fix(distributed): stop queue loops on agent nodes + dead-letter cap

pending_backend_ops rows targeting agent-type workers looped forever:
the reconciler fan-out hit a NATS subject the worker doesn't subscribe
to, returned ErrNoResponders, we marked the node unhealthy, and the
health monitor flipped it back to healthy on the next heartbeat. Next
tick, same row, same failure.

Three related fixes:

1. enqueueAndDrainBackendOp skips nodes whose NodeType != backend.
   Agent workers handle agent NATS subjects, not backend.install /
   delete / list, so enqueueing for them guarantees an infinite retry
   loop. Silent skip is correct — they aren't consumers of these ops.

2. Reconciler drain mirrors enqueueAndDrainBackendOp's behavior on
   nats.ErrNoResponders: mark the node unhealthy before recording the
   failure, so subsequent ListDuePendingBackendOps (filters by
   status=healthy) stops picking the row until the node actually
   recovers. Matches the synchronous fan-out path.

3. Dead-letter cap at maxPendingBackendOpAttempts (10). After ~1h of
   exponential backoff the row is a poison message; further retries
   just thrash NATS. Row is deleted and logged at ERROR so it stays
   visible without staying infinite.

Plus a one-shot startup cleanup in NewNodeRegistry: drop queue rows
that target agent-type nodes, non-existent nodes, or carry an empty
backend name. Guarded by the same schema-migration advisory lock so
only one instance performs it. The guards above prevent new rows of
this shape; this closes the migration gap for existing ones.

Tests: the prune migration (valid row stays, agent + empty-name rows
drop) on top of existing upsert / backoff coverage.

backend/cpp/turboquant/Makefile

@@ -1,7 +1,7 @@
 
 # Pinned to the HEAD of feature/turboquant-kv-cache on https://github.com/TheTom/llama-cpp-turboquant.
 # Auto-bumped nightly by .github/workflows/bump_deps.yaml.
-TURBOQUANT_VERSION?=627ebbc6e27727bd4f65422d8aa60b13404993c8
+TURBOQUANT_VERSION?=45f8a066ed5f5bb38c695cec532f6cef9f4efa9d
 LLAMA_REPO?=https://github.com/TheTom/llama-cpp-turboquant
 
 CMAKE_ARGS?=

backend/cpp/turboquant/patch-grpc-server.sh

@@ -1,22 +1,13 @@
 #!/bin/bash
-# Patch the shared backend/cpp/llama-cpp/grpc-server.cpp *copy* used by the
-# turboquant build to account for two gaps between upstream and the fork:
+# Augment the shared backend/cpp/llama-cpp/grpc-server.cpp allow-list of KV-cache
+# types so the gRPC `LoadModel` call accepts the TurboQuant-specific
+# `turbo2` / `turbo3` / `turbo4` cache types.
 #
-#   1. Augment the kv_cache_types[] allow-list so `LoadModel` accepts the
-#      fork-specific `turbo2` / `turbo3` / `turbo4` cache types.
-#   2. Replace `get_media_marker()` (added upstream in ggml-org/llama.cpp#21962,
-#      server-side random per-instance marker) with the legacy "<__media__>"
-#      literal. The fork branched before that PR, so server-common.cpp has no
-#      get_media_marker symbol. The fork's mtmd_default_marker() still returns
-#      "<__media__>", and Go-side tooling falls back to that sentinel when the
-#      backend does not expose media_marker, so substituting the literal keeps
-#      behavior identical on the turboquant path.
+# We do this on the *copy* sitting in turboquant-<flavor>-build/, never on the
+# original under backend/cpp/llama-cpp/, so the stock llama-cpp build keeps
+# compiling against vanilla upstream which does not know about GGML_TYPE_TURBO*.
 #
-# We patch the *copy* sitting in turboquant-<flavor>-build/, never the original
-# under backend/cpp/llama-cpp/, so the stock llama-cpp build keeps compiling
-# against vanilla upstream.
-#
-# Idempotent: skips each insertion if its marker is already present (so re-runs
+# Idempotent: skips the insertion if the marker is already present (so re-runs
 # of the same build dir don't double-insert).
 
