---
base_model: Kwaipilot/KAT-Coder-V2.5-Dev
base_model_relation: quantized
license: apache-2.0
pipeline_tag: text-generation
library_name: transformers
language:
  - en
  - zh
tags:
  - code
  - agent
  - agentic-coding
  - moe
  - coding
  - quantized
---

# Quantization of KAT-Coder-V2.5-Dev to FP8 Dynamic

Quantized using llm-compressor.

```
recipe = QuantizationModifier(
    targets="Linear",
    scheme="FP8_DYNAMIC",
     ignore=[
        "re:.*lm_head",
        "re:model.visual.*",
        "re:.*mlp.gate$",
        "re:.*embed_tokens$",
        "re:.*shared_expert_gate$",
      ],
)
```