GLM-4.5 is a Python-based foundation model family for agentic, reasoning, and coding tasks with multiple variants and tooling support. It includes deployment guidance, quick start instructions, and system requirements for FP8/BF16 on Nvidia GPUs.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
GLM-4.5 is a family of foundation models focused on agentic, reasoning, and coding capabilities, with GLM-4.5, GLM-4.5-Air, GLM-4.6, GLM-4.7 variants. The README describes model downloads, capabilities, and accompanying tooling support (sglang, vLLM, SGLang).
How it works
The project provides code references for model implementations and parsers across transformers, vLLM, and SGLang. Tool usage and reasoning components are designed to work with agentic tasks and interleaved/preserved/turn-level thinking modes. Specific model code references include: transformers glm4_moe models, vLLM glm4_moe_mtp.py, and sglang glm4_moe.py implementations.
Getting started
Quick Start
Install dependencies (sglang, vllm, etc.) according to the configuration requirements in requirements.txt.
transformers
Please refer to the trans_infer_cli.py code in the inference folder.
vLLM
vllm serve zai-org/GLM-4.7-FP8 \
--tensor-parallel-size 4 \
--speculative-config.method mtp \
--speculative-config.num_speculative_tokens 1 \
--tool-call-parser glm47 \
--reasoning-parser glm45 \
--enable-auto-tool-choice \
--served-model-name glm-4.7-fp8
SGLang
python3 -m sglang.launch_server \
--model-path zai-org/GLM-4.7-FP8 \
--tp-size 8 \
--tool-call-parser glm47 \
--reasoning-parser glm45 \
--speculative-algorithm EAGLE \
--speculative-num-steps 3 \
--speculative-eagle-topk 1 \
--speculative-num-draft-tokens 4 \
--mem-fraction-static 0.8 \
--served-model-name glm-4.7-fp8 \
--host 0.0.0.0 \
--port 8000
- The README also provides example commands for PD-Disaggregation and further parameter instructions for thinking modes.
Recent releases
RELEASES (latest 0): - none
Traction
Stars: 4410
Behind the repo
There is no separate linked startup/company section in the provided facts.
Caveats
License: Apache-2.0. Created: 2025-07-20. Last push: 2026-02-01. Open issues: 27. Language: Python. Models are hosted with multiple download options and FP8/BF16 configurations; system requirements specify Nvidia GPUs and memory constraints. The README references external resources and deployment guides, but no specific licensing changes beyond Apache-2.0 are noted here.