 set -euo pipefail
@@ -34,47 +25,33 @@ if [[ ! -f "$SRC" ]]; then
 fi
 
 if grep -q 'GGML_TYPE_TURBO2_0' "$SRC"; then
-    echo "==> $SRC already has TurboQuant cache types, skipping KV allow-list patch"
-else
-    echo "==> patching $SRC to allow turbo2/turbo3/turbo4 KV-cache types"
-
-    # Insert the three TURBO entries right after the first `    GGML_TYPE_Q5_1,`
-    # line (the kv_cache_types[] allow-list). Using awk because the builder image
-    # does not ship python3, and GNU sed's multi-line `a\` quoting is awkward.
-    awk '
-        /^    GGML_TYPE_Q5_1,$/ && !done {
-            print
-            print "    // turboquant fork extras — added by patch-grpc-server.sh"
-            print "    GGML_TYPE_TURBO2_0,"
-            print "    GGML_TYPE_TURBO3_0,"
-            print "    GGML_TYPE_TURBO4_0,"
-            done = 1
-            next
-        }
-        { print }
-        END {
-            if (!done) {
-                print "patch-grpc-server.sh: anchor `    GGML_TYPE_Q5_1,` not found" > "/dev/stderr"
-                exit 1
-            }
-        }
-    ' "$SRC" > "$SRC.tmp"
-    mv "$SRC.tmp" "$SRC"
-
-    echo "==> KV allow-list patch OK"
+    echo "==> $SRC already has TurboQuant cache types, skipping"
+    exit 0
 fi
 
-if grep -q 'get_media_marker()' "$SRC"; then
-    echo "==> patching $SRC to replace get_media_marker() with legacy \"<__media__>\" literal"
-    # Only one call site today (ModelMetadata), but replace all occurrences to
-    # stay robust if upstream adds more. Use a temp file to avoid relying on
-    # sed -i portability (the builder image uses GNU sed, but keeping this
-    # consistent with the awk block above).
-    sed 's/get_media_marker()/"<__media__>"/g' "$SRC" > "$SRC.tmp"
-    mv "$SRC.tmp" "$SRC"
-    echo "==> get_media_marker() substitution OK"
-else
-    echo "==> $SRC has no get_media_marker() call, skipping media-marker patch"
-fi
+echo "==> patching $SRC to allow turbo2/turbo3/turbo4 KV-cache types"
 
-echo "==> all patches applied"
+# Insert the three TURBO entries right after the first `    GGML_TYPE_Q5_1,`
+# line (the kv_cache_types[] allow-list). Using awk because the builder image
+# does not ship python3, and GNU sed's multi-line `a\` quoting is awkward.
+awk '
+    /^    GGML_TYPE_Q5_1,$/ && !done {
+        print
+        print "    // turboquant fork extras — added by patch-grpc-server.sh"
+        print "    GGML_TYPE_TURBO2_0,"
+        print "    GGML_TYPE_TURBO3_0,"
+        print "    GGML_TYPE_TURBO4_0,"
+        done = 1
+        next
+    }
+    { print }
+    END {
+        if (!done) {
+            print "patch-grpc-server.sh: anchor `    GGML_TYPE_Q5_1,` not found" > "/dev/stderr"
+            exit 1
+        }
+    }
+' "$SRC" > "$SRC.tmp"
+mv "$SRC.tmp" "$SRC"
+
+echo "==> patched OK"

