise PydanticUseDefault()
+        return v
+

 class CreateInstanceParams(BaseModel):
    instance: Instance
@@ -262,27 +270,6 @@ class DeleteInstanceResponse(BaseModel):
    instance_id: InstanceId


-ImageSize = Literal[
-    "auto",
-    "512x512",
-    "768x768",
-    "1024x768",
-    "768x1024",
-    "1024x1024",
-    "1024x1536",
-    "1536x1024",
-]
-
-
-def normalize_image_size(v: object) -> ImageSize:
-    """Shared validator for ImageSize fields: maps None → "auto" and rejects invalid values."""
-    if v is None:
-        return "auto"
-    if v not in get_args(ImageSize):
-        raise ValueError(f"Invalid size: {v!r}. Must be one of {get_args(ImageSize)}")
-    return v  # pyright: ignore[reportReturnType]
-
-
 class AdvancedImageParams(BaseModel):
    seed: Annotated[int, Field(ge=0)] | None = None
    num_inference_steps: Annotated[int, Field(ge=1, le=100)] | None = None
@@ -302,7 +289,7 @@ class ImageGenerationTaskParams(BaseModel):
    partial_images: int | None = 0
    quality: Literal["high", "medium", "low"] | None = "medium"
    response_format: Literal["url", "b64_json"] | None = "b64_json"
-    size: ImageSize = "auto"
+    size: str | None = "1024x1024"
    stream: bool | None = False
    style: str | None = "vivid"
    user: str | None = None
@@ -310,11 +297,6 @@ class ImageGenerationTaskParams(BaseModel):
    # Internal flag for benchmark mode - set by API, preserved through serialization
    bench: bool = False

-    @field_validator("size", mode="before")
-    @classmethod
-    def normalize_size(cls, v: object) -> ImageSize:
-        return normalize_image_size(v)
-

 class BenchImageGenerationTaskParams(ImageGenerationTaskParams):
    bench: bool = True
@@ -331,18 +313,13 @@ class ImageEditsTaskParams(BaseModel):
    quality: Literal["high", "medium", "low"] | None = "medium"
    output_format: Literal["png", "jpeg", "webp"] = "png"
    response_format: Literal["url", "b64_json"] | None = "b64_json"
-    size: ImageSize = "auto"
+    size: str | None = "1024x1024"
    image_strength: float | None = 0.7
    stream: bool = False
    partial_images: int | None = 0
    advanced_params: AdvancedImageParams | None = None
    bench: bool = False

-    @field_validator("size", mode="before")
-    @classmethod
-    def normalize_size(cls, v: object) -> ImageSize:
-        return normalize_image_size(v)
-
    def __repr_args__(self) -> Generator[tuple[str, Any], None, None]:
        for name, value in super().__repr_args__():  # pyright: ignore[reportAny]
            if name == "image_data":
--- a/src/exo/shared/types/commands.py
+++ b/src/exo/shared/types/commands.py
@@ -38,6 +38,8 @@ class PlaceInstance(BaseCommand):
    sharding: Sharding
    instance_meta: InstanceMeta
    min_nodes: int
+    draft_model: ModelId | None = None
+    num_draft_tokens: int = 4


 class CreateInstance(BaseCommand):
@@ -48,10 +50,6 @@ class DeleteInstance(BaseCommand):
    instance_id: InstanceId


-class TaskCancelled(BaseCommand):
-    cancelled_command_id: CommandId
-
-
 class TaskFinished(BaseCommand):
    finished_command_id: CommandId

@@ -76,6 +74,14 @@ class DeleteDownload(BaseCommand):
    model_id: ModelId


+class SetInstanceDraftModel(BaseCommand):
+    """Set or update the draft model for an existing instance."""
+
+    instance_id: InstanceId
+    draft_model: ModelId | None  # None to disable speculative decoding
+    num_draft_tokens: int = 4
+
+
 class CancelDownload(BaseCommand):
    target_node_id: NodeId
    model_id: ModelId
@@ -93,7 +99,7 @@ Command = (
    | PlaceInstance
    | CreateInstance
    | DeleteInstance
-    | TaskCancelled
+    | SetInstanceDraftModel
    | TaskFinished
    | SendInputChunk
 )
--- a/src/exo/shared/types/events.py
+++ b/src/exo/shared/types/events.py
@@ -5,7 +5,7 @@ from pydantic import Field

 from exo.shared.topology import Connection
 from exo.shared.types.chunks import GenerationChunk, InputImageChunk
-from exo.shared.types.common import CommandId, Id, NodeId, SessionId
+from exo.shared.types.common import CommandId, Id, ModelId, NodeId, SessionId
 from exo.shared.types.tasks import Task, TaskId, TaskStatus
 from exo.shared.types.worker.downloads import DownloadProgress
 from exo.shared.types.worker.instances import Instance, InstanceId
@@ -68,6 +68,14 @@ class InstanceDeleted(BaseEvent):
    instance_id: InstanceId


+class InstanceDraftModelUpdated(BaseEvent):
+    """Draft model updated on an existing instance."""
+
+    instance_id: InstanceId
+    draft_model: ModelId | None
+    num_draft_tokens: int
+
+
 class RunnerStatusUpdated(BaseEvent):
    runner_id: RunnerId
    runner_status: RunnerStatus
@@ -141,6 +149,7 @@ Event = (
    | TaskAcknowledged
    | InstanceCreated
    | InstanceDeleted
+    | InstanceDraftModelUpdated
    | RunnerStatusUpdated
    | RunnerDeleted
    | NodeTimedOut
--- a/src/exo/shared/types/tasks.py
+++ b/src/exo/shared/types/tasks.py
@@ -24,7 +24,6 @@ class TaskStatus(str, Enum):
    Complete = "Complete"
    TimedOut = "TimedOut"
    Failed = "Failed"
-    Cancelled = "Cancelled"


 class BaseTask(TaggedModel):
@@ -41,6 +40,12 @@ class DownloadModel(BaseTask):  # emitted by Worker
    shard_metadata: ShardMetadata


+class DownloadDraftModel(BaseTask):  # emitted by Worker
+    """Download a draft model for speculative decoding (rank 0 only)."""
+
+    model_id: str  # HuggingFace model ID
+
+
 class LoadModel(BaseTask):  # emitted by Worker
    pass

@@ -61,11 +66,6 @@ class TextGeneration(BaseTask):  # emitted by Master
    error_message: str | None = Field(default=None)


-class CancelTask(BaseTask):
-    cancelled_task_id: TaskId
-    runner_id: RunnerId
-
-
 class ImageGeneration(BaseTask):  # emitted by Master
    command_id: CommandId
    task_params: ImageGenerationTaskParams
@@ -86,15 +86,23 @@ class Shutdown(BaseTask):  # emitted by Worker
    runner_id: RunnerId


+class SetDraftModel(BaseTask):  # emitted by Worker
+    """Load or clear a draft model on an already-running instance."""
+
+    model_id: str | None  # HuggingFace model ID, or None to clear
+    num_draft_tokens: int = 4
+
+
 Task = (
    CreateRunner
    | DownloadModel
+    | DownloadDraftModel
    | ConnectToGroup
    | LoadModel
    | StartWarmup
    | TextGeneration
-    | CancelTask
    | ImageGeneration
    | ImageEdits
    | Shutdown
+    | SetDraftModel
 )
--- a/src/exo/shared/types/worker/downloads.py
+++ b/src/exo/shared/types/worker/downloads.py
@@ -26,7 +26,6 @@ class DownloadProgressData(CamelCaseModel):
 class BaseDownloadProgress(TaggedModel):
    node_id: NodeId
    shard_metadata: ShardMetadata
-    model_directory: str = ""


 class DownloadPending(BaseDownloadProgress):
--- a/src/exo/shared/types/worker/instances.py
+++ b/src/exo/shared/types/worker/instances.py
@@ -2,7 +2,7 @@ from enum import Enum

 from pydantic import model_validator

-from exo.shared.types.common import Host, Id, NodeId
+from exo.shared.types.common import Host, Id, ModelId, NodeId
 from exo.shared.types.worker.runners import RunnerId, ShardAssignments, ShardMetadata
 from exo.utils.pydantic_ext import CamelCaseModel, TaggedModel

@@ -19,6 +19,8 @@ class InstanceMeta(str, Enum):
 class BaseInstance(TaggedModel):
    instance_id: InstanceId
    shard_assignments: ShardAssignments
+    draft_model: ModelId | None = None  # For speculative decoding (rank 0 only)
+    num_draft_tokens: int = 4  # Tokens to draft per iteration (when draft_model is set)

    def shard(self, runner_id: RunnerId) -> ShardMetadata | None:
        return self.shard_assignments.runner_to_shard.get(runner_id, None)
--- a/src/exo/shared/types/worker/runner_response.py
+++ b/src/exo/shared/types/worker/runner_response.py
@@ -62,7 +62,6 @@ class PartialImageResponse(BaseRunnerResponse):
 class ToolCallResponse(BaseRunnerResponse):
    tool_calls: list[ToolCallItem]
    usage: Usage | None
-    stats: GenerationStats | None = None


 class FinishedResponse(BaseRunnerResponse):
--- a/src/exo/utils/banner.py
+++ b/src/exo/utils/banner.py
@@ -1,7 +1,5 @@
-import sys
-
-
 def print_startup_banner(port: int) -> None:
+    """Print a prominent startup banner with API endpoint information."""
    dashboard_url = f"http://localhost:{port}"
    banner = f"""
 ╔═══════════════════════════════════════════════════════════════════════╗
@@ -29,4 +27,4 @@ def print_startup_banner(port: int) -> None:

 """

