From 64703e0d2c054862e3eccbd9b11cfc307f9e9ede Mon Sep 17 00:00:00 2001
From: wanglin <2281216234@qq.com>
Date: Sat, 3 Oct 2026 19:21:57 +0800
Subject: [PATCH] =?UTF-8?q?enginex-bi100-compat:=20=E6=94=B9=E4=B8=BA?=
 =?UTF-8?q?=E6=8B=A6=E6=88=AA=20vllm=20=E4=BA=8C=E8=BF=9B=E5=88=B6?=
 =?UTF-8?q?=EF=BC=88=E5=AF=B9=E9=BD=90=20zhengzhongwei/vLLM-Iluvatar=5Fbi-?=
 =?UTF-8?q?100-opt=20=E6=A8=A1=E5=BC=8F=EF=BC=89=EF=BC=8C=E5=B9=B3?=
 =?UTF-8?q?=E5=8F=B0=E5=91=BD=E4=BB=A4=E6=97=A0=E9=9C=80=E6=94=B9=E5=8A=A8?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 Dockerfile | 35 +++++++++++++++++++----------------
 1 file changed, 19 insertions(+), 16 deletions(-)

diff --git a/Dockerfile b/Dockerfile
index 1c3c4b6..99ddd88 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -2,9 +2,10 @@
 #
 # 天数智芯 天垓100（Iluvatar_bi-100）· 文本生成 · vLLM「兼容增强」引擎。
 #
-# 不动模型、不动卡，只把「引擎镜像层」的结构性不兼容在启动前修掉。
-# 结构与社区已上线引擎 EngineX-Sunrise/enginex-S2-vllm-fix-tokenizer 完全同型
-# （Dockerfile + entrypoint.sh + 修补脚本 + README），该模式已在曦望 S2 生产验证。
+# 不动模型、不动卡，只把「引擎镜像层」的结构性不兼容在容器启动前修掉。
+# 做法照搬社区已上线引擎 zhengzhongwei/vLLM-Iluvatar_bi-100-opt：
+# **把镜像内的 vllm 二进制换成本仓的 wrapper**，平台原样下发
+# `vllm serve /model --port 80 ...` 就会流经 wrapper，无需平台改任何命令。
 #
 # 修什么（全部有日志实证，见 README.md）：
 #   R3  extra_special_tokens 是 list            -> transformers 崩（社区已上线修复，已合并）
@@ -13,33 +14,35 @@
 #   R4  architectures 是镜像未注册类名          -> KeyError / MODEL_NOT_SUPPORTED
 #   R7  镜像缺 ixformer.contrib.vllm.layers     -> ModuleNotFoundError（MoE）
 #
-# 为什么选 bi-100：本人 37 个验证失败里，bi-100 单卡 10 例为全平台最集中；
-# 且该卡社区通过率约 51%（15 张卡最低），修好收益最大。
-#
-# 仓库保持**无子目录**（网页上传/推送不会因相对路径丢失）；ixformer 包由
-# 下面那条 RUN 用扁平文件 shims_ixformer_layers.py 组装出来。
+# 基线镜像与 zhengzhongwei/vLLM-Iluvatar_bi-100-opt 一致（平台 bi-100 vllm 框架在用的那个），
+# 保证替换 vllm 二进制的路径与行为可预期。
 
-FROM git.modelhub.org.cn:9443/enginex-iluvatar/bi100-3.2.3-x86-ubuntu20.04-py3.10-poc-llm-infer:v1.2.3
+FROM harbor-contest.4pd.io/luopingyi/bi100/vllm:0.6.3
 
