---
title: Kimi-K2.6-w8a8-xllm
canonical_url: "https://www.modelscope.cn/models/Eco-Tech/Kimi-K2.6-w8a8-xllm"
md_url: "https://www.modelscope.cn/models/Eco-Tech/Kimi-K2.6-w8a8-xllm.md"
repository: Eco-Tech/Kimi-K2.6-w8a8-xllm
last_updated: 2026-09-04
license: apache-2.0
pipeline_tag: image-text-to-text
tasks:
  - image-text-to-text
model_type:
  - kimi_k25
architectures:
  - KimiK25ForConditionalGeneration
parameters: 1027.4B
tensor_type:
  - F32
  - I8
  - BF16
  - I32
library_name:
  - safetensors
  - other
frameworks:
  - other
downloads: 2008
stars: 0
---

# Kimi-K2.6-w8a8-xllm

> Kimi-K2.6-w8a8-xllm - Eco-Tech 在 ModelScope 开源的模型。1. 基本信息 | 项目 | 信息 | |:------:|:------:| | 原始模型名 | Kimi-K2.6 | | 原始模型链接 | moonshotai/Kimi-K2.6 | | msmodelslim commit id | 5519faf40801f47487b273486372058ce22e6036 |

Eco-Tech/Kimi-K2.6-w8a8-xllm 是 ModelScope 魔搭社区上的 1027.4B 参数image-text-to-text模型，采用 apache-2.0 许可。

- **Repository**: Eco-Tech/Kimi-K2.6-w8a8-xllm
- **License**: apache-2.0
- **Tasks**: image-text-to-text
- **Parameters**: 1027.4B
- **Downloads**: 2008
- **Stars**: 0
- **Last updated**: 2026-09-04

Source: https://www.modelscope.cn/models/Eco-Tech/Kimi-K2.6-w8a8-xllm

---

# Kimi-K2.6-w8a8-xllm

## 1. 基本信息
| 项目 | 信息 |
|:------:|:------:|
| 原始模型名  | Kimi-K2.6 |
| 原始模型链接  | [ moonshotai/Kimi-K2.6 ](https://www.modelscope.cn/models/moonshotai/Kimi-K2.6) |
| msmodelslim commit id | 5519faf40801f47487b273486372058ce22e6036 |


## 2 量化脚本：
```shell
msmodelslim quant --model_path  ${model_path} --save_path ${save_path} \
  --model_type Kimi-K2.5 --config_path ./kimi_w8a8.yaml  --trust_remote_code True
```

## 3 量化yaml文件
```yaml
apiversion: multimodal_vlm_modelslim_v1
metadata:
  config_id: kimi_w8a8_quarot
  score: 90
  verified_model_types:
    - Kimi-K2.5
  label:
    w_bit: 8
    a_bit: 8
    is_sparse: False
    kv_cache: False

default_w4a8_dynamic: &default_w4a8_dynamic
    act:
      scope: "per_token"
      dtype: "int8"
      symmetric: True
      method: "minmax"
    weight:
      scope: "per_channel"
      dtype: "int4"
      symmetric: True
      method: "ssz"

default_w8a8_dynamic: &default_w8a8_dynamic
  act:
    scope: "per_token"
    dtype: "int8"
    symmetric: True
    method: "minmax"
  weight:
    scope: "per_channel"
    dtype: "int8"
    symmetric: True
    method: "minmax"

default_w8a8: &default_w8a8
  act:
    scope: "per_tensor"
    dtype: "int8"
    symmetric: False
    method: "minmax"
  weight:
    scope: "per_channel"
    dtype: "int8"
    symmetric: True
    method: "minmax"

spec:
  process:
    - type: "group"
      configs:
        - type: "linear_quant"
          qconfig: *default_w8a8
          include:
            - "*self_attn*"
          exclude:
            - "*kv_b_proj"
        - type: "linear_quant"
          qconfig: *default_w8a8_dynamic
          include:
            - "*mlp*"
          exclude:
            - "*gate"
            - "*mlp.experts.*"
        - type: "linear_quant"
          qconfig: *default_w8a8_dynamic
          include:
            - "model.layers.61.mlp.experts.*"
        - type: "linear_quant"
          qconfig: *default_w8a8_dynamic
          include:
            - "*mlp.experts.*"
          exclude:
            - "model.layers.61.*"
  dataset: calibImages
  save:
    - type: "ascendv1_saver"
      part_file_size: 4
```