-    print(banner, file=sys.stderr)
+    print(banner)
--- a/src/exo/utils/channels.py
+++ b/src/exo/utils/channels.py
@@ -125,9 +125,7 @@ class MpSender[T]:
            self._state.buffer.put(item, block=True)

    async def send_async(self, item: T) -> None:
-        await to_thread.run_sync(
-            self.send, item, limiter=CapacityLimiter(1), abandon_on_cancel=True
-        )
+        await to_thread.run_sync(self.send, item, limiter=CapacityLimiter(1))

    def close(self) -> None:
        if not self._state.closed.is_set():
--- a/src/exo/worker/engines/image/generate.py
+++ b/src/exo/worker/engines/image/generate.py
@@ -14,7 +14,6 @@ from exo.shared.types.api import (
    ImageEditsTaskParams,
    ImageGenerationStats,
    ImageGenerationTaskParams,
-    ImageSize,
 )
 from exo.shared.types.memory import Memory
 from exo.shared.types.worker.runner_response import (
@@ -24,9 +23,9 @@ from exo.shared.types.worker.runner_response import (
 from exo.worker.engines.image.distributed_model import DistributedImageModel


-def parse_size(size_str: ImageSize) -> tuple[int, int]:
+def parse_size(size_str: str | None) -> tuple[int, int]:
    """Parse size parameter like '1024x1024' to (width, height) tuple."""
-    if size_str == "auto":
+    if not size_str:
        return (1024, 1024)

    try:
@@ -110,9 +109,6 @@ def generate_image(
            # Decode base64 image data and save to temp file
            image_path = Path(tmpdir) / "input.png"
            image_path.write_bytes(base64.b64decode(task.image_data))
-            if task.size == "auto":
-                with Image.open(image_path) as img:
-                    width, height = img.size

        for image_num in range(num_images):
            # Increment seed for each image to ensure unique results
--- a/src/exo/worker/engines/mlx/auto_parallel.py
+++ b/src/exo/worker/engines/mlx/auto_parallel.py
@@ -163,14 +163,11 @@ class PipelineLastLayer(CustomMlxLayer):
                output, (self.r + 1) % self.s, group=self.group
            )
            if cache is not None:
-                # CacheList (used by MLA models like DeepSeekV32, GLM MoE DSA)
-                # doesn't have .keys directly; access via first sub-cache.
-                _cache = cache[0] if hasattr(cache, "caches") else cache  # type: ignore
-                _cache.keys = mx.depends(_cache.keys, output)  # type: ignore
+                cache.keys = mx.depends(cache.keys, output)  # type: ignore[reportUnknownMemberType]
            if self.is_prefill:
                mx.eval(output)
                if cache is not None:
-                    mx.eval(_cache.keys)  # type: ignore
+                    mx.eval(cache.keys)  # type: ignore

        if not self.is_prefill:
            output = mx.distributed.all_gather(output, group=self.group)[
@@ -310,9 +307,7 @@ def patch_pipeline_model[T](model: T, group: mx.distributed.Group) -> T:

        # Add dependency to last cache entry to ensure distributed ops are evaluated
        if cache is not None:
-            last = cache[-1]  # type: ignore
-            dep_cache = last[0] if hasattr(last, "caches") else last  # type: ignore
-            dep_cache.keys = mx.depends(dep_cache.keys, logits)  # type: ignore
+            cache[-1].state = mx.depends(cache[-1].state, logits)  # type: ignore

        return logits

@@ -338,9 +333,7 @@ def patch_tensor_model[T](model: T) -> T:

        # Add dependency to last cache entry to ensure distributed ops are evaluated
        if cache is not None and len(cache) > 0:  # pyright: ignore[reportAny]
-            last = cache[-1]  # pyright: ignore[reportAny]
-            dep_cache = last[0] if hasattr(last, "caches") else last  # pyright: ignore[reportAny]
-            dep_cache.keys = mx.depends(dep_cache.keys, logits)  # pyright: ignore[reportAny,reportUnknownMemberType]
+            cache[-1].state = mx.depends(cache[-1].state, logits)  # pyright: ignore[reportAny,reportUnknownMemberType]

        return logits

@@ -554,12 +547,10 @@ class DeepSeekShardingStrategy(TensorParallelShardingStrategy):
        on_timeout: TimeoutCallback | None,
    ) -> nn.Module:
        model = cast(DeepseekV3Model, model)
-
        for layer in model.layers:
            eval_with_timeout(
                layer.parameters(), timeout_seconds / len(model.layers), on_timeout
            )
-
            # Shard the self attention
            if layer.self_attn.q_lora_rank is None:
                layer.self_attn.q_proj = self.all_to_sharded_linear(
@@ -590,18 +581,12 @@ class DeepSeekShardingStrategy(TensorParallelShardingStrategy):
                layer.mlp.down_proj = self.sharded_to_all_linear(layer.mlp.down_proj)
                layer.mlp.up_proj = self.all_to_sharded_linear(layer.mlp.up_proj)

-            # Shard the MoE.
+            # Shard the MoE. Shard in place since the MoE should be responsible
+            # for aggregating the results.
            else:
-                if getattr(layer.mlp, "shared_experts", None) is not None:
-                    self.all_to_sharded_linear_in_place(
-                        layer.mlp.shared_experts.gate_proj
-                    )
-                    self.sharded_to_all_linear_in_place(
-                        layer.mlp.shared_experts.down_proj
-                    )
-                    self.all_to_sharded_linear_in_place(
-                        layer.mlp.shared_experts.up_proj
-                    )
+                self.all_to_sharded_linear_in_place(layer.mlp.shared_experts.gate_proj)
+                self.sharded_to_all_linear_in_place(layer.mlp.shared_experts.down_proj)
+                self.all_to_sharded_linear_in_place(layer.mlp.shared_experts.up_proj)
                self.all_to_sharded_linear_in_place(layer.mlp.switch_mlp.gate_proj)
                self.sharded_to_all_linear_in_place(layer.mlp.switch_mlp.down_proj)
                self.all_to_sharded_linear_in_place(layer.mlp.switch_mlp.up_proj)
@@ -794,7 +779,8 @@ class MiniMaxShardingStrategy(TensorParallelShardingStrategy):

            layer.self_attn = WrappedMiniMaxAttention(layer.self_attn, self.group)  # pyright: ignore[reportAttributeAccessIssue,reportArgumentType]

-            # Shard the MoE.
+            # Shard the MoE. Shard in place since the MoE should be responsible
+            # for aggregating the results.
            self.all_to_sharded_linear_in_place(
                layer.block_sparse_moe.switch_mlp.gate_proj
            )
@@ -907,7 +893,8 @@ class QwenShardingStrategy(TensorParallelShardingStrategy):
                    layer.self_attn.num_attention_heads //= self.N
                    layer.self_attn.num_key_value_heads //= self.N

-            # Shard the MoE.
+            # Shard the MoE. Shard in place since the MoE should be responsible
+            # for aggregating the results.
            if isinstance(layer.mlp, (Qwen3MoeSparseMoeBlock, Qwen3NextSparseMoeBlock)):
                self.all_to_sharded_linear_in_place(layer.mlp.switch_mlp.gate_proj)
                self.sharded_to_all_linear_in_place(layer.mlp.switch_mlp.down_proj)
--- a/src/exo/worker/engines/mlx/generator/generate.py
+++ b/src/exo/worker/engines/mlx/generator/generate.py
@@ -57,7 +57,6 @@ def prefill(
    sampler: Callable[[mx.array], mx.array],
    prompt_tokens: mx.array,
    cache: KVCacheType,
-    group: mx.distributed.Group | None = None,
 ) -> tuple[float, int, list[CacheSnapshot]]:
    """Prefill the KV cache with prompt tokens.

@@ -87,9 +86,6 @@ def prefill(

    set_pipeline_prefill(model, is_prefill=True)

-    mx_barrier(group)
-    logger.info("Starting prefill")
-
    # Use max_tokens=1 because max_tokens=0 does not work.
    # We just throw away the generated token - we only care about filling the cache
    for _ in stream_generate(
@@ -309,9 +305,16 @@ def mlx_generate(
    )
    max_stop_len = max((len(s) for s in stop_sequences), default=0)

+    mx_barrier(group)
+    logger.info("Ready to prefill")
+
    # Prefill cache with all tokens except the last one
    prefill_tps, prefill_tokens, ssm_snapshots_list = prefill(
-        model, tokenizer, sampler, prompt_tokens[:-1], caches, group
+        model,
+        tokenizer,
+        sampler,
+        prompt_tokens[:-1],
+        caches,
    )
    cache_snapshots: list[CacheSnapshot] | None = ssm_snapshots_list or None

@@ -328,7 +331,6 @@ def mlx_generate(
    think_start = tokenizer.think_start
    think_end = tokenizer.think_end

-    logger.info("Starting decode")
    mx_barrier(group)

    for completion_tokens, out in enumerate(
@@ -391,11 +393,10 @@ def mlx_generate(
                    f"Model generated unexpected finish_reason: {out.finish_reason}"
                )

-            total_prompt_tokens = len(all_prompt_tokens)
            usage = Usage(
-                prompt_tokens=total_prompt_tokens,
+                prompt_tokens=int(out.prompt_tokens),
                completion_tokens=completion_tokens,
-                total_tokens=total_prompt_tokens + completion_tokens,
+                total_tokens=int(out.prompt_tokens) + completion_tokens,
                prompt_tokens_details=PromptTokensDetails(
                    cached_tokens=prefix_hit_length
                ),
--- a/src/exo/worker/engines/mlx/utils_mlx.py
+++ b/src/exo/worker/engines/mlx/utils_mlx.py
@@ -64,6 +64,8 @@ from exo.worker.runner.bootstrap import logger
 Group = mx.distributed.Group


+# TODO: Test this
+#  ALSO https://github.com/exo-explore/exo/pull/233#discussion_r2549683673
 def get_weights_size(model_shard_meta: ShardMetadata) -> Memory:
    return Memory.from_float_kb(
        (model_shard_meta.end_layer - model_shard_meta.start_layer)
@@ -81,6 +83,30 @@ class ModelLoadingTimeoutError(Exception):
    pass


