codestral

Codestral is Mistral AI’s first-ever code model designed for code generation tasks.

22b
Hızlı Kurulum (Ollama kuruluysa)
ollama run codestral

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 13GB 32K Text ollama run codestral:latest
v0.1 13GB 32K Text ollama run codestral:v0.1
22b 13GB 32K Text ollama run codestral:22b
22b-v0.1-q2_K 8.3GB 32K Text ollama run codestral:22b-v0.1-q2_K
22b-v0.1-q3_K_S 9.6GB 32K Text ollama run codestral:22b-v0.1-q3_K_S
22b-v0.1-q3_K_M 11GB 32K Text ollama run codestral:22b-v0.1-q3_K_M
22b-v0.1-q3_K_L 12GB 32K Text ollama run codestral:22b-v0.1-q3_K_L
22b-v0.1-q4_0 13GB 32K Text ollama run codestral:22b-v0.1-q4_0
22b-v0.1-q4_1 14GB 32K Text ollama run codestral:22b-v0.1-q4_1
22b-v0.1-q4_K_S 13GB 32K Text ollama run codestral:22b-v0.1-q4_K_S
22b-v0.1-q4_K_M 13GB 32K Text ollama run codestral:22b-v0.1-q4_K_M
22b-v0.1-q5_0 15GB 32K Text ollama run codestral:22b-v0.1-q5_0
22b-v0.1-q5_1 17GB 32K Text ollama run codestral:22b-v0.1-q5_1
22b-v0.1-q5_K_S 15GB 32K Text ollama run codestral:22b-v0.1-q5_K_S
22b-v0.1-q5_K_M 16GB 32K Text ollama run codestral:22b-v0.1-q5_K_M
22b-v0.1-q6_K 18GB 32K Text ollama run codestral:22b-v0.1-q6_K
22b-v0.1-q8_0 24GB 32K Text ollama run codestral:22b-v0.1-q8_0

Model Detayları ve Benchmarklar (kaynak: ollama.com)

Codestral

Codestral Hello World

Codestral is Mistral AI’s first-ever code model designed for code generation tasks. It is a 22B model.

Fluent in 80+ programming languages

Codestral is trained on a dataset of over 80 programming languages, including Python, Java, C, C++, JavaScript, Swift, Fortran and Bash.

The model can complete coding functions, write tests, and complete any partial code using a fill-in-the-middle mechanism.

Benchmarks

With its larger context window of 32k (compared to 4k, 8k or 16k for competitors), Codestral outperforms all other models in RepoBench, a long-range eval for code generation

Benchmarks

FIM benchmarks

Reference

Mistral AI - Codestral: Hello, World