update

2026-07-08 16:40:51 +02:00
parent 8a8b1db65e
commit 8ddf750b4b
2 changed files with 4 additions and 4 deletions
@@ -4,8 +4,8 @@ MODEL_PATH=/models/deepseek-coder-33b-gptq
 SERVED_NAME=deepseek-coder
 TP_SIZE=4
 DTYPE=float16
-MAX_MODEL_LEN=8192
-GPU_MEM_UTIL=0.88
+MAX_MODEL_LEN=65536
+GPU_MEM_UTIL=0.60 # 0.88
 MAX_NUM_SEQS=4
 VLLM_EXTRA_ARGS=--quantization gptq --disable-custom-all-reduce --enforce-eager