+def mx_barrier(group: Group | None = None):
+    mx.eval(
+        mx.distributed.all_sum(
+            mx.array(1.0),
+            stream=mx.default_stream(mx.Device(mx.cpu)),
+            group=group,
+        )
+    )
+
+
+def broadcast_from_zero(value: int, group: Group | None = None):
+    if group is None:
+        return value
+
+    if group.rank() == 0:
+        a = mx.array([value], dtype=mx.int32)
+    else:
+        a = mx.array([0], dtype=mx.int32)
+
+    m = mx.distributed.all_sum(a, stream=mx.Device(mx.DeviceType.cpu), group=group)
+    mx.eval(m)
+    return int(m.item())
+
+
 class HostList(RootModel[list[str]]):
    @classmethod
    def from_hosts(cls, hosts: list[Host]) -> "HostList":
@@ -197,6 +223,27 @@ def load_mlx_items(
    return cast(Model, model), tokenizer


+def load_draft_model(model_id: ModelId) -> nn.Module:
+    """Load a draft model for speculative decoding (rank 0 only).
+
+    Draft models are small models (typically 0.5B-2B parameters) used to
+    generate candidate tokens quickly, which are then verified by the main
+    model in a single forward pass.
+
+    Assumes the model has already been downloaded by the worker.
+
+    Args:
+        model_id: HuggingFace model ID for the draft model
+
+    Returns:
+        The loaded draft model
+    """
+    model_path = build_model_path(model_id)
+    draft_model, _ = load_model(model_path, strict=True)
+    logger.info(f"Loaded draft model from {model_path}")
+    return draft_model
+
+
 def shard_and_load(
    shard_metadata: ShardMetadata,
    group: Group,
@@ -285,12 +332,10 @@ def get_eos_token_ids_for_model(model_id: ModelId) -> list[int] | None:
    model_id_lower = model_id.lower()
    if "kimi-k2" in model_id_lower:
        return [163586]
-    elif "glm-5" in model_id_lower or "glm-4.7" in model_id_lower:
-        # For GLM-5 and GLM-4.7
+    elif "glm-4.7-flash" in model_id_lower:
        # 154820: <|endoftext|>, 154827: <|user|>, 154829: <|observation|>
        return [154820, 154827, 154829]
    elif "glm" in model_id_lower:
-        # For GLM-4.5 and older
        return [151336, 151329, 151338]
    return None

@@ -355,13 +400,7 @@ def load_tokenizer_for_model_id(
            return list(hf_tokenizer.model.encode(text, allowed_special="all"))  # pyright: ignore[reportUnknownMemberType,reportUnknownArgumentType]

        hf_tokenizer.encode = _patched_encode
-        return TokenizerWrapper(
-            hf_tokenizer,
-            eos_token_ids=eos_token_ids,
-            tool_call_start="<|tool_calls_section_begin|>",
-            tool_call_end="<|tool_calls_section_end|>",
-            tool_parser=_parse_kimi_tool_calls,
-        )
+        return TokenizerWrapper(hf_tokenizer, eos_token_ids=eos_token_ids)

    tokenizer = load_tokenizer(
        model_path,
@@ -573,61 +612,3 @@ def mlx_cleanup(
    import gc

    gc.collect()
-
-
-def mx_any(bool_: bool, group: Group | None) -> bool:
-    if group is None:
-        return bool_
-    num_true = mx.distributed.all_sum(
-        mx.array(bool_), group=group, stream=mx.default_stream(mx.Device(mx.cpu))
-    )
-    mx.eval(num_true)
-    return num_true.item() > 0
-
-
-def mx_barrier(group: Group | None):
-    if group is None:
-        return
-    mx.eval(
-        mx.distributed.all_sum(
-            mx.array(1.0), group=group, stream=mx.default_stream(mx.Device(mx.cpu))
-        )
-    )
-
-
-def _parse_kimi_tool_calls(text: str):
-    import regex as re
-
-    # kimi has a fixed function naming scheme, with a json formatted arg
-    #   functions.multiply:0<|tool_call_argument_begin|>{"a": 2, "b": 3}
-    _func_name_regex = re.compile(
-        r"^\s*((?:functions\.)?(.+?):\d+)\s*<\|tool_call_argument_begin\|>", re.DOTALL
-    )
-    _func_arg_regex = re.compile(r"<\|tool_call_argument_begin\|>\s*(.*)\s*", re.DOTALL)
-    _tool_call_split_regex = re.compile(
-        r"<\|tool_call_begin\|>(.*?)<\|tool_call_end\|>", re.DOTALL
-    )
-
-    def _parse_single_tool(text: str) -> dict[str, Any]:
-        func_name_match = _func_name_regex.search(text)
-        if func_name_match is None:
-            raise ValueError("No tool call found.")
-        tool_call_id = func_name_match.group(1)  # e.g. "functions.get_weather:0"
-        func_name = func_name_match.group(2)  # e.g. "get_weather"
-
-        func_args_match = _func_arg_regex.search(text)
-        if func_args_match is None:
-            raise ValueError("No tool call arguments found.")
-        func_args = func_args_match.group(1)
-        try:
-            arg_dct = json.loads(func_args)  # pyright: ignore[reportAny]
-        except Exception:
-            arg_dct = None
-
-        return dict(id=tool_call_id, name=func_name, arguments=arg_dct)
-
-    tool_matches = _tool_call_split_regex.findall(text)
-    if tool_matches:
-        return [_parse_single_tool(match) for match in tool_matches]  # pyright: ignore[reportAny]
-    else:
-        return [_parse_single_tool(text)]
--- a/src/exo/worker/main.py
+++ b/src/exo/worker/main.py
@@ -33,7 +33,6 @@ from exo.shared.types.events import (
 from exo.shared.types.multiaddr import Multiaddr
 from exo.shared.types.state import State
 from exo.shared.types.tasks import (
-    CancelTask,
    CreateRunner,
    DownloadModel,
    ImageEdits,
@@ -225,22 +224,15 @@ class Worker:
                        )
                    )
                case Shutdown(runner_id=runner_id):
-                    runner = self.runners.pop(runner_id)
                    try:
                        with fail_after(3):
-                            await runner.start_task(task)
+                            await self.runners.pop(runner_id).start_task(task)
                    except TimeoutError:
                        await self.event_sender.send(
                            TaskStatusUpdated(
                                task_id=task.task_id, task_status=TaskStatus.TimedOut
                            )
                        )
-                    finally:
-                        runner.shutdown()
-                case CancelTask(
-                    cancelled_task_id=cancelled_task_id, runner_id=runner_id
-                ):
-                    await self.runners[runner_id].cancel_task(cancelled_task_id)
                case ImageEdits() if task.task_params.total_input_chunks > 0:
                    # Assemble image from chunks and inject into task
                    cmd_id = task.command_id
@@ -278,18 +270,18 @@ class Worker:
                        del self.input_chunk_buffer[cmd_id]
                    if cmd_id in self.input_chunk_counts:
                        del self.input_chunk_counts[cmd_id]
-                    await self._start_runner_task(modified_task)
+                    await self.runners[self._task_to_runner_id(task)].start_task(
+                        modified_task
+                    )
                case task:
-                    await self._start_runner_task(task)
+                    await self.runners[self._task_to_runner_id(task)].start_task(task)

    def shutdown(self):
        self._tg.cancel_scope.cancel()

-    async def _start_runner_task(self, task: Task):
-        if (instance := self.state.instances.get(task.instance_id)) is not None:
-            await self.runners[
-                instance.shard_assignments.node_to_runner[self.node_id]
-            ].start_task(task)
+    def _task_to_runner_id(self, task: Task):
+        instance = self.state.instances[task.instance_id]
+        return instance.shard_assignments.node_to_runner[self.node_id]

    async def _nack_request(self, since_idx: int) -> None:
        # We request all events after (and including) the missing index.
@@ -328,6 +320,8 @@ class Worker:
            for event in self.out_for_delivery.copy().values():
                await self.local_event_sender.send(event)

+    ## Op Executors
+
    def _create_supervisor(self, task: CreateRunner) -> RunnerSupervisor:
        """Creates and stores a new AssignedRunner with initial downloading status."""
        runner = RunnerSupervisor.create(
--- a/src/exo/worker/plan.py
+++ b/src/exo/worker/plan.py
@@ -4,7 +4,6 @@ from collections.abc import Mapping, Sequence

 from exo.shared.types.common import CommandId, NodeId
 from exo.shared.types.tasks import (
-    CancelTask,
    ConnectToGroup,
    CreateRunner,
    DownloadModel,
@@ -54,14 +53,13 @@ def plan(
 ) -> Task | None:
    # Python short circuiting OR logic should evaluate these sequentially.
    return (
-        _cancel_tasks(runners, tasks)
-        or _kill_runner(runners, all_runners, instances)
+        _kill_runner(runners, all_runners, instances)
        or _create_runner(node_id, runners, instances)
        or _model_needs_download(node_id, runners, global_download_status)
        or _init_distributed_backend(runners, all_runners)
        or _load_model(runners, all_runners, global_download_status)
        or _ready_to_warmup(runners, all_runners)
-        or _pending_tasks(runners, tasks, all_runners, input_chunk_buffer or {})
+        or _pending_tasks(runners, tasks, all_runners, input_chunk_buffer)
    )


@@ -272,7 +270,7 @@ def _pending_tasks(
    runners: Mapping[RunnerId, RunnerSupervisor],
    tasks: Mapping[TaskId, Task],
    all_runners: Mapping[RunnerId, RunnerStatus],
-    input_chunk_buffer: Mapping[CommandId, dict[int, str]],
+    input_chunk_buffer: Mapping[CommandId, dict[int, str]] | None = None,
 ) -> Task | None:
    for task in tasks.values():
        # for now, just forward chat completions
@@ -286,7 +284,7 @@ def _pending_tasks(
        if isinstance(task, ImageEdits) and task.task_params.total_input_chunks > 0:
            cmd_id = task.command_id
            expected = task.task_params.total_input_chunks
-            received = len(input_chunk_buffer.get(cmd_id, {}))
+            received = len((input_chunk_buffer or {}).get(cmd_id, {}))
            if received < expected:
                continue  # Wait for all chunks to arrive

