fix tokenizer

.gitea/workflows/docker-build-push.yml

@@ -129,4 +129,4 @@ jobs:
               \"createBy\": \"${{ gitea.actor }}\",
               \"repoUrl\": \"${{ gitea.server_url }}/${{ gitea.repository }}\",
               \"tag\": \"${{ github.ref_name }}\"
-            }"
+            }"

.gitea/workflows/task_info.env

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

vllm_wrapper.sh

@@ -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" "$@"
+    fi
+fi
+
+exec /opt/vllm_kunlun/bin/vllm_real "$@"