minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[NPU]qwen3-8b and 32b md bugfix (#22687)

Browse Source
This commit is contained in:
Liwansi
2026-04-13 22:20:17 +08:00
committed by GitHub
parent 2089ac86a7
commit 8d904e50f2
1 changed files with 4 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/platforms/ascend/ascend_npu_best_practice.md
+4 -8
View File
@@ -2397,7 +2397,6 @@ LOCAL_HOST2=`hostname -I|awk -F " " '{print$2}'`
echo "${LOCAL_HOST1}" echo "${LOCAL_HOST1}"
echo "${LOCAL_HOST2}" echo "${LOCAL_HOST2}"
export HCCL_BUFFSIZE=400
export HCCL_SOCKET_IFNAME=lo export HCCL_SOCKET_IFNAME=lo
export GLOO_SOCKET_IFNAME=lo export GLOO_SOCKET_IFNAME=lo
export HCCL_OP_EXPANSION_MODE="AIV" export HCCL_OP_EXPANSION_MODE="AIV"
@@ -2410,7 +2409,7 @@ python -m sglang.launch_server --model-path $MODEL_PATH \
--max-running-requests 16 \ --max-running-requests 16 \
--disable-radix-cache \ --disable-radix-cache \
--speculative-draft-model-quantization unquant \ --speculative-draft-model-quantization unquant \
--speculative-algorithm EAGLE3 --speculative-draft-model-path xxx --speculative-draft-model-quantization unquant \ --speculative-algorithm EAGLE3 --speculative-draft-model-path xxx \
--speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4 \ --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4 \
--chunked-prefill-size -1 --max-prefill-tokens 16384 \ --chunked-prefill-size -1 --max-prefill-tokens 16384 \
--tp-size 4 --mem-fraction-static 0.843 --cuda-graph-bs 1 4 8 16 --dtype bfloat16 --tp-size 4 --mem-fraction-static 0.843 --cuda-graph-bs 1 4 8 16 --dtype bfloat16
@@ -2470,7 +2469,6 @@ LOCAL_HOST2=`hostname -I|awk -F " " '{print$2}'`
echo "${LOCAL_HOST1}" echo "${LOCAL_HOST1}"
echo "${LOCAL_HOST2}" echo "${LOCAL_HOST2}"
export HCCL_BUFFSIZE=400
export HCCL_SOCKET_IFNAME=lo export HCCL_SOCKET_IFNAME=lo
export GLOO_SOCKET_IFNAME=lo export GLOO_SOCKET_IFNAME=lo
export HCCL_OP_EXPANSION_MODE="AIV" export HCCL_OP_EXPANSION_MODE="AIV"
@@ -2483,7 +2481,7 @@ python -m sglang.launch_server --model-path $MODEL_PATH \
--max-running-requests 16 \ --max-running-requests 16 \
--disable-radix-cache \ --disable-radix-cache \
--speculative-draft-model-quantization unquant \ --speculative-draft-model-quantization unquant \
--speculative-algorithm EAGLE3 --speculative-draft-model-path xxx --speculative-draft-model-quantization unquant \ --speculative-algorithm EAGLE3 --speculative-draft-model-path xxx \
--speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4 \ --speculative-num-steps 3 --speculative-eagle-topk 1 --speculative-num-draft-tokens 4 \
--chunked-prefill-size -1 --max-prefill-tokens 16384 \ --chunked-prefill-size -1 --max-prefill-tokens 16384 \
--tp-size 4 --mem-fraction-static 0.843 --cuda-graph-bs 1 4 10 15 16 --dtype bfloat16 --tp-size 4 --mem-fraction-static 0.843 --cuda-graph-bs 1 4 10 15 16 --dtype bfloat16
@@ -2543,7 +2541,6 @@ LOCAL_HOST2=`hostname -I|awk -F " " '{print$2}'`
echo "${LOCAL_HOST1}" echo "${LOCAL_HOST1}"
echo "${LOCAL_HOST2}" echo "${LOCAL_HOST2}"
export HCCL_BUFFSIZE=400
export HCCL_SOCKET_IFNAME=lo export HCCL_SOCKET_IFNAME=lo
export GLOO_SOCKET_IFNAME=lo export GLOO_SOCKET_IFNAME=lo
export HCCL_OP_EXPANSION_MODE="AIV" export HCCL_OP_EXPANSION_MODE="AIV"
@@ -2556,7 +2553,7 @@ python -m sglang.launch_server --model-path $MODEL_PATH \
--max-running-requests 16 \ --max-running-requests 16 \
--disable-radix-cache \ --disable-radix-cache \
--speculative-draft-model-quantization unquant \ --speculative-draft-model-quantization unquant \
--speculative-algorithm EAGLE3 --speculative-draft-model-path xxx --speculative-draft-model-quantization unquant \ --speculative-algorithm EAGLE3 --speculative-draft-model-path xxx \
--speculative-num-steps 4 --speculative-eagle-topk 1 --speculative-num-draft-tokens 5 \ --speculative-num-steps 4 --speculative-eagle-topk 1 --speculative-num-draft-tokens 5 \
--chunked-prefill-size -1 --max-prefill-tokens 16384 \ --chunked-prefill-size -1 --max-prefill-tokens 16384 \
--tp-size 2 --mem-fraction-static 0.894 --cuda-graph-bs 1 2 4 6 9 10 15 16 --dtype bfloat16 --tp-size 2 --mem-fraction-static 0.894 --cuda-graph-bs 1 2 4 6 9 10 15 16 --dtype bfloat16
@@ -2616,7 +2613,6 @@ LOCAL_HOST2=`hostname -I|awk -F " " '{print$2}'`
echo "${LOCAL_HOST1}" echo "${LOCAL_HOST1}"
echo "${LOCAL_HOST2}" echo "${LOCAL_HOST2}"
export HCCL_BUFFSIZE=400
export HCCL_SOCKET_IFNAME=lo export HCCL_SOCKET_IFNAME=lo
export GLOO_SOCKET_IFNAME=lo export GLOO_SOCKET_IFNAME=lo
export HCCL_OP_EXPANSION_MODE="AIV" export HCCL_OP_EXPANSION_MODE="AIV"
@@ -2629,7 +2625,7 @@ python -m sglang.launch_server --model-path $MODEL_PATH \
--max-running-requests 16 \ --max-running-requests 16 \
--disable-radix-cache \ --disable-radix-cache \
--speculative-draft-model-quantization unquant \ --speculative-draft-model-quantization unquant \
--speculative-algorithm EAGLE3 --speculative-draft-model-path xxx --speculative-draft-model-quantization unquant \ --speculative-algorithm EAGLE3 --speculative-draft-model-path xxx \
--speculative-num-steps 4 --speculative-eagle-topk 1 --speculative-num-draft-tokens 5 \ --speculative-num-steps 4 --speculative-eagle-topk 1 --speculative-num-draft-tokens 5 \
--chunked-prefill-size -1 --max-prefill-tokens 16384 \ --chunked-prefill-size -1 --max-prefill-tokens 16384 \
--tp-size 2 --mem-fraction-static 0.894 --cuda-graph-bs 1 5 15 16 --dtype bfloat16 --tp-size 2 --mem-fraction-static 0.894 --cuda-graph-bs 1 5 15 16 --dtype bfloat16