@@ -294,33 +292,16 @@ def _pending_tasks(
            if task.instance_id != runner.bound_instance.instance.instance_id:
                continue

-            # the task status _should_ be set to completed by the LAST runner
-            # it is currently set by the first
-            # this is definitely a hack
+            # I have a design point here; this is a state race in disguise as the task status doesn't get updated to completed fast enough
+            # however, realistically the task status should be set to completed by the LAST runner, so this is a true race
+            # the actual solution is somewhat deeper than this bypass - TODO!
            if task.task_id in runner.completed:
                continue

+            # TODO: Check ordering aligns with MLX distributeds expectations.
+
            if isinstance(runner.status, RunnerReady) and all(
                isinstance(all_runners[global_runner_id], (RunnerReady, RunnerRunning))
                for global_runner_id in runner.bound_instance.instance.shard_assignments.runner_to_shard
            ):
                return task
-
-
-def _cancel_tasks(
-    runners: Mapping[RunnerId, RunnerSupervisor],
-    tasks: Mapping[TaskId, Task],
-) -> Task | None:
-    for task in tasks.values():
-        if task.task_status != TaskStatus.Cancelled:
-            continue
-        for runner_id, runner in runners.items():
-            if task.instance_id != runner.bound_instance.instance.instance_id:
-                continue
-            if task.task_id in runner.cancelled:
-                continue
-            return CancelTask(
-                instance_id=task.instance_id,
-                cancelled_task_id=task.task_id,
-                runner_id=runner_id,
-            )
--- a/src/exo/worker/runner/bootstrap.py
+++ b/src/exo/worker/runner/bootstrap.py
@@ -3,7 +3,7 @@ import os
 import loguru

 from exo.shared.types.events import Event, RunnerStatusUpdated
-from exo.shared.types.tasks import Task, TaskId
+from exo.shared.types.tasks import Task
 from exo.shared.types.worker.instances import BoundInstance, MlxJacclInstance
 from exo.shared.types.worker.runners import RunnerFailed
 from exo.utils.channels import ClosedResourceError, MpReceiver, MpSender
@@ -15,7 +15,6 @@ def entrypoint(
    bound_instance: BoundInstance,
    event_sender: MpSender[Event],
    task_receiver: MpReceiver[Task],
-    cancel_receiver: MpReceiver[TaskId],
    _logger: "loguru.Logger",
 ) -> None:
    fast_synch_override = os.environ.get("EXO_FAST_SYNCH")
@@ -39,7 +38,7 @@ def entrypoint(
    try:
        from exo.worker.runner.runner import main

-        main(bound_instance, event_sender, task_receiver, cancel_receiver)
+        main(bound_instance, event_sender, task_receiver)
    except ClosedResourceError:
        logger.warning("Runner communication closed unexpectedly")
    except Exception as e:
--- a/src/exo/worker/runner/runner.py
+++ b/src/exo/worker/runner/runner.py
@@ -1,10 +1,10 @@
 import base64
-import math
+import json
 import resource
 import time
 from collections.abc import Generator
 from functools import cache
-from typing import Literal
+from typing import Any, Callable, Literal

 import mlx.core as mx
 from mlx_lm.models.gpt_oss import Model as GptOssModel
@@ -15,6 +15,7 @@ from openai_harmony import (  # pyright: ignore[reportMissingTypeStubs]
    StreamableParser,
    load_harmony_encoding,
 )
+from pydantic import ValidationError

 from exo.shared.constants import EXO_MAX_CHUNK_SIZE, EXO_TRACING_ENABLED
 from exo.shared.models.model_cards import ModelId, ModelTask
@@ -87,12 +88,9 @@ from exo.worker.engines.mlx.utils_mlx import (
    initialize_mlx,
    load_mlx_items,
    mlx_force_oom,
-    mx_any,
 )
 from exo.worker.runner.bootstrap import logger

-from .tool_parsers import ToolParser, make_mlx_parser
-

 def _is_primary_output_node(shard_metadata: ShardMetadata) -> bool:
    """Check if this node is the primary output node for image generation.
@@ -114,7 +112,6 @@ def main(
    bound_instance: BoundInstance,
    event_sender: MpSender[Event],
    task_receiver: MpReceiver[Task],
-    cancel_receiver: MpReceiver[TaskId],
 ):
    soft, hard = resource.getrlimit(resource.RLIMIT_NOFILE)
    resource.setrlimit(resource.RLIMIT_NOFILE, (min(max(soft, 2048), hard), hard))
@@ -132,16 +129,11 @@ def main(
        time.sleep(timeout)

    setup_start_time = time.time()
-    cancelled_tasks = set[TaskId]()

-    # type checker was unhappy with me - splitting these fixed it
-    inference_model: Model | None = None
-    image_model: DistributedImageModel | None = None
+    model: Model | DistributedImageModel | None = None
    tokenizer = None
-    tool_parser: ToolParser | None = None
    group = None
    kv_prefix_cache: KVPrefixCache | None = None
-    check_for_cancel_every: int | None = None

    current_status: RunnerStatus = RunnerIdle()
    logger.info("runner created")
@@ -154,7 +146,6 @@ def main(
            if task.task_id in seen:
                logger.warning("repeat task - potential error")
            seen.add(task.task_id)
-            cancelled_tasks.discard(TaskId("CANCEL_CURRENT_TASK"))
            event_sender.send(
                TaskStatusUpdated(task_id=task.task_id, task_status=TaskStatus.Running)
            )
@@ -200,28 +191,19 @@ def main(
                        time.sleep(0.5)

                    if ModelTask.TextGeneration in shard_metadata.model_card.tasks:
-                        inference_model, tokenizer = load_mlx_items(
+                        model, tokenizer = load_mlx_items(
                            bound_instance, group, on_timeout=on_model_load_timeout
                        )
                        logger.info(
-                            f"model has_tool_calling={tokenizer.has_tool_calling} using tokens {tokenizer.tool_call_start}, {tokenizer.tool_call_end}"
+                            f"model has_tool_calling={tokenizer.has_tool_calling}"
                        )
-                        if tokenizer.has_tool_calling:
-                            assert tokenizer.tool_call_start
-                            assert tokenizer.tool_call_end
-                            assert tokenizer.tool_parser  # pyright: ignore[reportAny]
-                            tool_parser = make_mlx_parser(
-                                tokenizer.tool_call_start,
-                                tokenizer.tool_call_end,
-                                tokenizer.tool_parser,  # pyright: ignore[reportAny]
-                            )
                        kv_prefix_cache = KVPrefixCache(group)

                    elif (
                        ModelTask.TextToImage in shard_metadata.model_card.tasks
                        or ModelTask.ImageToImage in shard_metadata.model_card.tasks
                    ):
-                        image_model = initialize_image_model(bound_instance)
+                        model = initialize_image_model(bound_instance)
                    else:
                        raise ValueError(
                            f"Unknown model task(s): {shard_metadata.model_card.tasks}"
@@ -229,6 +211,8 @@ def main(
                    current_status = RunnerLoaded()
                    logger.info("runner loaded")
                case StartWarmup() if isinstance(current_status, RunnerLoaded):
+                    assert model
+
                    current_status = RunnerWarmingUp()
                    logger.info("runner warming up")
                    event_sender.send(
@@ -240,31 +224,16 @@ def main(

                    logger.info(f"warming up inference for instance: {instance}")
                    if ModelTask.TextGeneration in shard_metadata.model_card.tasks:
-                        assert inference_model
+                        assert not isinstance(model, DistributedImageModel)
                        assert tokenizer

-                        t = time.monotonic()
                        toks = warmup_inference(
-                            model=inference_model,
+                            model=model,
                            tokenizer=tokenizer,
                            group=group,
+                            # kv_prefix_cache=kv_prefix_cache,  # supply for warmup-time prefix caching
                        )
                        logger.info(f"warmed up by generating {toks} tokens")
-                        check_for_cancel_every = min(
-                            math.ceil(toks / min(time.monotonic() - t, 0.001)), 100
-                        )
-                        if group is not None:
-                            check_for_cancel_every = int(
-                                mx.max(
-                                    mx.distributed.all_gather(
-                                        mx.array([check_for_cancel_every]), group=group
-                                    )
-                                ).item()
-                            )
-
-                        logger.info(
-                            f"runner checking for cancellation every {check_for_cancel_every} tokens"
-                        )
                        logger.info(
                            f"runner initialized in {time.time() - setup_start_time} seconds"
                        )
@@ -272,8 +241,8 @@ def main(
                        ModelTask.TextToImage in shard_metadata.model_card.tasks
                        or ModelTask.ImageToImage in shard_metadata.model_card.tasks
                    ):
-                        assert image_model
-                        image = warmup_image_generator(model=image_model)
+                        assert isinstance(model, DistributedImageModel)
+                        image = warmup_image_generator(model=model)
                        if image is not None:
                            logger.info(f"warmed up by generating {image.size} image")
                        else:
@@ -293,9 +262,9 @@ def main(
                        )
                    )
                    event_sender.send(TaskAcknowledged(task_id=task.task_id))
-                    assert inference_model
+
+                    assert model and not isinstance(model, DistributedImageModel)
                    assert tokenizer
-                    assert check_for_cancel_every

                    try:
                        _check_for_debug_prompts(task_params)
@@ -305,7 +274,7 @@ def main(

                        # Generate responses using the actual MLX generation
                        mlx_generator = mlx_generate(
-                            model=inference_model,
+                            model=model,
                            tokenizer=tokenizer,
                            task=task_params,
                            prompt=prompt,
@@ -320,25 +289,34 @@ def main(
                                mlx_generator, tokenizer
                            )