 LABEL org.opencontainers.image.title="enginex-bi100-compat" \
-      org.opencontainers.image.description="Iluvatar bi-100 vLLM text-generation engine with preflight compatibility patches (R2/R3/R3b/R4/R7)" \
+      org.opencontainers.image.description="Iluvatar bi-100 vLLM engine with preflight compatibility patches (R2/R3/R3b/R4/R7)" \
       com.modelhubxc.engine.target-card="Iluvatar_bi-100" \
       com.modelhubxc.engine.framework="vllm" \
       com.modelhubxc.engine.task-type="text-generation" \
-      com.modelhubxc.engine.baseline="EngineX-Iluvatar/enginex-vllm-bi100-qwen36" \
-      com.modelhubxc.engine.pattern="EngineX-Sunrise/enginex-S2-vllm-fix-tokenizer"
+      com.modelhubxc.engine.pattern="zhengzhongwei/vLLM-Iluvatar_bi-100-opt"
 
 # 修补脚本 + R7 纯 PyTorch shim（都是小文本文件，无权重）
 COPY preflight.py /opt/
-COPY entrypoint.sh /opt/
 COPY detect_tokenizer.py /opt/
+COPY vllm_wrapper.sh /opt/
 COPY shims_ixformer_layers.py /opt/shims_src/
 
 # 组装 ixformer.contrib.vllm.layers 包（R7 兜底，只有真缺模块时才会被 import）
-RUN set -eux && mkdir -p /opt/shims/ixformer/contrib/vllm/layers && for d in /opt/shims/ixformer /opt/shims/ixformer/contrib /opt/shims/ixformer/contrib/vllm /opt/shims/ixformer/contrib/vllm/layers; do printf '# ModelHub XC compat shim package\n' > "$d/__init__.py"; done && cp /opt/shims_src/shims_ixformer_layers.py /opt/shims/ixformer/contrib/vllm/layers/__init__.py && rm -rf /opt/shims_src && chmod +x /opt/entrypoint.sh /opt/preflight.py && python3 -c "import ast,io;[ast.parse(io.open(f,encoding='utf-8').read()) for f in ['/opt/preflight.py','/opt/detect_tokenizer.py','/opt/shims/ixformer/contrib/vllm/layers/__init__.py']]" && bash -n /opt/entrypoint.sh && echo "[enginex-bi100-compat] preflight syntax OK"
+RUN set -eux && mkdir -p /opt/shims/ixformer/contrib/vllm/layers && for d in /opt/shims/ixformer /opt/shims/ixformer/contrib /opt/shims/ixformer/contrib/vllm /opt/shims/ixformer/contrib/vllm/layers; do printf '# ModelHub XC compat shim package\n' > "$d/__init__.py"; done && cp /opt/shims_src/shims_ixformer_layers.py /opt/shims/ixformer/contrib/vllm/layers/__init__.py && rm -rf /opt/shims_src && chmod +x /opt/vllm_wrapper.sh /opt/preflight.py && python3 -c "import ast,io;[ast.parse(io.open(f,encoding='utf-8').read()) for f in ['/opt/preflight.py','/opt/detect_tokenizer.py','/opt/shims/ixformer/contrib/vllm/layers/__init__.py']]" && bash -n /opt/vllm_wrapper.sh && echo "[enginex-bi100-compat] preflight syntax OK"
 
-# R7：让 shims 全局可 import（平台若覆盖 entrypoint 也仍然生效）
+# R7：让 shims 全局可 import
 ENV PYTHONPATH=/opt/shims:${PYTHONPATH}
 ENV MODEL_DIR=/model
 
-ENTRYPOINT ["/opt/entrypoint.sh"]
+# 关键一步：把 vllm 二进制换成 wrapper（真身改名 vllm_real），
+# 平台下发的 `vllm serve ...` 因此流经本引擎的启动前修补。
+RUN set -eux && VLLM_BIN="$(command -v vllm || echo /usr/local/corex/lib64/python3/dist-packages/bin/vllm)" && \
+    if [ ! -f "$VLLM_BIN" ]; then echo "vllm binary not found at $VLLM_BIN"; exit 1; fi && \
+    mv "$VLLM_BIN" "${VLLM_BIN}_real" && \
+    ln -s /opt/vllm_wrapper.sh "$VLLM_BIN" && \
+    echo "[enginex-bi100-compat] vllm -> wrapper installed"