backend/cpp/turboquant/patches/0001-server-respect-the-ignore-eos-flag.patch

@@ -0,0 +1,83 @@
+From 660600081fb7b9b769ded5c805a2d39a419f0a0d Mon Sep 17 00:00:00 2001
+From: Yuri Khrustalev <ykhrustalev@users.noreply.github.com>
+Date: Wed, 8 Apr 2026 11:12:15 -0400
+Subject: [PATCH] server: respect the ignore eos flag (#21203)
+
+---
+ tools/server/server-context.cpp | 3 +++
+ tools/server/server-context.h   | 3 +++
+ tools/server/server-task.cpp    | 3 ++-
+ tools/server/server-task.h      | 1 +
+ 4 files changed, 9 insertions(+), 1 deletion(-)
+
+diff --git a/tools/server/server-context.cpp b/tools/server/server-context.cpp
+index 9d3ac538..b31981c5 100644
+--- a/tools/server/server-context.cpp
++++ b/tools/server/server-context.cpp
+@@ -3033,6 +3033,8 @@ server_context_meta server_context::get_meta() const {
+         /* fim_rep_token          */ llama_vocab_fim_rep(impl->vocab),
+         /* fim_sep_token          */ llama_vocab_fim_sep(impl->vocab),
+ 
++        /* logit_bias_eog         */ impl->params_base.sampling.logit_bias_eog,
++
+         /* model_vocab_type       */ llama_vocab_type(impl->vocab),
+         /* model_vocab_n_tokens   */ llama_vocab_n_tokens(impl->vocab),
+         /* model_n_ctx_train      */ llama_model_n_ctx_train(impl->model),
+@@ -3117,6 +3119,7 @@ std::unique_ptr<server_res_generator> server_routes::handle_completions_impl(
+                     ctx_server.vocab,
+                     params,
+                     meta->slot_n_ctx,
++                    meta->logit_bias_eog,
+                     data);
+             task.id_slot = json_value(data, "id_slot", -1);
+ 
+diff --git a/tools/server/server-context.h b/tools/server/server-context.h
+index d7ce8735..6ea9afc0 100644
+--- a/tools/server/server-context.h
++++ b/tools/server/server-context.h
+@@ -39,6 +39,9 @@ struct server_context_meta {
+     llama_token fim_rep_token;
+     llama_token fim_sep_token;
+ 
++    // sampling
++    std::vector<llama_logit_bias> logit_bias_eog;
++
+     // model meta
+     enum llama_vocab_type model_vocab_type;
+     int32_t model_vocab_n_tokens;
+diff --git a/tools/server/server-task.cpp b/tools/server/server-task.cpp
+index 4cc87bc5..856b3f0e 100644
+--- a/tools/server/server-task.cpp
++++ b/tools/server/server-task.cpp
+@@ -239,6 +239,7 @@ task_params server_task::params_from_json_cmpl(
+         const llama_vocab * vocab,
+         const common_params & params_base,
+         const int n_ctx_slot,
++        const std::vector<llama_logit_bias> & logit_bias_eog,
+         const json & data) {
+     task_params params;
+ 
+@@ -562,7 +563,7 @@ task_params server_task::params_from_json_cmpl(
+         if (params.sampling.ignore_eos) {
+             params.sampling.logit_bias.insert(
+                     params.sampling.logit_bias.end(),
+-                    defaults.sampling.logit_bias_eog.begin(), defaults.sampling.logit_bias_eog.end());
++                    logit_bias_eog.begin(), logit_bias_eog.end());
+         }
+     }
+ 
+diff --git a/tools/server/server-task.h b/tools/server/server-task.h
+index d855bf08..243e47a8 100644
+--- a/tools/server/server-task.h
++++ b/tools/server/server-task.h
+@@ -209,6 +209,7 @@ struct server_task {
+         const llama_vocab * vocab,
+         const common_params & params_base,
+         const int n_ctx_slot,
++        const std::vector<llama_logit_bias> & logit_bias_eog,
+         const json & data);
+ 
+     // utility function
+-- 
+2.43.0
+

backend/index.yaml

@@ -1008,20 +1008,6 @@
     nvidia-cuda-12: "cuda12-turboquant-development"
     nvidia-l4t-cuda-12: "nvidia-l4t-arm64-turboquant-development"
     nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-turboquant-development"
-- !!merge <<: *stablediffusionggml
-  name: "stablediffusion-ggml-development"
-  capabilities:
-    default: "cpu-stablediffusion-ggml-development"
-    nvidia: "cuda12-stablediffusion-ggml-development"
-    intel: "intel-sycl-f16-stablediffusion-ggml-development"
-    # amd: "rocm-stablediffusion-ggml-development"
-    vulkan: "vulkan-stablediffusion-ggml-development"
-    nvidia-l4t: "nvidia-l4t-arm64-stablediffusion-ggml-development"
-    metal: "metal-stablediffusion-ggml-development"
-    nvidia-cuda-13: "cuda13-stablediffusion-ggml-development"
-    nvidia-cuda-12: "cuda12-stablediffusion-ggml-development"
-    nvidia-l4t-cuda-12: "nvidia-l4t-arm64-stablediffusion-ggml-development"
-    nvidia-l4t-cuda-13: "cuda13-nvidia-l4t-arm64-stablediffusion-ggml-development"
 - !!merge <<: *neutts
   name: "cpu-neutts"
   uri: "quay.io/go-skynet/local-ai-backends:latest-cpu-neutts"

backend/rust/kokoros/src/service.rs

@@ -341,16 +341,6 @@ impl Backend for KokorosService {
         Err(Status::unimplemented("Not supported"))
     }
 