+                        # Kimi-K2 has tool call sections - we don't care about them
+                        if "kimi" in shard_metadata.model_card.model_id.lower():
+                            mlx_generator = filter_kimi_tokens(mlx_generator)
+                            patch_kimi_tokenizer(tokenizer)
+
+                        # GLM models need patched parser (upstream has bug with None regex match)
+                        elif "glm" in shard_metadata.model_card.model_id.lower():
+                            patch_glm_tokenizer(tokenizer)
+
                        # GPT-OSS specific parsing to match other model formats.
-                        if isinstance(inference_model, GptOssModel):
+                        elif isinstance(model, GptOssModel):
                            mlx_generator = parse_gpt_oss(mlx_generator)
-                        elif tool_parser:
-                            mlx_generator = parse_tool_calls(mlx_generator, tool_parser)
+
+                        if tokenizer.has_tool_calling and not isinstance(
+                            model, GptOssModel
+                        ):
+                            assert tokenizer.tool_call_start
+                            assert tokenizer.tool_call_end
+                            assert tokenizer.tool_parser  # pyright: ignore[reportAny]
+                            mlx_generator = parse_tool_calls(
+                                mlx_generator,
+                                tokenizer.tool_call_start,
+                                tokenizer.tool_call_end,
+                                tokenizer.tool_parser,  # pyright: ignore[reportAny]
+                            )

                        completion_tokens = 0
-                        tokens_since_last_cancel_check = 0
                        for response in mlx_generator:
-                            tokens_since_last_cancel_check += 1
-                            if tokens_since_last_cancel_check >= check_for_cancel_every:
-                                tokens_since_last_cancel_check = 0
-                                cancelled_tasks.update(cancel_receiver.collect())
-                                want_to_cancel = (task.task_id in cancelled_tasks) or (
-                                    TaskId("CANCEL_CURRENT_TASK") in cancelled_tasks
-                                )
-                                if mx_any(want_to_cancel, group):
-                                    break
-
                            match response:
                                case GenerationResponse():
                                    completion_tokens += 1
@@ -386,7 +364,6 @@ def main(
                                                    tool_calls=response.tool_calls,
                                                    model=shard_metadata.model_card.model_id,
                                                    usage=response.usage,
-                                                    stats=response.stats,
                                                ),
                                            )
                                        )
@@ -411,7 +388,7 @@ def main(
                case ImageGeneration(
                    task_params=task_params, command_id=command_id
                ) if isinstance(current_status, RunnerReady):
-                    assert image_model
+                    assert isinstance(model, DistributedImageModel)
                    logger.info(f"received image generation request: {str(task)[:500]}")
                    current_status = RunnerRunning()
                    logger.info("runner running")
@@ -424,9 +401,7 @@ def main(

                    try:
                        image_index = 0
-                        for response in generate_image(
-                            model=image_model, task=task_params
-                        ):
+                        for response in generate_image(model=model, task=task_params):
                            is_primary_output = _is_primary_output_node(shard_metadata)

                            if is_primary_output:
@@ -476,7 +451,7 @@ def main(
                case ImageEdits(task_params=task_params, command_id=command_id) if (
                    isinstance(current_status, RunnerReady)
                ):
-                    assert image_model
+                    assert isinstance(model, DistributedImageModel)
                    logger.info(f"received image edits request: {str(task)[:500]}")
                    current_status = RunnerRunning()
                    logger.info("runner running")
@@ -489,9 +464,7 @@ def main(

                    try:
                        image_index = 0
-                        for response in generate_image(
-                            model=image_model, task=task_params
-                        ):
+                        for response in generate_image(model=model, task=task_params):
                            if _is_primary_output_node(shard_metadata):
                                match response:
                                    case PartialImageResponse():
@@ -550,20 +523,14 @@ def main(
                    raise ValueError(
                        f"Received {task.__class__.__name__} outside of state machine in {current_status=}"
                    )
-            was_cancelled = (task.task_id in cancelled_tasks) or (
-                TaskId("CANCEL_CURRENT_TASK") in cancelled_tasks
+            event_sender.send(
+                TaskStatusUpdated(task_id=task.task_id, task_status=TaskStatus.Complete)
            )
-            if not was_cancelled:
-                event_sender.send(
-                    TaskStatusUpdated(
-                        task_id=task.task_id, task_status=TaskStatus.Complete
-                    )
-                )
            event_sender.send(
                RunnerStatusUpdated(runner_id=runner_id, runner_status=current_status)
            )
            if isinstance(current_status, RunnerShutdown):
-                del inference_model, image_model, tokenizer, group
+                del model, tokenizer, group
                mx.clear_cache()
                import gc

@@ -577,8 +544,21 @@ def get_gpt_oss_encoding():
    return encoding


+def filter_kimi_tokens(
+    responses: Generator[GenerationResponse | ToolCallResponse],
+) -> Generator[GenerationResponse]:
+    for resp in responses:
+        assert isinstance(resp, GenerationResponse)
+        if (
+            resp.text == "<|tool_calls_section_begin|>"
+            or resp.text == "<|tool_calls_section_end|>"
+        ):
+            continue
+        yield resp
+
+
 def parse_gpt_oss(
-    responses: Generator[GenerationResponse],
+    responses: Generator[GenerationResponse | ToolCallResponse],
 ) -> Generator[GenerationResponse | ToolCallResponse]:
    encoding = get_gpt_oss_encoding()
    stream = StreamableParser(encoding, role=Role.ASSISTANT)
@@ -635,9 +615,9 @@ def parse_gpt_oss(


 def parse_thinking_models(
-    responses: Generator[GenerationResponse],
+    responses: Generator[GenerationResponse | ToolCallResponse],
    tokenizer: TokenizerWrapper,
-) -> Generator[GenerationResponse]:
+) -> Generator[GenerationResponse | ToolCallResponse]:
    """
    For models that inject thinking tags in the prompt (like GLM-4.7),
    prepend the thinking tag to the output stream so the frontend
@@ -758,55 +738,218 @@ def _process_image_response(


 def parse_tool_calls(
-    responses: Generator[GenerationResponse], tool_parser: ToolParser
+    responses: Generator[GenerationResponse | ToolCallResponse],
+    tool_call_start: str,
+    tool_call_end: str,
+    tool_parser: Callable[[str], dict[str, Any] | list[dict[str, Any]]],
 ) -> Generator[GenerationResponse | ToolCallResponse]:
    in_tool_call = False
    tool_call_text_parts: list[str] = []
    for response in responses:
-        if response.text.startswith(tool_parser.start_parsing):
+        assert isinstance(response, GenerationResponse)
+        # assumption: the tool call start is one token
+        if response.text == tool_call_start:
            in_tool_call = True
-
-        if in_tool_call:
-            tool_call_text_parts.append(response.text)
-            if response.text.endswith(tool_parser.end_parsing):
-                # parse the actual tool calls from the tool call text
-                parsed = tool_parser.parse_tool_calls(
-                    "".join(tool_call_text_parts).strip()
-                )
+            continue
+        # assumption: the tool call end is one token
+        if in_tool_call and response.text == tool_call_end:
+            try:
+                # tool_parser returns an arbitrarily nested python dictionary
+                # we actually don't want the python dictionary, we just want to
+                # parse the top level { function: ..., arguments: ... } structure
+                # as we're just gonna hand it back to the api anyway
+                parsed = tool_parser("".join(tool_call_text_parts).strip())
                logger.info(f"parsed {tool_call_text_parts=} into {parsed=}")
-                if parsed is not None:
-                    yield ToolCallResponse(
-                        tool_calls=parsed, usage=response.usage, stats=response.stats
-                    )
+                if isinstance(parsed, list):
+                    tools = [_validate_single_tool(tool) for tool in parsed]
                else:
-                    logger.warning(
-                        f"tool call parsing failed for text {''.join(tool_call_text_parts)}"
-                    )
-                    response.text = "".join(tool_call_text_parts)
-                    yield response
+                    tools = [_validate_single_tool(parsed)]
+                yield ToolCallResponse(tool_calls=tools, usage=response.usage)

-                in_tool_call = False
-                tool_call_text_parts = []
-                continue
-
-            if response.finish_reason is not None:
-                logger.info(
-                    "tool call parsing interrupted, yield partial tool call as text"
-                )
-                response = response.model_copy(
-                    update={
-                        "text": "".join(tool_call_text_parts),
-                        "token": 0,
-                    }
+            except (
+                json.JSONDecodeError,
+                ValidationError,
+                ValueError,
+                AttributeError,
+            ) as e:
+                # ValueError: our parsers raise this for malformed tool calls
+                # AttributeError: upstream parsers (e.g. glm47) may raise this when regex doesn't match
+                logger.opt(exception=e).warning("tool call parsing failed")
+                # assumption: talking about tool calls, not making a tool call
+                response.text = (
+                    tool_call_start + "".join(tool_call_text_parts) + tool_call_end
                )
                yield response

+            in_tool_call = False
+            tool_call_text_parts = []
            continue

+        if in_tool_call:
+            tool_call_text_parts.append(response.text)
+            if response.finish_reason is not None:
+                logger.info(
+                    "toll call parsing interrupted, yield partial tool call as text"
+                )
+                yield GenerationResponse(
+                    text=tool_call_start + "".join(tool_call_text_parts),
+                    token=0,
+                    finish_reason=response.finish_reason,
+                    usage=None,
+                )
+            continue
        # fallthrough
        yield response


