添加模型服务 Dockerfile 与启动脚本

Dockerfile

@@ -0,0 +1,25 @@
+FROM registry.maas.sunrise-ai.com/public/vllm:S2-v1.1.1
+ENV LD_LIBRARY_PATH=/usr/local/pccl/lib:\
+/usr/local/tangrt/targets/linux-x86_64/lib:\
+/usr/local/tangrt/targets/linux-x86_64/lib/stub:\
+/root/pt200/gcc-11.3.0/install/lib64:\
+/root:/root/gcc-11.5.0/lib64:\
+/usr/local/pccl/lib:\
+/usr/local/tangrt/targets/linux-x86_64/lib:\
+/usr/local/tangrt/targets/linux-x86_64/lib/stub:\
+/usr/local/tangrt/lib/linux-x86_64:\
+/root/pt200/gcc-11.3.0/install/lib64:\
+/root:\
+/usr/lib64:\
+/usr/local/lib/python3.10/site-packages/torch/lib
+ENV TORCH_DEVICE_BACKEND_AUTOLOAD=0
+ENV PATH=/root/gcc-11.5.0/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
+ENV PYTHONPATH=/sunrise_code/vllm:/sunrise_code/sunrise_vllm:/usr/local/lib/python3.10/site-packages:
+RUN ln -sf /usr/local/bin/python3.10 /usr/bin/python3
+COPY fix_tokenizer.py /opt/
+COPY detect_tokenizer.py /opt/
+COPY entrypoint.sh /opt/
+WORKDIR /model
+COPY . /model
+RUN chmod +x /opt/entrypoint.sh
+ENTRYPOINT ["/opt/entrypoint.sh"]

detect_tokenizer.py

@@ -0,0 +1,18 @@
+#!/usr/bin/env python3
+# 检测 tokenizer 是否需要修复
+import json
+import os
+
+MODEL_DIR = os.environ.get("MODEL_DIR", "/model")
+cfg_path = os.path.join(MODEL_DIR, "tokenizer_config.json")
+if not os.path.exists(cfg_path):
+    print("no tokenizer_config.json")
+    raise SystemExit(0)
+with open(cfg_path, "r", encoding="utf-8") as f:
+    cfg = json.load(f)
+if cfg.get("tokenizer_class") in {"TokenizersBackend", "TiktokenTokenizer"}:
+    print("need_fix")
+elif isinstance(cfg.get("extra_special_tokens"), list):
+    print("need_fix")
+else:
+    print("ok")

entrypoint.sh

@@ -0,0 +1,38 @@
+#!/bin/bash
+set -e
+
+MODEL_DIR=${1:-/model}
+shift || true
+
+FIX_TOKENIZER_DIR=/tmp/fixed_tokenizer
+AUTO_FIX=${AUTO_FIX_TOKENIZER:-auto}
+
+echo "[entrypoint] model dir: $MODEL_DIR"
+
+NEED_FIX=0
+
+if [ "$AUTO_FIX" = "1" ] || [ "$AUTO_FIX" = "true" ]; then
+    NEED_FIX=1
+elif [ "$AUTO_FIX" = "auto" ]; then
+    if [ -f "$MODEL_DIR/tokenizer_config.json" ]; then
+        if grep -q "TokenizersBackend\|TiktokenTokenizer" "$MODEL_DIR/tokenizer_config.json"; then
+            NEED_FIX=1
+        fi
+        if grep -q '"extra_special_tokens":\s*\[' "$MODEL_DIR/tokenizer_config.json"; then
+            NEED_FIX=1
+        fi
+    fi
+fi
+
+if [ $NEED_FIX -eq 1 ]; then
+    echo "[entrypoint] fixing tokenizer..."
+    python3 /opt/fix_tokenizer.py
+    TOKENIZER_ARG="--tokenizer $FIX_TOKENIZER_DIR"
+else
+    echo "[entrypoint] tokenizer OK, skip fix"
+    TOKENIZER_ARG=""
+fi
+
+echo "[entrypoint] starting vllm..."
+
+exec vllm serve "$MODEL_DIR" $TOKENIZER_ARG "$@"

fix_tokenizer.py

@@ -0,0 +1,32 @@
+#!/usr/bin/env python3
+# 自动修复 tokenizer 配置,供 entrypoint.sh 调用
+import json
+import os
+import shutil
+
+MODEL_DIR = os.environ.get("MODEL_DIR", "/model")
+FIX_DIR = "/tmp/fixed_tokenizer"
+
+os.makedirs(FIX_DIR, exist_ok=True)
+for name in ["tokenizer.json", "tokenizer_config.json", "special_tokens_map.json", "vocab.json", "merges.txt", "tokenizer.model"]:
+    src = os.path.join(MODEL_DIR, name)
+    if os.path.exists(src):
+        shutil.copy(src, os.path.join(FIX_DIR, name))
+
+cfg_path = os.path.join(FIX_DIR, "tokenizer_config.json")
+if os.path.exists(cfg_path):
+    with open(cfg_path, "r", encoding="utf-8") as f:
+        cfg = json.load(f)
+    changed = False
+    bad = {"TokenizersBackend", "TiktokenTokenizer"}
+    if cfg.get("tokenizer_class") in bad:
+        cfg["tokenizer_class"] = "PreTrainedTokenizerFast"
+        changed = True
+    extra = cfg.get("extra_special_tokens")
+    if isinstance(extra, list):
+        cfg["extra_special_tokens"] = {t: t for t in extra}
+        changed = True
+    if changed:
+        with open(cfg_path, "w", encoding="utf-8") as f:
+            json.dump(cfg, f, ensure_ascii=False, indent=2)
+        print("[fix] tokenizer_config.json 已修复")