-    type AudioTranscriptionStreamStream =
-        ReceiverStream<Result<backend::TranscriptStreamResponse, Status>>;
-
-    async fn audio_transcription_stream(
-        &self,
-        _: Request<backend::TranscriptRequest>,
-    ) -> Result<Response<Self::AudioTranscriptionStreamStream>, Status> {
-        Err(Status::unimplemented("Not supported"))
-    }
-
     async fn sound_generation(
         &self,
         _: Request<backend::SoundGenerationRequest>,

core/services/nodes/managers_distributed.go

@@ -106,6 +106,13 @@ func (d *DistributedBackendManager) enqueueAndDrainBackendOp(ctx context.Context
 		if node.Status == StatusPending {
 			continue
 		}
+		// Backend lifecycle ops only make sense on backend-type workers.
+		// Agent workers don't subscribe to backend.install/delete/list, so
+		// enqueueing for them guarantees a forever-retrying row that the
+		// reconciler can never drain. Silently skip — they aren't consumers.
+		if node.NodeType != "" && node.NodeType != NodeTypeBackend {
+			continue
+		}
 		if err := d.registry.UpsertPendingBackendOp(ctx, node.ID, backend, op, galleriesJSON); err != nil {
 			xlog.Warn("Failed to enqueue backend op", "op", op, "node", node.Name, "backend", backend, "error", err)
 			result.Nodes = append(result.Nodes, NodeOpStatus{

core/services/nodes/reconciler.go

@@ -3,12 +3,14 @@ package nodes
 import (
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"time"
 
 	"github.com/mudler/LocalAI/core/services/advisorylock"
 	grpcclient "github.com/mudler/LocalAI/pkg/grpc"
 	"github.com/mudler/xlog"
+	"github.com/nats-io/nats.go"
 	"gorm.io/gorm"
 )
 
@@ -206,12 +208,47 @@ func (rc *ReplicaReconciler) drainPendingBackendOps(ctx context.Context) {
 			}
 			continue
 		}
+
+		// ErrNoResponders means the node has no active NATS subscription for
+		// this subject. Either its connection dropped, or it's the wrong
+		// node type entirely. Mark unhealthy so the health monitor's
+		// heartbeat-only pass doesn't immediately flip it back — and so
+		// ListDuePendingBackendOps (which filters by status=healthy) stops
+		// picking the row until the node genuinely recovers.
+		if errors.Is(applyErr, nats.ErrNoResponders) {
+			xlog.Warn("Reconciler: no NATS responders — marking node unhealthy",
+				"op", op.Op, "backend", op.Backend, "node", op.NodeID)
+			_ = rc.registry.MarkUnhealthy(ctx, op.NodeID)
+		}
+
+		// Dead-letter cap: after maxAttempts the row is the reconciler
+		// equivalent of a poison message. Delete it loudly so the queue
+		// doesn't churn NATS every tick forever — operators can re-issue
+		// the op from the UI if they still want it applied.
+		if op.Attempts+1 >= maxPendingBackendOpAttempts {
+			xlog.Error("Reconciler: abandoning pending backend op after max attempts",
+				"op", op.Op, "backend", op.Backend, "node", op.NodeID,
+				"attempts", op.Attempts+1, "last_error", applyErr)
+			if err := rc.registry.DeletePendingBackendOp(ctx, op.ID); err != nil {
+				xlog.Warn("Reconciler: failed to delete abandoned op row", "id", op.ID, "error", err)
+			}
+			continue
+		}
+
 		_ = rc.registry.RecordPendingBackendOpFailure(ctx, op.ID, applyErr.Error())
 		xlog.Warn("Reconciler: pending backend op retry failed",
 			"op", op.Op, "backend", op.Backend, "node", op.NodeID, "attempts", op.Attempts+1, "error", applyErr)
 	}
 }
 
+// maxPendingBackendOpAttempts caps how many times the reconciler retries a
+// failing row before dead-lettering it. Ten attempts at exponential backoff
+// (30s → 15m cap) is >1h of wall-clock patience — well past any transient
+// worker restart or network blip. Poisoned rows beyond that are almost
+// certainly structural (wrong node type, non-existent gallery entry) and no
+// amount of further retrying will help.
+const maxPendingBackendOpAttempts = 10
+
 // probeLoadedModels gRPC-health-checks model addresses that the DB says are
 // loaded. If a model's backend process is gone (OOM, crash, manual restart)
 // we remove the row so ghosts don't linger. Only probes rows older than