+def patch_kimi_tokenizer(tokenizer: TokenizerWrapper):
+    """
+    Version of to-be-upstreamed kimi-k2 tool parser
+    """
+    import ast
+    import json
+    from typing import Any
+
+    import regex as re
+
+    # kimi has a fixed function naming scheme, with a json formatted arg
+    #   functions.multiply:0 <|tool_call_argument_begin|> {"a": 2, "b": 3}
+    #   Also needs to handle tools like call_0<|tool_call_argument_begin|>{"filePath": "..."}
+    _func_name_regex = re.compile(
+        r"^\s*(.+)[:](\d+)\s*<\|tool_call_argument_begin\|>", re.DOTALL
+    )
+    _func_arg_regex = re.compile(r"<\|tool_call_argument_begin\|>\s*(.*)\s*", re.DOTALL)
+
+    # kimi has a tool_calls_section - we're leaving this up to the caller to handle
+    tool_call_start = "<|tool_call_begin|>"
+    tool_call_end = "<|tool_call_end|>"
+
+    def _deserialize(value: str) -> Any:  # pyright: ignore[reportAny]
+        try:
+            return json.loads(value)  # pyright: ignore[reportAny]
+        except Exception:
+            pass
+
+        try:
+            return ast.literal_eval(value)  # pyright: ignore[reportAny]
+        except Exception:
+            pass
+        return value
+
+    def parse_tool_call(text: str, tools: Any | None = None):
+        func_name_match = _func_name_regex.search(text)
+        if func_name_match is None:
+            raise ValueError(f"Could not parse function name from tool call: {text!r}")
+        original_func_name = func_name_match.group(1)
+        tool_id = func_name_match.group(2)
+        # strip off the `functions.` prefix, if it exists.
+        func_name = original_func_name[original_func_name.find(".") + 1 :]
+
+        func_args_match = _func_arg_regex.search(text)
+        if func_args_match is None:
+            raise ValueError(f"Could not parse function args from tool call: {text!r}")
+        func_args = func_args_match.group(1)
+        # the args should be valid json - no need to check against our tools to deserialize
+        arg_dct = _deserialize(func_args)  # pyright: ignore[reportAny]
+
+        return dict(
+            id=f"{original_func_name}:{tool_id}",
+            name=func_name,
+            arguments=arg_dct,  # pyright: ignore[reportAny]
+        )
+
+    tokenizer._tool_call_start = tool_call_start
+    tokenizer._tool_call_end = tool_call_end
+    tokenizer._tool_parser = parse_tool_call
+
+
+def patch_glm_tokenizer(tokenizer: TokenizerWrapper):
+    """
+    Fixed version of mlx_lm's glm47 tool parser that handles regex match failures.
+    """
+    import ast
+    import json
+    from typing import Any
+
+    import regex as re
+
+    _func_name_regex = re.compile(r"^(.*?)<arg_key>", re.DOTALL)
+    _func_arg_regex = re.compile(
+        r"<arg_key>(.*?)</arg_key>(?:\n|\s)*<arg_value>(.*?)(?:</arg_value>|(?=<arg_key>)|$)",
+        re.DOTALL,
+    )
+
+    tool_call_start = "<tool_call>"
+    tool_call_end = "</tool_call>"
+
+    def _is_string_type(
+        tool_name: str,
+        arg_name: str,
+        tools: list[Any] | None,
+    ) -> bool:
+        if tools is None:
+            return False
+        for tool in tools:  # pyright: ignore[reportAny]
+            func = tool["function"]  # pyright: ignore[reportAny]
+            if func["name"] == tool_name:
+                params = func["parameters"]  # pyright: ignore[reportAny]
+                if params is None:
+                    return False
+                props = params.get("properties", {})  # pyright: ignore[reportAny]
+                arg_props = props.get(arg_name, {})  # pyright: ignore[reportAny]
+                arg_type = arg_props.get("type", None)  # pyright: ignore[reportAny]
+                return arg_type == "string"  # pyright: ignore[reportAny]
+        return False
+
+    def _deserialize(value: str) -> Any:  # pyright: ignore[reportAny]
+        try:
+            return json.loads(value)  # pyright: ignore[reportAny]
+        except Exception:
+            pass
+        try:
+            return ast.literal_eval(value)  # pyright: ignore[reportAny]
+        except Exception:
+            pass
+        return value
+
+    def parse_tool_call(text: str, tools: list[Any] | None = None):
+        func_name_match = _func_name_regex.search(text)
+        if func_name_match is None:
+            raise ValueError(f"Could not parse function name from tool call: {text!r}")
+        func_name = func_name_match.group(1)
+
+        pairs = _func_arg_regex.findall(text)
+        arg_dct: dict[str, Any] = {}
+        for key, value in pairs:  # pyright: ignore[reportAny]
+            arg_key = key.strip()  # pyright: ignore[reportAny]
+            arg_val = value.strip()  # pyright: ignore[reportAny]
+            if not _is_string_type(func_name, arg_key, tools):  # pyright: ignore[reportAny]
+                arg_val = _deserialize(arg_val)  # pyright: ignore[reportAny]
+            arg_dct[arg_key] = arg_val
+        return dict(name=func_name, arguments=arg_dct)
+
+    tokenizer._tool_call_start = tool_call_start
+    tokenizer._tool_call_end = tool_call_end
+    tokenizer._tool_parser = parse_tool_call
+
+
+def _validate_single_tool(obj: dict[str, Any]) -> ToolCallItem:
+    if (
+        ((name := obj.get("name")) is not None)
+        and ((args := obj.get("arguments")) is not None)
+        and isinstance(name, str)
+    ):
+        raw_id: object = obj.get("id")
+        extra = {"id": str(raw_id)} if raw_id is not None else {}
+        return ToolCallItem(
+            **extra,
+            name=name,
+            arguments=json.dumps(args),
+        )
+    else:
+        raise ValidationError
+
+
 EXO_RUNNER_MUST_FAIL = "EXO RUNNER MUST FAIL"
 EXO_RUNNER_MUST_OOM = "EXO RUNNER MUST OOM"
 EXO_RUNNER_MUST_TIMEOUT = "EXO RUNNER MUST TIMEOUT"
--- a/src/exo/worker/runner/runner_supervisor.py
+++ b/src/exo/worker/runner/runner_supervisor.py
@@ -47,11 +47,9 @@ class RunnerSupervisor:
    _ev_recv: MpReceiver[Event]
    _task_sender: MpSender[Task]
    _event_sender: Sender[Event]
-    _cancel_sender: MpSender[TaskId]
    status: RunnerStatus = field(default_factory=RunnerIdle, init=False)
    pending: dict[TaskId, anyio.Event] = field(default_factory=dict, init=False)
    completed: set[TaskId] = field(default_factory=set, init=False)
-    cancelled: set[TaskId] = field(default_factory=set, init=False)

    @classmethod
    def create(
@@ -62,8 +60,8 @@ class RunnerSupervisor:
        initialize_timeout: float = 400,
    ) -> Self:
        ev_send, ev_recv = mp_channel[Event]()
+        # A task is kind of a runner command
        task_sender, task_recv = mp_channel[Task]()
-        cancel_sender, cancel_recv = mp_channel[TaskId]()

        runner_process = Process(
            target=entrypoint,
@@ -71,7 +69,6 @@ class RunnerSupervisor:
                bound_instance,
                ev_send,
                task_recv,
-                cancel_recv,
                logger,
            ),
            daemon=True,
@@ -86,7 +83,6 @@ class RunnerSupervisor:
            initialize_timeout=initialize_timeout,
            _ev_recv=ev_recv,
            _task_sender=task_sender,
-            _cancel_sender=cancel_sender,
            _event_sender=event_sender,
        )

@@ -101,8 +97,6 @@ class RunnerSupervisor:
        self._ev_recv.close()
        self._task_sender.close()
        self._event_sender.close()
-        self._cancel_sender.send(TaskId("CANCEL_CURRENT_TASK"))
-        self._cancel_sender.close()
        self.runner_process.join(1)
        if not self.runner_process.is_alive():
            logger.info("Runner process succesfully terminated")
@@ -118,6 +112,14 @@ class RunnerSupervisor:
        logger.critical("Runner process didn't respond to SIGTERM, killing")
        self.runner_process.kill()

+        self.runner_process.join(1)
+        if not self.runner_process.is_alive():
+            return
+
+        logger.critical(
+            "Runner process didn't respond to SIGKILL. System resources may have leaked"
+        )
+
    async def start_task(self, task: Task):
        if task.task_id in self.pending:
            logger.warning(
@@ -139,17 +141,6 @@ class RunnerSupervisor:
            return
        await event.wait()

-    async def cancel_task(self, task_id: TaskId):
-        if task_id in self.completed:
-            logger.info(f"Unable to cancel {task_id} as it has been completed")
-            return
-        self.cancelled.add(task_id)
-        with anyio.move_on_after(0.5) as scope:
-            await self._cancel_sender.send_async(task_id)
-        if scope.cancel_called:
-            logger.error("RunnerSupervisor cancel pipe blocked")
-            await self._check_runner(TimeoutError("cancel pipe blocked"))
-
    async def _forward_events(self):
        with self._ev_recv as events:
            try:
@@ -191,7 +182,7 @@ class RunnerSupervisor:
        logger.info("Checking runner's status")
        if self.runner_process.is_alive():
            logger.info("Runner was found to be alive, attempting to join process")
-            await to_thread.run_sync(self.runner_process.join, 5)
+            await to_thread.run_sync(self.runner_process.join, 1)
        rc = self.runner_process.exitcode
        logger.info(f"RunnerSupervisor exited with exit code {rc}")
        if rc == 0:
--- a/src/exo/worker/runner/tool_parsers.py
+++ b/src/exo/worker/runner/tool_parsers.py
@@ -1,72 +0,0 @@
-import json
-from dataclasses import dataclass
-from typing import Any, Callable
-
-from exo.shared.types.api import ToolCallItem
-
-
-@dataclass
-class ToolParser:
-    start_parsing: str
-    end_parsing: str
-    parse_tool_calls: Callable[[str], list[ToolCallItem] | None]
-
-
-def make_mlx_parser(
-    tool_call_start: str,
-    tool_call_end: str,
-    tool_parser: Callable[[str], dict[str, Any] | list[dict[str, Any]]],
-) -> ToolParser:
-    def parse_tool_calls(text: str) -> list[ToolCallItem] | None:
-        try:
-            text = text.removeprefix(tool_call_start)
-            text = text.removesuffix(tool_call_end)
-            parsed = tool_parser(text)
-            if isinstance(parsed, list):
-                return [ToolCallItem.model_validate(_flatten(p)) for p in parsed]
-            else:
-                return [ToolCallItem.model_validate(_flatten(parsed))]
-
-        except Exception:
-            return None
-
-    return ToolParser(
-        start_parsing=tool_call_start,
-        end_parsing=tool_call_end,
-        parse_tool_calls=parse_tool_calls,
-    )
-
-
-# TODO / example code:
-def _parse_json_calls(text: str) -> list[ToolCallItem] | None:
-    try:
-        text = text.removeprefix("<tool_call>")
-        text = text.removesuffix("</tool_call>")
-        top_level = {
-            k: json.dumps(v) if isinstance(v, (dict, list)) else v
-            for k, v in json.loads(text).items()  # pyright: ignore[reportAny]
-        }
-        return [ToolCallItem.model_validate(top_level)]
-    except Exception:
-        return None
-
-
-def _flatten(p: dict[str, Any]) -> dict[str, str]:
-    return {
-        k: json.dumps(v) if isinstance(v, (dict, list)) else str(v)  # pyright: ignore[reportAny]
-        for k, v in p.items()  # pyright: ignore[reportAny]
-    }
-
-
-json_tool_parser = ToolParser(
-    start_parsing="<tool_call>",
-    end_parsing="</tool_call>",
-    parse_tool_calls=_parse_json_calls,
-)
-
-
-def infer_tool_parser(chat_template: str) -> ToolParser | None:
-    """Attempt to auto-infer a tool parser from the chat template."""
-    if "<tool_call>" in chat_template and "tool_call.name" in chat_template:
-        return json_tool_parser
-    return None
--- a/src/exo/worker/tests/unittests/test_runner/test_event_ordering.py
+++ b/src/exo/worker/tests/unittests/test_runner/test_event_ordering.py
@@ -1,9 +1,7 @@
 # Check tasks are complete before runner is ever ready.
-import unittest.mock
 from collections.abc import Iterable
 from typing import Callable

