fenyo
Qwen2.5-7B-base2instruct-SFT
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Get help setting up a custom Dedicated Endpoints.
Talk with our engineer to get a quote for reserved GPU instances with discounts.
README
License: apache-2.0Entraînement
Full fine-tuning (TRL SFTTrainer), format ChatML, loss sur la réponse assistant uniquement
(assistant_only_loss + balises {% generation %}), liger-kernel, packing, bf16, attention SDPA.
1 epoch, lr 5e-6 cosine, seq_len 4096. Données :
allenai/tulu-3-sft-mixture (180k).
Résultats (lm-eval, backend vLLM)
| étape | IFEval (prompt strict) | GSM8K (flexible) | MMLU |
|---|---|---|---|
| base Qwen2.5-7B | 27.4 | 83.0 | 71.8 |
| + SFT (ce modèle) | 44.9 | 77.5 | 69.1 |
| + DPO | 44.7 | 77.1 | 69.9 |
| + RLVR | 45.1 | 77.4 | 69.9 |
C'est le SFT qui apporte l'essentiel du gain en suivi d'instructions (IFEval 27→45). Détails et analyse : dépôt GitHub.
Usage
Format ChatML, via tokenizer.apply_chat_template. Voir l'exemple sur le
modèle final.
Model provider
fenyo
Model tree
Base
Qwen/Qwen2.5-7B
Fine-tuned
this model
Modalities
Input
Text
Output
Text
Pricing
Dedicated Endpoints
View detailsSupported Functionality
Model APIs
Dedicated Endpoints
Container
More information