fix: Update eagle_one configs with speculative_model_dir field (#2283)

rmccorm4 · web-flow · commit 039c061a5784 · 2025-08-05T00:45:39.000+01:00
diff --git a/components/backends/trtllm/engine_configs/llama4/eagle_one_model/eagle_agg.yml b/components/backends/trtllm/engine_configs/llama4/eagle_one_model/eagle_agg.yml
@@ -24,7 +24,7 @@ disable_overlap_scheduler: true # disable_overlap_scheduler is having acc issue
 speculative_config:
   decoding_type: Eagle
   max_draft_len: 3
-  pytorch_weights_path: nvidia/Llama-4-Maverick-17B-128E-Eagle3
+  speculative_model_dir: nvidia/Llama-4-Maverick-17B-128E-Eagle3
   eagle3_one_model: true
 
 kv_cache_config:
diff --git a/components/backends/trtllm/engine_configs/llama4/eagle_one_model/eagle_decode.yaml b/components/backends/trtllm/engine_configs/llama4/eagle_one_model/eagle_decode.yaml
@@ -26,7 +26,7 @@ disable_overlap_scheduler: true
 speculative_config:
   decoding_type: Eagle
   max_draft_len: 3
-  pytorch_weights_path: nvidia/Llama-4-Maverick-17B-128E-Eagle3
+  speculative_model_dir: nvidia/Llama-4-Maverick-17B-128E-Eagle3
   eagle3_one_model: True
 
 kv_cache_config:
diff --git a/components/backends/trtllm/engine_configs/llama4/eagle_one_model/eagle_prefill.yaml b/components/backends/trtllm/engine_configs/llama4/eagle_one_model/eagle_prefill.yaml
@@ -26,7 +26,7 @@ disable_overlap_scheduler: true
 speculative_config:
   decoding_type: Eagle
   max_draft_len: 3
-  pytorch_weights_path: nvidia/Llama-4-Maverick-17B-128E-Eagle3
+  speculative_model_dir: nvidia/Llama-4-Maverick-17B-128E-Eagle3
   eagle3_one_model: True
 
 kv_cache_config:
