Downloads · 30 days
7
29% of all-time downloads
mistral-7b-arc-stage2
minkyungpark/mistral-7b-arc-stage2
mistral-7b-arc-stage2 is a text generation model from minkyungpark. Use it when you need the model to write or continue text. It is set up for peft. The card lists the license as apache-2.0.
This is the final PEFT LoRA adapter trained with a two-stage setup for ARC.
All-time downloads
24
Public
Repo size
7.3 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors6.8 MB · 75%
At a glance
- Task
- Text Generation
- Library
- peft
- License
- apache-2.0
- Access
- Public
- Created
- Aug 10, 2026
- Updated
- Aug 10, 2026
- SHA
- d00e55b3
Try a prompt
Base models
Model card
From the Hugging Face model README
Mistral-7B ARC Stage 2
This is the final PEFT LoRA adapter trained with a two-stage setup for ARC.
Model
- Base model:
mistralai/Mistral-7B-v0.1 - Base revision:
26bca36bde8333b5d7f72e9ed20ccda6a618af24 - Format: PEFT LoRA adapter with tokenizer
- LoRA: rank 8, alpha 16, dropout 0.05,
q_projandv_proj - Training: Stage 1 answer-completion SFT followed by Stage 2 answer NLL plus choice-ranking loss
Evaluation
ARC-Challenge test, 25-shot, 1,172 samples, lm-evaluation-harness commit ae79b1217aad7738b91e88a4017c86a5d5e45aa7:
acc: 0.6237201365acc_norm: 0.6638225256
The result uses this adapter with the pinned base model. The base-model weights are not included in this repository.
Limitation
Stage 2 hardness scores were computed with an NF4 4-bit model, while Stage 2 training used BF16. This precision mismatch may affect examples near the hard-set selection boundary.
- Task
- Text Generation
- Library
- peft
- License
- apache-2.0
- Languages
- en
- Created
- Aug 10, 2026
- Updated
- Aug 10, 2026