-import mlx.core as mx
 import pytest

 import exo.worker.runner.runner as mlx_runner
@@ -21,7 +19,6 @@ from exo.shared.types.tasks import (
    Shutdown,
    StartWarmup,
    Task,
-    TaskId,
    TaskStatus,
    TextGeneration,
 )
@@ -116,7 +113,6 @@ def patch_out_mlx(monkeypatch: pytest.MonkeyPatch):
    monkeypatch.setattr(mlx_runner, "load_mlx_items", make_nothin((1, MockTokenizer)))
    monkeypatch.setattr(mlx_runner, "warmup_inference", make_nothin(1))
    monkeypatch.setattr(mlx_runner, "_check_for_debug_prompts", nothin)
-    monkeypatch.setattr(mlx_runner, "mx_any", make_nothin(False))
    # Mock apply_chat_template since we're using a fake tokenizer (integer 1).
    # Returns a prompt without thinking tag so detect_thinking_prompt_suffix returns None.
    monkeypatch.setattr(mlx_runner, "apply_chat_template", make_nothin("test prompt"))
@@ -167,7 +163,6 @@ def _run(tasks: Iterable[Task]):
    )

    task_sender, task_receiver = mp_channel[Task]()
-    _cancel_sender, cancel_receiver = mp_channel[TaskId]()
    event_sender = EventCollector()

    with task_sender:
@@ -178,16 +173,8 @@ def _run(tasks: Iterable[Task]):
        # this is some c++ nonsense
        task_receiver.close = nothin
        task_receiver.join = nothin
-        with unittest.mock.patch(
-            "exo.worker.runner.runner.mx.distributed.all_gather",
-            make_nothin(mx.array([1])),
-        ):
-            mlx_runner.main(
-                bound_instance,
-                event_sender,  # pyright: ignore[reportArgumentType]
-                task_receiver,
-                cancel_receiver,
-            )
+
+        mlx_runner.main(bound_instance, event_sender, task_receiver)  # type: ignore[arg-type]

        return event_sender.events

--- a/src/exo/worker/tests/unittests/test_runner/test_parse_tool_calls.py
+++ b/src/exo/worker/tests/unittests/test_runner/test_parse_tool_calls.py
@@ -5,13 +5,12 @@ from typing import Any

 from exo.shared.types.worker.runner_response import GenerationResponse, ToolCallResponse
 from exo.worker.runner.runner import parse_tool_calls
-from exo.worker.runner.tool_parsers import make_mlx_parser


 def _make_responses(
    texts: list[str],
    finish_on_last: bool = True,
-) -> Generator[GenerationResponse]:
+) -> Generator[GenerationResponse | ToolCallResponse]:
    """Create a sequence of GenerationResponses from text strings."""
    for i, text in enumerate(texts):
        is_last = i == len(texts) - 1
@@ -23,13 +22,10 @@ def _make_responses(
        )


-def _dummier_parser(text: str) -> dict[str, Any]:
+def _dummy_parser(text: str) -> dict[str, Any]:
    return {"name": "test_fn", "arguments": {"arg": text}}


-_dummy_parser = make_mlx_parser("<tool_call>", "</tool_call>", _dummier_parser)
-
-
 class TestParseToolCalls:
    """Tests for parse_tool_calls generator."""

@@ -39,6 +35,8 @@ class TestParseToolCalls:
        results = list(
            parse_tool_calls(
                _make_responses(texts, finish_on_last=False),
+                "<tool_call>",
+                "</tool_call>",
                _dummy_parser,
            )
        )
@@ -52,6 +50,8 @@ class TestParseToolCalls:
        results = list(
            parse_tool_calls(
                _make_responses(texts),
+                "<tool_call>",
+                "</tool_call>",
                _dummy_parser,
            )
        )
@@ -76,7 +76,9 @@ class TestParseToolCalls:
        results = list(
            parse_tool_calls(
                _make_responses(texts, finish_on_last=False),
-                make_mlx_parser("<tool_call>", "</tool_call>", _failing_parser),
+                "<tool_call>",
+                "</tool_call>",
+                _failing_parser,
            )
        )

--- a/uv.lock
+++ b/uv.lock
@@ -377,8 +377,8 @@ dependencies = [
    { name = "hypercorn", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
    { name = "loguru", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
    { name = "mflux", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
-    { name = "mlx", version = "0.30.6", source = { registry = "https://pypi.org/simple" }, extra = ["cpu"], marker = "sys_platform == 'linux'" },
-    { name = "mlx", version = "0.30.7.dev20260218+14841977", source = { git = "https://github.com/rltakashige/mlx-jaccl-fix-small-recv.git?branch=address-rdma-gpu-locks#1484197707f35186ad3bd614357c7c47fdf86ebc" }, marker = "sys_platform == 'darwin'" },
+    { name = "mlx", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
+    { name = "mlx", extra = ["cpu"], marker = "sys_platform == 'linux'" },
    { name = "mlx-lm", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
    { name = "msgspec", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
    { name = "openai-harmony", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
@@ -416,9 +416,9 @@ requires-dist = [
    { name = "hypercorn", specifier = ">=0.18.0" },
    { name = "loguru", specifier = ">=0.7.3" },
    { name = "mflux", specifier = "==0.15.5" },
-    { name = "mlx", marker = "sys_platform == 'darwin'", git = "https://github.com/rltakashige/mlx-jaccl-fix-small-recv.git?branch=address-rdma-gpu-locks" },
+    { name = "mlx", marker = "sys_platform == 'darwin'", specifier = "==0.30.6" },
    { name = "mlx", extras = ["cpu"], marker = "sys_platform == 'linux'", specifier = "==0.30.6" },
-    { name = "mlx-lm", specifier = "==0.30.7" },
+    { name = "mlx-lm", specifier = "==0.30.6" },
    { name = "msgspec", specifier = ">=0.19.0" },
    { name = "openai-harmony", specifier = ">=0.0.8" },
    { name = "pillow", specifier = ">=11.0,<12.0" },
@@ -1020,8 +1020,8 @@ dependencies = [
    { name = "fonttools", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
    { name = "huggingface-hub", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
    { name = "matplotlib", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
-    { name = "mlx", version = "0.30.6", source = { registry = "https://pypi.org/simple" }, extra = ["cuda13"], marker = "sys_platform == 'linux'" },
-    { name = "mlx", version = "0.30.7.dev20260218+14841977", source = { git = "https://github.com/rltakashige/mlx-jaccl-fix-small-recv.git?branch=address-rdma-gpu-locks#1484197707f35186ad3bd614357c7c47fdf86ebc" }, marker = "sys_platform == 'darwin'" },
+    { name = "mlx", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
+    { name = "mlx", extra = ["cuda13"], marker = "sys_platform == 'linux'" },
    { name = "numpy", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
    { name = "opencv-python", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
    { name = "piexif", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
@@ -1048,12 +1048,18 @@ wheels = [
 name = "mlx"
 version = "0.30.6"
 source = { registry = "https://pypi.org/simple" }
-resolution-markers = [
-    "sys_platform == 'linux'",
+dependencies = [
+    { name = "mlx-metal", marker = "sys_platform == 'darwin'" },
 ]
 wheels = [
+    { url = "https://files.pythonhosted.org/packages/ae/5b/e460e144a34d5529e010056cccf50b538d56ed001473bc6b246018fd58cb/mlx-0.30.6-cp313-cp313-macosx_14_0_arm64.whl", hash = "sha256:ed86f8bffc174c2f259ca589ea25464c96cf69d1bb457074a2bf2ef53737e54f", size = 573515, upload-time = "2026-02-06T03:45:23.405Z" },
+    { url = "https://files.pythonhosted.org/packages/60/25/69833fefb9a3fef30b56792b1bcd022496c4fea83e45411d289b77ef7546/mlx-0.30.6-cp313-cp313-macosx_15_0_arm64.whl", hash = "sha256:c52294958269e20f300639a17c1900ca8fc737d859ddda737f9811e94bd040e5", size = 573516, upload-time = "2026-02-06T03:45:24.618Z" },
+    { url = "https://files.pythonhosted.org/packages/9c/6a/7e7fbeebc5cb51b6a5eba96b263a6298707bcbdc059f4b0b73e088bc3dea/mlx-0.30.6-cp313-cp313-macosx_26_0_arm64.whl", hash = "sha256:b5b6636f7c49a4d86d8ec82643b972f45a144a7a9f3a967b27b2e6e22cf71e6a", size = 573592, upload-time = "2026-02-06T03:45:25.928Z" },
    { url = "https://files.pythonhosted.org/packages/93/06/280f6f2ba80520a7109730425eda0d966658793aa0d02d8be8d351f75253/mlx-0.30.6-cp313-cp313-manylinux_2_35_aarch64.whl", hash = "sha256:67e6c9e30a9faeacc209917ef5523177cf9b086914b6b5d83ff886e4294b727d", size = 622011, upload-time = "2026-02-06T03:45:28.165Z" },
    { url = "https://files.pythonhosted.org/packages/fe/35/f872afbee9c079cc69924d9e9c46f5663adb7da58cba3511db082dd307c1/mlx-0.30.6-cp313-cp313-manylinux_2_35_x86_64.whl", hash = "sha256:47db8b16fcb6f6c5a47c0bdb24ed377b41237017ac93aa6cb6aa206c9bdf82e4", size = 663650, upload-time = "2026-02-06T03:45:30.315Z" },
+    { url = "https://files.pythonhosted.org/packages/60/23/361dc7a5797634e4d7e9bdd6564c6b28f9b1246672632def2f91bf066b18/mlx-0.30.6-cp314-cp314-macosx_14_0_arm64.whl", hash = "sha256:78804a89dcff4a838f7c2da72392fe87a523e95122a3c840e53df019122aad45", size = 575028, upload-time = "2026-02-06T03:45:31.549Z" },
+    { url = "https://files.pythonhosted.org/packages/a8/69/1854484d414171586814dfbe8def95f75c4ea2c7341ba13ba8ee675f7c62/mlx-0.30.6-cp314-cp314-macosx_15_0_arm64.whl", hash = "sha256:ec13584ab069665cc7ad34a05494d9291cd623aef6ae96be48875fc87cfc25d6", size = 575026, upload-time = "2026-02-06T03:45:33.072Z" },
+    { url = "https://files.pythonhosted.org/packages/6b/b8/3adbc441924209a7e4c568308b2a0b54bd09aee6a68db5bae85304791e54/mlx-0.30.6-cp314-cp314-macosx_26_0_arm64.whl", hash = "sha256:b2c5e8a090a753ef99a1380a4d059c983083f36198864f6df9faaf1223d083df", size = 575041, upload-time = "2026-02-06T03:45:34.814Z" },
    { url = "https://files.pythonhosted.org/packages/3f/54/9d9e06804fb2088202a2cdf60458e00b221f71420bea285720b60f9e82b5/mlx-0.30.6-cp314-cp314-manylinux_2_35_aarch64.whl", hash = "sha256:9ceddede4af0de31d1f6b3099f70e5469d60cd7c546975dedbdbeab3519cab3f", size = 624002, upload-time = "2026-02-06T03:45:36Z" },
    { url = "https://files.pythonhosted.org/packages/42/92/3140a15a50cb1f9267a6552171e1dfa577861de53e093124bc43707f2a0e/mlx-0.30.6-cp314-cp314-manylinux_2_35_x86_64.whl", hash = "sha256:4a6ffd2d16728cf95f63a1b555d7c2eaeea686a0e6b73228bd265411cb5d77a4", size = 663569, upload-time = "2026-02-06T03:45:37.242Z" },
 ]
@@ -1066,14 +1072,6 @@ cuda13 = [
    { name = "mlx-cuda-13", marker = "sys_platform == 'linux'" },
 ]

