Référence des modèles Ollama
Parcourez la base de modèles Ollama : tailles, capacités et usages.
llama3.1
Meta's latest Llama model. Great general-purpose performance.
Paramètres: 8B/70B/405BTaille: 4.7GB/40GB/231GBContexte: 128K
generalcodechat
$ ollama run llama3.1
llama3.2
Lightweight Llama models optimized for edge and mobile.
Paramètres: 1B/3BTaille: 1.3GB/2.0GBContexte: 128K
edgesmall
$ ollama run llama3.2
mistral
Mistral 7B - efficient model with strong performance.
Paramètres: 7BTaille: 4.1GBContexte: 32K
generalfast
$ ollama run mistral
mixtral
Mixture of Experts model. Sparse activation for efficiency.
Paramètres: 8x7B/8x22BTaille: 26GB/80GBContexte: 64K
generalMoE
$ ollama run mixtral
codellama
Code-specialized Llama. Supports code completion and infilling.
Paramètres: 7B/13B/34B/70BTaille: 3.8GB/7.4GB/19GB/40GBContexte: 16K
codecompletion
$ ollama run codellama
deepseek-coder-v2
DeepSeek Coder v2. State-of-the-art coding model.
Paramètres: 16B/236BTaille: 8.9GB/133GBContexte: 128K
codemath
$ ollama run deepseek-coder-v2
phi3
Microsoft Phi-3. Small but capable reasoning model.
Paramètres: 3.8B/14BTaille: 2.2GB/7.9GBContexte: 128K
reasoningsmall
$ ollama run phi3
gemma2
Google Gemma 2. Efficient open model family.
Paramètres: 2B/9B/27BTaille: 1.6GB/5.4GB/16GBContexte: 8K
generalgoogle
$ ollama run gemma2
qwen2.5
Alibaba Qwen 2.5. Strong multilingual and coding.
Paramètres: 0.5B-72BTaille: 0.4-41GBContexte: 128K
multilingualcode
$ ollama run qwen2.5
nomic-embed-text
Text embedding model for RAG and semantic search.
Paramètres: 137MTaille: 274MBContexte: 8K
embeddingRAG
$ ollama run nomic-embed-text