---
title: VibeThinker-1.5B-mlx-4bit
canonical_url: "https://www.modelscope.cn/models/mlx-community/VibeThinker-1.5B-mlx-4bit"
md_url: "https://www.modelscope.cn/models/mlx-community/VibeThinker-1.5B-mlx-4bit.md"
repository: mlx-community/VibeThinker-1.5B-mlx-4bit
last_updated: 2025-11-19
license: mit
pipeline_tag: text-generation
tasks:
- text-generation
model_type:
- qwen2
architectures:
- Qwen2ForCausalLM
base_model:
- WeiboAI/VibeThinker-1.5B
base_model_relation: quantized
library_name:
- mlx
- safetensors
- pytorch
frameworks:
- pytorch
language:
- en
inference_backends:
- "deploy_task text/emb"
- "lmdeploy 0.9.1"
- "lmdeploy_turbomind 0.9.1"
- "sglang 0.5.2"
- "vllm 0.9.2"
downloads: 52
stars: 0
tags:
- math
- code
- gpqa
- mlx
---
# VibeThinker-1.5B-mlx-4bit
> VibeThinker-1.5B-mlx-4bit - mlx-community 在 ModelScope 开源的模型。VibeThinker-1.5B-mlx-4bit
mlx-community/VibeThinker-1.5B-mlx-4bit 是 ModelScope 魔搭社区上的text-generation模型，采用 mit 许可，基于 WeiboAI/VibeThinker-1.5B 构建，可用 deploy_task text/emb、lmdeploy 0.9.1、lmdeploy_turbomind 0.9.1 部署。
- **Repository**: mlx-community/VibeThinker-1.5B-mlx-4bit
- **License**: mit
- **Tasks**: text-generation
- **Base model**: WeiboAI/VibeThinker-1.5B
- **Inference backends**: deploy_task text/emb, lmdeploy 0.9.1, lmdeploy_turbomind 0.9.1, sglang 0.5.2, vllm 0.9.2
- **Tags**: math, code, gpqa, mlx
- **Downloads**: 52
- **Stars**: 0
- **Last updated**: 2025-11-19
Source: https://www.modelscope.cn/models/mlx-community/VibeThinker-1.5B-mlx-4bit
---
# VibeThinker-1.5B-mlx-4bit
This is a 4-bit quantized version of the [VibeThinker-1.5B](https://huggingface.co/WeiboAI/VibeThinker-1.5B) model, optimized for Apple Silicon using the [MLX framework](https://github.com/ml-explore/mlx).
## Model Details
- **Original Model**: [WeiboAI/VibeThinker-1.5B](https://huggingface.co/WeiboAI/VibeThinker-1.5B)
- **Quantization**: 4-bit
- **Framework**: MLX
## Installation
To use this model, you need to install the `mlx` and `mlx-lm` packages:
```bash
pip install mlx mlx-lm
```
## Usage
You can generate text using the `mlx_lm.generate` command:
```bash
mlx_lm.generate --model mlx-community/VibeThinker-1.5B-mlx-4bit --prompt "A café sells 6 types of drinks. A customer orders exactly 8 drinks (types may repeat), but no drink type may be chosen more than 3 times. How many distinct combinations of 8 drinks can the customer order? Give your answer as an integer." --temp 0.6 --top-p 0.95 --top-k -1 --max-tokens 40960
```
Or using Python:
```python
from mlx_lm import load, generate
model, tokenizer = load("mlx-community/VibeThinker-1.5B-mlx-4bit")
response = generate(model, tokenizer, prompt="A café sells 6 types of drinks. A customer orders exactly 8 drinks (types may repeat), but no drink type may be chosen more than 3 times. How many distinct combinations of 8 drinks can the customer order? Give your answer as an integer.", verbose=True)
```
---
# Original Model: VibeThinker-1.5B
🚨 We recommend using this model for competitive-style math and algorithm coding problems. It works better to ask the question in English. We do not advise using it for other tasks, as this is an experimental release aimed at exploring the reasoning capabilities of small models.
📁 Github | 🤖 Model Scope | 📄 Techical Report
## Introduction
VibeThinker-1.5B is a 1.5-billion parameter dense language model. With a total training cost of only $7,800 USD, it achieves reasoning performance comparable to larger models like GPT OSS-20B Medium.
## Key Performance Data
💡 Mathematical Reasoning: On the three major math benchmarks AIME24, AIME25, and HMMT25, its scores (80.3, 74.4, and 50.4, respectively) all surpass those of the initial DeepSeek R1 model, which has over 400 times the parameters (scores of 79.8, 70.0, and 41.7, respectively).
🌱 Code Generation: It achieved scores of 55.9 on LiveCodeBench v5 and 51.1 on v6. Its v6 score slightly leads Magistral Medium (50.3), underscoring its strong reasoning performance.
🔁 On the AIME 25 benchmark, VibeThinker-1.5B significantly extends the Pareto frontier of reasoning accuracy versus model scale, demonstrating that exceptional performance can be achieved with extreme parameter efficiency.
## Training Pipeline
VibeThinker-1.5B's core innovation lies in the "Spectrum-to-Signal Principle" (SSP) training framework: it first explores solution diversity during the Supervised Fine-Tuning (SFT) stage, and then optimizes its policy to reinforce correct signals in the Reinforcement Learning (RL) stage. By systematically integrating these two phases, our approach establishes diversity as the central technical design principle, enabling VibeThinker-1.5B to achieve robust performance that surpasses conventional training paradigms.