---
title: RakutenAI-3.0-MLX-4bit
canonical_url: "https://www.modelscope.cn/models/mlx-community/RakutenAI-3.0-MLX-4bit"
md_url: "https://www.modelscope.cn/models/mlx-community/RakutenAI-3.0-MLX-4bit.md"
repository: mlx-community/RakutenAI-3.0-MLX-4bit
last_updated: 2026-04-04
license: apache-2.0
pipeline_tag: text-generation
tasks:
  - text-generation
model_type:
  - deepseek_v3
architectures:
  - DeepseekV3ForCausalLM
base_model:
  - Rakuten/RakutenAI-3.0
base_model_relation: quantized
parameters: 104.9B
tensor_type:
  - F32
  - BF16
  - U32
library_name:
  - mlx
  - safetensors
  - pytorch
frameworks:
  - pytorch
language:
  - en
  - ja
inference_backends:
  - "deploy_task text/emb"
  - "sglang 0.5.2"
  - "vllm 0.9.2"
downloads: 11
stars: 0
tags:
  - mlx
---

# RakutenAI-3.0-MLX-4bit

> RakutenAI-3.0-MLX-4bit - mlx-community 在 ModelScope 开源的模型。This model was converted to MLX format from Rakuten/RakutenAI-3.0 using mlx-lm version 0.31.0. Refer to the original model card for more details on the model.

mlx-community/RakutenAI-3.0-MLX-4bit 是 ModelScope 魔搭社区上的 104.9B 参数text-generation模型，采用 apache-2.0 许可，基于 Rakuten/RakutenAI-3.0 构建，可用 deploy_task text/emb、sglang 0.5.2、vllm 0.9.2 部署。

- **Repository**: mlx-community/RakutenAI-3.0-MLX-4bit
- **License**: apache-2.0
- **Tasks**: text-generation
- **Parameters**: 104.9B
- **Base model**: Rakuten/RakutenAI-3.0
- **Inference backends**: deploy_task text/emb, sglang 0.5.2, vllm 0.9.2
- **Tags**: mlx
- **Downloads**: 11
- **Stars**: 0
- **Last updated**: 2026-04-04

Source: https://www.modelscope.cn/models/mlx-community/RakutenAI-3.0-MLX-4bit

---

This model was converted to MLX format from [`Rakuten/RakutenAI-3.0`](https://huggingface.co/Rakuten/RakutenAI-3.0) using mlx-lm version **0.31.0**.
Refer to the [original model card](https://huggingface.co/Rakuten/RakutenAI-3.0) for more details on the model.

Convert:❯ mlx_lm.convert --hf-path ./RakutenAI-3.0 --mlx-path ./lm-studio/models/limegreenpeper1/rakutenai3-mlx-4bit  -q --q-bits 4 --trust-remote-code

[INFO] Loading
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
[INFO] Using dtype: bfloat16
[INFO] Quantizing
[INFO] Quantized model with 4.502 bits per weight.


Inference: 20.06 tok/sec M3 Ultra (CTX 128K - LM Studio MLX v1.3.0)