-[[package]]
-name = "mlx"
-version = "0.30.7.dev20260218+14841977"
-source = { git = "https://github.com/rltakashige/mlx-jaccl-fix-small-recv.git?branch=address-rdma-gpu-locks#1484197707f35186ad3bd614357c7c47fdf86ebc" }
-resolution-markers = [
-    "sys_platform == 'darwin'",
-]
-
 [[package]]
 name = "mlx-cpu"
 version = "0.30.6"
@@ -1100,20 +1098,30 @@ wheels = [

 [[package]]
 name = "mlx-lm"
-version = "0.30.7"
+version = "0.30.6"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
    { name = "jinja2", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
-    { name = "mlx", version = "0.30.7.dev20260218+14841977", source = { git = "https://github.com/rltakashige/mlx-jaccl-fix-small-recv.git?branch=address-rdma-gpu-locks#1484197707f35186ad3bd614357c7c47fdf86ebc" }, marker = "sys_platform == 'darwin'" },
+    { name = "mlx", marker = "sys_platform == 'darwin'" },
    { name = "numpy", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
    { name = "protobuf", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
    { name = "pyyaml", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
    { name = "sentencepiece", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
    { name = "transformers", marker = "sys_platform == 'darwin' or sys_platform == 'linux'" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/66/0d/56542e2ae13ec6f542d3977d7cff89a205d4f6c5122e0ce23f33265f61c9/mlx_lm-0.30.7.tar.gz", hash = "sha256:e5f31ac58d9f2381f28e1ba639ff903e64f7cff1bdc245c0bc97f72264be329c", size = 275764, upload-time = "2026-02-12T18:41:11.86Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/76/cb/815deddc8699b1f694d7e1f9cbed52934c03a8b49432c8add72932bb2f0b/mlx_lm-0.30.6.tar.gz", hash = "sha256:807e042d7040268f1b19190b7eaefd8b2efbff5590a65460974ad4225b91dda1", size = 271733, upload-time = "2026-02-04T21:27:45.741Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/1e/17/a41c798a3d9cbdc47f39c6db5bba4c2cd199203ead26bf911cb03b644070/mlx_lm-0.30.7-py3-none-any.whl", hash = "sha256:17442a4bf01c4c2d3bca1e647712fe44f19890c3f1eadc8589d389e57b44b9bf", size = 386591, upload-time = "2026-02-12T18:41:10.236Z" },
+    { url = "https://files.pythonhosted.org/packages/20/5f/01d281f1fa8a1521d5936659beb4f5ab1f32b463d059263cf9d4cef969d9/mlx_lm-0.30.6-py3-none-any.whl", hash = "sha256:a7405bd581eacc4bf8209d7a6b7f23629585a0d7c6740c2a97e51fee35b3b0e1", size = 379451, upload-time = "2026-02-04T21:27:43.222Z" },
+]
+
+[[package]]
+name = "mlx-metal"
+version = "0.30.6"
+source = { registry = "https://pypi.org/simple" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/f3/85/44406b521f920248fad621334d4dc15e77660a494edf890e7cbee33bf38d/mlx_metal-0.30.6-py3-none-macosx_14_0_arm64.whl", hash = "sha256:ea6d0c973def9a5b4f652cc77036237db3f88c9d0af63701d76b5fddde99b820", size = 38437818, upload-time = "2026-02-06T03:44:56.19Z" },
+    { url = "https://files.pythonhosted.org/packages/d0/cb/10a516995f7d0c154b0d7e633c54b51e96977a86a355105b6474cfcbe0d0/mlx_metal-0.30.6-py3-none-macosx_15_0_arm64.whl", hash = "sha256:0f8cb94634d07e06a372d6ad9a090f38a18bab1ff19a140aede60eacf707bb94", size = 38433701, upload-time = "2026-02-06T03:44:59.678Z" },
+    { url = "https://files.pythonhosted.org/packages/4c/7d/70cb272f7373c334709f210ed8420511fc9d64d05a7a646c0b3b94c29c04/mlx_metal-0.30.6-py3-none-macosx_26_0_arm64.whl", hash = "sha256:d761ae26304f2c4b454eeea7f612a56919d9e5e57dbb1dc0788f8e34aa6f41c2", size = 47718448, upload-time = "2026-02-06T03:45:03.133Z" },
 ]

 [[package]]
Author	SHA1	Message	Date
Alex Cheema	3d4f8130c9	Merge remote-tracking branch 'origin/main' into alexcheema/speculative-decoding	2026-02-13 09:39:11 -08:00
Alex Cheema	6762a0a8ba	Add draft_model and num_draft_tokens fields to PlaceInstance command PlaceInstance was missing these fields that are accessed in placement.py when creating MlxJaccl and MlxRing instances with speculative decoding support, causing type errors after merging main. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>	2026-02-13 05:51:11 -08:00
Alex Cheema	6524820e5b	Merge remote-tracking branch 'origin/main' into alexcheema/speculative-decoding # Conflicts: # src/exo/shared/types/commands.py	2026-02-13 05:46:52 -08:00
Alex Cheema	90e2a20091	Merge remote-tracking branch 'origin/main' into alexcheema/speculative-decoding Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>	2026-02-05 06:09:50 -08:00
Alex Cheema	55152fa99d	Merge remote-tracking branch 'origin/main' into alexcheema/speculative-decoding # Conflicts: # dashboard/src/routes/+page.svelte # src/exo/master/api.py # src/exo/shared/types/api.py # src/exo/shared/types/commands.py # src/exo/worker/engines/mlx/generator/generate.py # src/exo/worker/main.py # src/exo/worker/plan.py # src/exo/worker/runner/runner.py	2026-02-05 06:07:51 -08:00
Alex Cheema	e7f61c3494	Add speculative decoding support with draft models This adds support for speculative decoding using draft models to accelerate inference. Key changes: - Add draft_model and num_draft_tokens fields to Instance for configuration - Add SetDraftModel task to load/clear draft models on running instances - Add InstanceDraftModelUpdated event to propagate draft model changes - Add SetInstanceDraftModel command and API endpoint for runtime updates - Update plan.py to download draft models in parallel with main model - Update runner to load draft model during LoadModel phase - Add draft model UI to dashboard instances panel (both views) The draft model can be configured when creating an instance or updated on a running instance via the dashboard or API. Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>	2026-01-22 13:09:26 +00:00