fix: update gpt-oss image

Author: Evrard-Nil

gpt-oss-single.yaml

@@ -26,7 +26,7 @@ services:
       - --gpu-memory-utilization=0.95
       - --max-num-seqs=32
       - --max-num-batched-tokens=8192
-      - --max-model-len=32768
+      - --max-model-len=131072
       - --no-enable-prefix-caching
       - --speculative-config={"method":"eagle3","model":"nvidia/gpt-oss-120b-Eagle3-v2","num_speculative_tokens":3}
       - --async-scheduling

small-models.yaml

@@ -62,7 +62,7 @@ x-sglang-env:
 
 x-gpt-oss-common: &gpt-oss-common
   <<: *vllm-common
-  image: vllm/vllm-openai@sha256:014a95f21c9edf6abe0aea6b07353f96baa4ec291c427bb1176dc7c93a85845c
+  image: vllm/vllm-openai@sha256:6766ce0c459e24b76f3e9ba14ffc0442131ef4248c904efdcbf0d89e38be01fe
   command: >
       openai/gpt-oss-120b
       --tensor-parallel-size 1
@@ -73,14 +73,19 @@ x-gpt-oss-common: &gpt-oss-common
       --tool-call-parser openai
       --enable-auto-tool-choice
       --max-model-len 128K
-      --max-num-batched-tokens 16K
+      --max-num-batched-tokens 8192
       --speculative-config '{"model":"nvidia/gpt-oss-120b-Eagle3-v2","num_speculative_tokens":3,"method":"eagle3","draft_tensor_parallel_size":1}'
       --load-format runai_streamer
       --model-loader-extra-config '{"distributed":true, "concurrency":48}'
   volumes:
     - hugginface_cache:/root/.cache/huggingface
     - vllm_cache:/root/.cache/vllm
-  environment: *vllm-env
+  environment:
+    - HUGGING_FACE_HUB_TOKEN=${HUGGING_FACE_HUB_TOKEN}
+    - VLLM_LOGGING_LEVEL=INFO
+    - NVIDIA_DRIVER_CAPABILITIES=compute,utility
+    - VLLM_MXFP4_USE_MARLIN=1
+    - VLLM_CACHE_ROOT=/root/.cache/vllm
 
 x-flux-common: &flux-common
   <<: *vllm-common