GUI-Owl-1.5-8B-Think-FP8-Dynamic

FP8 dynamic quantization of mPLUG/GUI-Owl-1.5-8B-Think, produced with llmcompressor and saved in compressed format.

Quantization

• Method: FP8_DYNAMIC
• Quantization mode: data-free (no calibration dataset)
• Target modules: Linear
• Ignored modules: lm_head, visual tower
• MAX_SEQUENCE_LENGTH: 32768
• Reference recipe: llm-compressor Qwen2.5-VL FP8 example

Evaluation

Evaluation results coming soon.