D

Deepseek R1 0528 GPTQ Int4 Int8Mix Compact

Developed by QuantTrio
The GPTQ quantized version of the DeepSeek-R1-0528 model, using a quantization scheme of Int4 + selective Int8, which reduces the file size while ensuring the generation quality.
Downloads 258
Release Time : 6/1/2025

Model Overview

This model is a quantized version of DeepSeek-R1-0528. Through the mixed Int4 and Int8 quantization technology, it optimizes the model inference speed and video memory usage, and is suitable for deployment scenarios with different hardware configurations.

Model Features

Mixed quantization technology
Adopt a quantization scheme of Int4 + selective Int8. Only the quantization-sensitive layers use Int8, and the rest use Int4 to balance the generation quality and file size.
Multiple quantization variants
Provide three quantization variants: Lite, Compact, and Medium to adapt to different hardware configurations and quality requirements.
Optimized inference performance
Through layer-by-layer fine-grained quantization, it significantly alleviates the problem of decreased inference accuracy caused by pure Int4 quantization.
Enhanced reasoning ability
Compared with the previous version, it has significant improvements in handling complex reasoning tasks, such as mathematical problems and programming challenges.

Model Capabilities

Complex logical reasoning
Mathematical problem solving
Code generation and understanding
Long text generation
Multi-round dialogue

Use Cases

Education
Mathematical competition problem solving
Solve math competition problems such as AIME
The accuracy rate reached 87.5% in the AIME 2025 test
Programming teaching
Assist in programming learning and code debugging
The Pass@1 reached 73.3% in the LiveCodeBench test
Software development
Code generation
Generate high-quality code according to requirements
The solution rate reached 57.6% in the SWE Verified test
Code review
Analyze the code and provide improvement suggestions
Research
Academic Q&A
Answer complex academic questions
The Pass@1 reached 81.0% in the GPQA-Diamond test
Featured Recommended AI Models
AIbase
Empowering the Future, Your AI Solution Knowledge Base
© 2025AIbase