metadata
base_model:
- THUDM/GLM-4.1V-9B-Thinking
pipeline_tag: image-text-to-text
library_name: transformers
GLM‑4.1V‑9B‑Thinking • Quantized
🚀 Model Description
This is a quantized version of GLM‑4.1V‑9B‑Thinking, a powerful 9B‑parameter vision‑language model using the “thinking paradigm” and reinforced reasoning. The quantization enables significantly lighter memory usage and faster inference on consumer-grade GPUs while preserving its strong performance on multimodal reasoning tasks.
Quantization Details
Method: torchao quantization Weight Precision: int8 Activation Precision: int8 dynamic Technique: Symmetric mapping Impact: Significant reduction in model size with minimal loss in reasoning, coding, and general instruction-following capabilities.
🎯 Intended Use
Perfect for:
- Vision‑language applications with long contexts and heavy reasoning
- On-device or low-VRAM inference for tempo‑sensitive environments
- Challenging multimodal tasks: image Q&A, reasoning over diagrams, high-resolution visual analysis
- Research into quantized vision‑language deployment
⚠️ Limitations
- Minor drop in detailed reasoning accuracy vs full-precision
- Maintains original model’s general LLM caveats: hallucinations, bias, and prompting sensitivity