Downloads · 30 days
9
18% of all-time downloads
math_nothink_lora_0.1
Montalte/math_nothink_lora_0.1
math_nothink_lora_0.1 is a text generation model from Montalte. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
Localize-and-Stitch merge of a math nothink SFT into Qwen/Qwen3-4B-Base.
All-time downloads
50
Public
Parameters
4B
8.1 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors8 GB · 100%
At a glance
- Task
- Text Generation
- Library
- transformers
- License
- apache-2.0
- Model type
- qwen3
- Access
- Public
- Created
- Jul 19, 2026
- Updated
- Jul 19, 2026
- SHA
- adb0a189
Try a prompt
Base models
Model card
From the Hugging Face model README
math_nothink_lora_0.1
Localize-and-Stitch merge of a math nothink SFT into Qwen/Qwen3-4B-Base.
Lineage
| Role | Checkpoint |
|---|---|
| Base | Qwen/Qwen3-4B-Base |
| SFT | modrill/Qwen3-4B-nothink-path3-ckpt2500 |
Recipe
Task vector (\tau = W_{\text{SFT}} - W_{\text{base}}), then a binary mask is trained on MergeBench/math_val (64-shot) and the masked vector is stitched back into the base.
| Setting | Value |
|---|---|
| Task | math |
| Sparsity | 0.1 |
| Mask LR | 1e7 |
| Mask epochs | 10 |
| L1 | 1e-5 |
| Mask train | SGD, batch 1, grad_accum 2, seq ≈ 2048, sdpa |
- Task
- Text Generation
- Library
- transformers
- Type
- qwen3
- License
- apache-2.0
- Created
- Jul 19, 2026
- Updated
- Jul 19, 2026