diff --git a/composte-hermes.yml b/composte-hermes.yml index 0b16116..6185ef0 100644 --- a/composte-hermes.yml +++ b/composte-hermes.yml @@ -26,7 +26,8 @@ services: - --quantization=awq - --tensor-parallel-size=2 - --gpu-memory-utilization=0.82 - - --max-model-len=32768 + - --max-model-len=65536 + - --rope-scaling={"rope_type":"yarn","factor":2.0,"original_max_position_embeddings":32768} - --max-num-seqs=16 - --kv-cache-dtype=fp8 - --served-model-name=qwen3-14b-awq