Architecture
data/models.yaml. Every label is auditable
against the model's sources.
Specs
- Architecture
- dense
- Total params
- 9B
- Active params
- 9B
- Context window
- 131K tokens
- Attention
- gqa
- Position encoding
- rope
- Post-training
- sft, rlhf
- OSI-approved
- no
- Data released
- no
- Training code
- not released
Benchmarks
Each score carries the date it was published; we never infer or interpolate missing scores.
General reasoning
| MMLU | 72.4 | as of 2024-06-05 | source ↗ |
Code
| HumanEval | 71.8 | as of 2024-06-05 | source ↗ |
Math
| MATH | 50.6 | as of 2024-06-05 | source ↗ |
Held-out / arena
| IFEval | 69.0 | as of 2024-06-05 | source ↗ |
Recommended use cases
- long-context summarization
- agentic tool use
- Chinese-English bilingual chat
Available quantizations
GGUF llama.cpp's container; the common local format, k-quants from Q2 to Q8.
runs on llama.cpp, Ollama
Verified via the Hugging Face model tree ↗. Community quantizations change over time; the families shown are those with published weights at audit time.
Notable innovations
- · 128K context at 9B scale
- · Native tool use and function calling
- · 26-language coverage
Known limitations
- · GLM-4 Model License is source-available, not OSI-approved. source ↗
Lineage
Descended from the GLM-130B and ChatGLM lines at Tsinghua's KEG lab; the 1M-context sibling extends GLM-4-9B-Chat-1M.