From 7a46d61e502a9044d3447cf272f27e49e7ea655d Mon Sep 17 00:00:00 2001 From: Marc Blatter Date: Sun, 26 Jul 2026 13:54:06 +0000 Subject: [PATCH] composte-hermes.yml aktualisiert --- composte-hermes.yml | 26 ++++++++++++++------------ 1 file changed, 14 insertions(+), 12 deletions(-) diff --git a/composte-hermes.yml b/composte-hermes.yml index 15a13b8..7b6ce11 100644 --- a/composte-hermes.yml +++ b/composte-hermes.yml @@ -11,7 +11,6 @@ services: - NVIDIA_VISIBLE_DEVICES=0,1 - HUGGING_FACE_HUB_TOKEN=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 - PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True - - VLLM_ALLOW_LONG_MAX_MODEL_LEN=1 volumes: - /opt/vllm/huggingface:/root/.cache/huggingface deploy: @@ -22,17 +21,19 @@ services: count: 2 capabilities: [gpu] command: - - --model=Qwen/Qwen3-14B-AWQ - - --tensor-parallel-size=2 - - --gpu-memory-utilization=0.92 - - --max-model-len=65536 - - --hf-overrides={"rope_scaling":{"rope_type":"yarn","factor":2.0,"original_max_position_embeddings":40960}} - - --enable-prefix-caching - - --enable-auto-tool-choice - - --tool-call-parser=hermes - - --max-num-seqs=4 - - --generation-config=vllm - - --disable-custom-all-reduce + - --model=Qwen/Qwen3-14B-AWQ + - --port=8005 + - --quantization=awq + - --tensor-parallel-size=2 + - --gpu-memory-utilization=0.82 + - --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 + - --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 @@ -57,6 +58,7 @@ services: volumes: - /opt/hermes:/opt/data - /opt/hermes/.ssh:/root/.ssh + - /var/run/docker.sock:/var/run/docker.sock environment: - OPENAI_BASE_URL=http://vllm:8005/v1