5 Commits
v1.0.0 ... v1.0

Author SHA1 Message Date
4paradigm
e083709837 fix tokenizer
All checks were successful
Docker Build and Push / docker (push) Successful in 1m30s
2026-06-30 16:44:45 +08:00
4paradigm
de449c2971 fix tokenizer
Some checks failed
Docker Build and Push / docker (push) Failing after 12s
2026-06-30 16:40:41 +08:00
4paradigm
b50b645cc7 fix tokenizer
Some checks failed
Docker Build and Push / docker (push) Failing after 10s
2026-06-30 16:34:40 +08:00
4paradigm
4cb1930e4a fix tokenizer
Some checks failed
Docker Build and Push / docker (push) Failing after 13s
2026-06-30 16:29:11 +08:00
4paradigm
252d761afc fix tokenizer 2026-06-30 15:35:45 +08:00
3 changed files with 19 additions and 2 deletions

View File

@@ -1,3 +1,3 @@
FRAMEWORK=vllm_tokenizer_patch
GPU_TYPE=Kunlun_P800
GPU_TYPE=Kunlunxin_p-800
TASK_TYPE=text-generation

17
vllm_wrapper.sh Normal file
View File

@@ -0,0 +1,17 @@
#!/bin/bash
set -e
# 只拦截 "serve <model_dir>" 子命令,其他子命令直接透传
if [ "$1" = "serve" ] && [ -n "$2" ]; then
MODEL_DIR="$2"
shift 2
FIXED_DIR=$(python3 /opt/fix_tokenizer.py "$MODEL_DIR")
if [ -n "$FIXED_DIR" ]; then
exec /opt/vllm_kunlun/bin/vllm_real serve "$MODEL_DIR" --tokenizer "$FIXED_DIR" "$@"
else
exec /opt/vllm_kunlun/bin/vllm_real serve "$MODEL_DIR" "$@"
fi
fi
exec /opt/vllm_kunlun/bin/vllm_real "$@"