Neue Korrektur mit ChatGPT

composte-hermes.yml

@@ -25,15 +25,13 @@ services:
        - --port=8005
        - --quantization=awq
        - --tensor-parallel-size=2
-      - --gpu-memory-utilization=0.82
+       - --gpu-memory-utilization=0.84
        - --max-model-len=65536
        - --hf-overrides={"rope_scaling":{"rope_type":"yarn","factor":2.0,"original_max_position_embeddings":32768}}
-      - --max-num-seqs=16
-      - --kv-cache-dtype=fp8
+       - --max-num-seqs=4
        - --served-model-name=qwen3-14b-awq
        - --enable-auto-tool-choice
        - --tool-call-parser=hermes
-      - --reasoning-parser=deepseek_r1
     healthcheck:
       test: ["CMD", "curl", "-f", "http://localhost:8005/health"]
       interval: 30s