Lower context to 32k, fix duplicate lines, remove prefix caching

2026-08-08 17:03:16 -07:00
parent 9227e1a8b9
commit 9d00f36d2b
1 changed files with 1 additions and 2 deletions
@@ -24,12 +24,11 @@ exec docker run --rm --name vllm-awq-qwen36-dflash-ampere \
    --served-model-name "frank-default-small" \
    --speculative-config '{"method": "dflash", "model": "z-lab/Qwen3.6-35B-A3B-DFlash", "num_speculative_tokens": 4}' \
    --max-num-seqs 2 \
-    --max-model-len 262144 \
+    --max-model-len 32768 \
    --tensor-parallel-size 1 \
    --enable-chunked-prefill \
    --max-num-batched-tokens 1536 \
    --performance-mode throughput \
-    # --enable-prefix-caching  # Disabled: Mamba cache align mode conflicts with prefix caching \\
    --enable-auto-tool-choice --tool-call-parser qwen3_xml \
    --reasoning-parser qwen3 \
    --trust-remote-code \