add GLM-5 model support

- Add model cards for GLM-5 variants: MXFP4-Q8, 4bit, 8bit-MXFP8, bf16
- Add GlmMoeDsaForCausalLM to supports_tensor whitelist
- Fix EOS tokens: GLM-5 and GLM-4.7 share the same EOS tokens
  (154820, 154827, 154829), consolidate the match condition

Note: full tensor parallel support for GLM-5 requires additional
auto_parallel changes (CacheList compatibility, NullIndexer for DSA)
and upstream MLX fixes for MXFP quantized matmul.

Closes #1468

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

resources/inference_model_cards/mlx-community--GLM-5-4bit.toml

@@ -0,0 +1,12 @@
+model_id = "mlx-community/GLM-5-4bit"
+n_layers = 78
+hidden_size = 6144
+supports_tensor = true
+tasks = ["TextGeneration"]
+family = "glm"
+quantization = "4bit"
+base_model = "GLM 5"
+capabilities = ["text", "thinking"]
+
+[storage_size]
+in_bytes = 418621403136

resources/inference_model_cards/mlx-community--GLM-5-8bit-MXFP8.toml

@@ -0,0 +1,12 @@
+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 = 767273926656

resources/inference_model_cards/mlx-community--GLM-5-MXFP4-Q8.toml

@@ -0,0 +1,12 @@
+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 = 405480321024

resources/inference_model_cards/mlx-community--GLM-5.toml

@@ -0,0 +1,12 @@
+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

src/exo/master/api.py

@@ -149,12 +149,7 @@ from exo.shared.types.openai_responses import (
     ResponsesResponse,
 )
 from exo.shared.types.state import State
-from exo.shared.types.worker.instances import (
-    Instance,
-    InstanceId,
-    InstanceMeta,
-    MlxJacclInstance,
-)
+from exo.shared.types.worker.instances import Instance, InstanceId, InstanceMeta
 from exo.shared.types.worker.shards import Sharding
 from exo.utils.banner import print_startup_banner
 from exo.utils.channels import Receiver, Sender, channel
@@ -496,14 +491,6 @@ class API:
             shard_assignments = instance.shard_assignments
             placement_node_ids = list(shard_assignments.node_to_runner.keys())
 
-            # Derive instance_meta from the actual instance type, since
-            # place_instance() may override it (e.g., single-node → MlxRing)
-            actual_instance_meta = (
-                InstanceMeta.MlxJaccl
-                if isinstance(instance, MlxJacclInstance)
-                else InstanceMeta.MlxRing
-            )
-
             memory_delta_by_node: dict[str, int] = {}
             if placement_node_ids:
                 total_bytes = model_card.storage_size.in_bytes
@@ -516,14 +503,14 @@ class API:
             if (
                 model_card.model_id,
                 sharding,
-                actual_instance_meta,
+                instance_meta,
                 len(placement_node_ids),
             ) not in seen:
                 previews.append(
                     PlacementPreview(
                         model_id=model_card.model_id,
                         sharding=sharding,
-                        instance_meta=actual_instance_meta,
+                        instance_meta=instance_meta,
                         instance=instance,
                         memory_delta_by_node=memory_delta_by_node or None,
                         error=None,
@@ -533,7 +520,7 @@ class API:
                 (
                     model_card.model_id,
                     sharding,
-                    actual_instance_meta,
+                    instance_meta,
                     len(placement_node_ids),
                 )
             )

src/exo/shared/models/model_cards.py

@@ -182,6 +182,7 @@ class ConfigData(BaseModel):
     def supports_tensor(self) -> bool:
         return self.architectures in [
             ["Glm4MoeLiteForCausalLM"],
+            ["GlmMoeDsaForCausalLM"],
             ["DeepseekV32ForCausalLM"],
             ["DeepseekV3ForCausalLM"],
             ["Qwen3NextForCausalLM"],

src/exo/worker/engines/mlx/utils_mlx.py

@@ -285,7 +285,7 @@ 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-4.7-flash" in model_id_lower:
+    elif "glm-5" in model_id_lower or "glm-4.7" in model_id_lower:
         # 154820: <|endoftext|>, 154827: <|user|>, 154829: <|observation|>
         return [154820, 154827, 154829]
     elif "glm" in model_id_lower: