Dockerfile

ARG BASE_IMAGE=vllm/vllm-openai:glm51-cu130
FROM ${BASE_IMAGE}

# Patch tool parser for GLM regex fix
COPY glm4_moe_tool_parser.py /usr/local/lib/python3.12/dist-packages/vllm/tool_parsers/glm4_moe_tool_parser.py
COPY utils.py /usr/local/lib/python3.12/dist-packages/vllm/tool_parsers/utils.py

# Patch hf renderer to force string content format for GLM models
# This fixes the issue where tool response content is dropped
COPY vllm_patches/hf.py /usr/local/lib/python3.12/dist-packages/vllm/renderers/hf.py
GLM-5.1 tool parser with incremental streaming support 2026-04-08 18:23:12 +00:00			`ARG BASE_IMAGE=vllm/vllm-openai:glm51-cu130`
			`FROM ${BASE_IMAGE}`

Add hf.py patch to force string content format for GLM models - Tool response content was being dropped because vLLM detected 'openai' content format incorrectly for GLM templates - Added _is_glm_model() detection to force 'string' format - Updated Dockerfile to include hf.py patch - Added debug tests for tool visibility 2026-04-09 05:20:47 +00:00			`# Patch tool parser for GLM regex fix`
GLM-5.1 tool parser with incremental streaming support 2026-04-08 18:23:12 +00:00			`COPY glm4_moe_tool_parser.py /usr/local/lib/python3.12/dist-packages/vllm/tool_parsers/glm4_moe_tool_parser.py`
			`COPY utils.py /usr/local/lib/python3.12/dist-packages/vllm/tool_parsers/utils.py`
Add hf.py patch to force string content format for GLM models - Tool response content was being dropped because vLLM detected 'openai' content format incorrectly for GLM templates - Added _is_glm_model() detection to force 'string' format - Updated Dockerfile to include hf.py patch - Added debug tests for tool visibility 2026-04-09 05:20:47 +00:00
			`# Patch hf renderer to force string content format for GLM models`
			`# This fixes the issue where tool response content is dropped`
			`COPY vllm_patches/hf.py /usr/local/lib/python3.12/dist-packages/vllm/renderers/hf.py`