mathstral

MathΣtral: a 7B model designed for math reasoning and scientific discovery by Mistral AI.

7b
Hızlı Kurulum (Ollama kuruluysa)
ollama run mathstral

Ollama kurulu değil mi? ollama.com/download — Windows, macOS ve Linux için ücretsiz. İlk çalıştırmada model indirilir, sonrası tamamen çevrimdışıdır.

Varyantlar

Boyut büyüdükçe kalite artar, donanım ihtiyacı yükselir. Başlangıç için küçük varyantı deneyin.

EtiketBoyutBağlamGirdiKomut
latest 4.1GB 32K Text ollama run mathstral:latest
7b 4.1GB 32K Text ollama run mathstral:7b
7b-v0.1-q2_K 2.7GB 32K Text ollama run mathstral:7b-v0.1-q2_K
7b-v0.1-q3_K_S 3.2GB 32K Text ollama run mathstral:7b-v0.1-q3_K_S
7b-v0.1-q3_K_M 3.5GB 32K Text ollama run mathstral:7b-v0.1-q3_K_M
7b-v0.1-q3_K_L 3.8GB 32K Text ollama run mathstral:7b-v0.1-q3_K_L
7b-v0.1-q4_0 4.1GB 32K Text ollama run mathstral:7b-v0.1-q4_0
7b-v0.1-q4_1 4.6GB 32K Text ollama run mathstral:7b-v0.1-q4_1
7b-v0.1-q4_K_S 4.1GB 32K Text ollama run mathstral:7b-v0.1-q4_K_S
7b-v0.1-q4_K_M 4.4GB 32K Text ollama run mathstral:7b-v0.1-q4_K_M
7b-v0.1-q5_0 5.0GB 32K Text ollama run mathstral:7b-v0.1-q5_0
7b-v0.1-q5_1 5.4GB 32K Text ollama run mathstral:7b-v0.1-q5_1
7b-v0.1-q5_K_S 5.0GB 32K Text ollama run mathstral:7b-v0.1-q5_K_S
7b-v0.1-q5_K_M 5.1GB 32K Text ollama run mathstral:7b-v0.1-q5_K_M
7b-v0.1-q6_K 5.9GB 32K Text ollama run mathstral:7b-v0.1-q6_K
7b-v0.1-q8_0 7.7GB 32K Text ollama run mathstral:7b-v0.1-q8_0
7b-v0.1-fp16 14GB 32K Text ollama run mathstral:7b-v0.1-fp16

Model Detayları ve Benchmarklar (kaynak: ollama.com)

Mistral AI is contributing Mathstral to the science community to bolster efforts in advanced mathematical problems requiring complex, multi-step logical reasoning. The Mathstral release is part of their broader effort to support academic projects—it was produced in the context of Mistral AI’s collaboration with Project Numina.

Akin to Isaac Newton in his time, Mathstral stands on the shoulders of Mistral 7B and specializes in STEM subjects. It achieves state-of-the-art reasoning capacities in its size category across various industry-standard benchmarks.

mathstral-breakdown.png

Benchmarks

Mathstral can achieve significantly better results with more inference-time computation: Mathstral 7B scores 68.37% on MATH with majority voting and 74.59% with a strong reward model among 64 candidates.

mathstral-benchmarks.png

References

Blog Post

Hugging Face