core/services/nodes/reconciler_test.go

@@ -373,4 +373,30 @@ var _ = Describe("ReplicaReconciler — state reconciliation", func() {
 			Expect(row.NextRetryAt).To(BeTemporally(">", before))
 		})
 	})
+
+	Describe("NewNodeRegistry malformed-row pruning", func() {
+		It("drops queue rows for agent nodes and non-existent nodes on startup", func() {
+			agent := &BackendNode{Name: "agent-1", NodeType: NodeTypeAgent, Address: "x"}
+			Expect(registry.Register(context.Background(), agent, true)).To(Succeed())
+			backend := &BackendNode{Name: "backend-1", NodeType: NodeTypeBackend, Address: "y"}
+			Expect(registry.Register(context.Background(), backend, true)).To(Succeed())
+
+			// Three rows: one for a valid backend node (should survive),
+			// one for an agent node (pruned), one for an empty backend name
+			// on the valid node (pruned).
+			Expect(registry.UpsertPendingBackendOp(context.Background(), backend.ID, "foo", OpBackendInstall, nil)).To(Succeed())
+			Expect(registry.UpsertPendingBackendOp(context.Background(), agent.ID, "foo", OpBackendInstall, nil)).To(Succeed())
+			Expect(registry.UpsertPendingBackendOp(context.Background(), backend.ID, "", OpBackendInstall, nil)).To(Succeed())
+
+			// Re-instantiating the registry runs the cleanup migration.
+			_, err := NewNodeRegistry(db)
+			Expect(err).ToNot(HaveOccurred())
+
+			var rows []PendingBackendOp
+			Expect(db.Find(&rows).Error).To(Succeed())
+			Expect(rows).To(HaveLen(1))
+			Expect(rows[0].NodeID).To(Equal(backend.ID))
+			Expect(rows[0].Backend).To(Equal("foo"))
+		})
+	})
 })

core/services/nodes/registry.go

@@ -148,6 +148,30 @@ func NewNodeRegistry(db *gorm.DB) (*NodeRegistry, error) {
 	}); err != nil {
 		return nil, fmt.Errorf("migrating node tables: %w", err)
 	}
+
+	// One-shot cleanup of queue rows that can never drain: ops targeted at
+	// agent workers (wrong subscription set), at non-existent nodes, or with
+	// an empty backend name. The guard in enqueueAndDrainBackendOp prevents
+	// new ones from being written, but rows persisted by earlier versions
+	// keep the reconciler busy retrying a permanently-failing NATS request
+	// every 30s. Guarded by the same migration advisory lock so only one
+	// frontend runs it.
+	_ = advisorylock.WithLockCtx(context.Background(), db, advisorylock.KeySchemaMigrate, func() error {
+		res := db.Exec(`
+			DELETE FROM pending_backend_ops
+			WHERE backend = ''
+			   OR node_id NOT IN (SELECT id FROM backend_nodes WHERE node_type = ? OR node_type = '')
+		`, NodeTypeBackend)
+		if res.Error != nil {
+			xlog.Warn("Failed to prune malformed pending_backend_ops rows", "error", res.Error)
+			return res.Error
+		}
+		if res.RowsAffected > 0 {
+			xlog.Info("Pruned pending_backend_ops rows (wrong node type or empty backend)", "count", res.RowsAffected)
+		}
+		return nil
+	})
+
 	return &NodeRegistry{db: db}, nil
 }
 

gallery/index.yaml

@@ -15186,10 +15186,10 @@
     - gpu
   overrides:
     parameters:
-      model: wan2.1_t2v_1.3b-q8_0.gguf
+      model: wan2.1-t2v-1.3B-Q8_0.gguf
   files:
-    - filename: "wan2.1_t2v_1.3b-q8_0.gguf"
-      uri: "huggingface://calcuis/wan-gguf/wan2.1_t2v_1.3b-q8_0.gguf"
+    - filename: "wan2.1-t2v-1.3B-Q8_0.gguf"
+      uri: "huggingface://calcuis/wan-gguf/wan2.1-t2v-1.3B-Q8_0.gguf"
     - filename: "wan_2.1_vae.safetensors"
       uri: "https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors"
     - filename: "umt5-xxl-encoder-Q8_0.gguf"

gallery/wan-ggml.yaml

@@ -9,6 +9,11 @@ config_file: |
     - "diffusion_model"
     - "vae_decode_only:false"
     - "sampler:euler"
+    - "scheduler:discrete"
     - "flow_shift:3.0"
+    - "diffusion_flash_attn:true"
+    - "offload_params_to_cpu:true"
+    - "keep_vae_on_cpu:true"
+    - "keep_clip_on_cpu:true"
     - "t5xxl_path:umt5-xxl-encoder-Q8_0.gguf"
     - "vae_path:wan_2.1_vae.safetensors"