| Les deux révisions précédentesRévision précédenteProchaine révision | Révision précédente |
| informatique:ai_lm:ai_coding [04/06/2026 08:52] – [Models] cyrille | informatique:ai_lm:ai_coding [03/08/2026 22:02] (Version actuelle) – [Models] cyrille |
|---|
| Essais de quelques modèles pour de l'assistance au code. Sur mon portable ''Core™ i7-1360P'' (2023 Q1) avec ''RTX 3060 en Thunderbolt 4 (thb4)'' le modèle le plus adapté avec llama.cpp est ''gpt-oss-20b-UD-Q4_K_XL.gguf''. | Essais de quelques modèles pour de l'assistance au code. Sur mon portable ''Core™ i7-1360P'' (2023 Q1) avec ''RTX 3060 en Thunderbolt 4 (thb4)'' le modèle le plus adapté avec llama.cpp est ''gpt-oss-20b-UD-Q4_K_XL.gguf''. |
| |
| [[https://www.cpu-monkey.com/en/compare_cpu-intel_core_ultra_7_270k_plus-vs-intel_core_i7_1360p|Core i7-1360P vs Core Ultra 7 270K Plus]] | Comparaison [[https://www.cpu-monkey.com/en/compare_cpu-intel_core_ultra_7_270k_plus-vs-intel_core_i7_1360p|Core i7-1360P vs Core Ultra 7 270K Plus]] |
| |
| [[https://lab.cyrille.giquello.fr/AI-compare/models-metadata/#state=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|GGUF Models Metadata Viewer]] : Un viewer des meta-données des modèles que j'essaye en local réalisé sans coder, juste assistant IA et "[[https://www.crackedaiengineering.com/ai-models/opencode-big-pickle|OpenCode Zen Big Pickle]]" et "[[https://mistral.ai/news/devstral-2-vibe-cli|Mistral Devstral 2]]". | [[https://lab.cyrille.giquello.fr/AI-compare/models-metadata/#state=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|GGUF Models Metadata Viewer]] : Un viewer des meta-données des modèles que j'essaye en local réalisé sans coder, juste assistant IA et "[[https://www.crackedaiengineering.com/ai-models/opencode-big-pickle|OpenCode Zen Big Pickle]]" et "[[https://mistral.ai/news/devstral-2-vibe-cli|Mistral Devstral 2]]". |
| * [[https://huggingface.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF|Qwen3-Coder-30B-A3B-Instruct]] | * [[https://huggingface.co/unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF|Qwen3-Coder-30B-A3B-Instruct]] |
| * layers=48, n_ctx_train=262144, n_embd=2048, n_rot=128, n_expert=128, n_expert_used=8, n_vocab=151936, n_merges=151387, max token length=256 | * layers=48, n_ctx_train=262144, n_embd=2048, n_rot=128, n_expert=128, n_expert_used=8, n_vocab=151936, n_merges=151387, max token length=256 |
| * avec 12Go ''<nowiki>--ctx-size 70000 --n-gpu-layers 23</nowiki>'' | * RTX 3060 12Go ''<nowiki>--ctx-size 70000 --n-gpu-layers 23</nowiki>'' |
| * avec 12Go ''<nowiki>--ctx-size 40000 --n-gpu-layers 26</nowiki>'' | * RTX 3060 12Go ''<nowiki>--ctx-size 40000 --n-gpu-layers 26</nowiki>'' |
| | * [[https://huggingface.co/unsloth/Qwen3-Coder-Next-GGUF|Qwen3-Coder-Next 80B]] |
| | * RTX 5060 16 Go: ''<nowiki>-c 101888 --cache-type-k q8_0 --cache-type-v q8_0 --n-cpu-moe 37</nowiki>'' nvtop: 15.617Gi/15.929Gi |
| * DeepSeek2 | * DeepSeek2 |
| * [[https://huggingface.co/bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF|DeepSeek-Coder-V2-Lite-Instruct]] **16B** by bartowski | * [[https://huggingface.co/bartowski/DeepSeek-Coder-V2-Lite-Instruct-GGUF|DeepSeek-Coder-V2-Lite-Instruct]] **16B** by bartowski |