Downloads · 30 days
0
slm-question-gen-sft
nayan90k/slm-question-gen-sft
slm-question-gen-sft is a machine learning model from nayan90k. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
A 114-million-parameter Small Language Model fine-tuned to generate educational questions from a given passage of text.
Access
Public
Updated Apr 18, 2026
Repo size
913 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt913 MB · 100%
At a glance
- License
- mit
- Access
- Public
- Created
- Mar 18, 2026
- Updated
- Apr 18, 2026
- SHA
- 12957b09
Model card
From the Hugging Face model README
SLM Question Generator - SFT (114M)
A 114-million-parameter Small Language Model fine-tuned to generate educational questions from a given passage of text.
Model Details
- Architecture: Decoder-only Transformer
- Parameters: 114.1 M
- Layers: 12
- d_model: 768
- Attention: GQA (12 query heads / 4 KV heads)
- Vocabulary: tiktoken
r50k_base+ 3 special tokens (50,260 total) - Context window: 4096 tokens
Evaluation Metrics
Evaluated on SQuAD v2 validation set (50 samples):
| Metric | Score |
|---|---|
| BERTScore Precision | 0.8972 |
| BERTScore Recall | 0.8920 |
| BERTScore F1 | 0.8945 |
Dataset
- Training samples: 83,861
- Validation samples: 4,414
- Sources: SQuAD v2, HotpotQA
Usage
from src.inference.generate import load_model, generate_from_text
model, enc, device = load_model()
questions = generate_from_text(
"Photosynthesis is the process by which green plants prepare food using sunlight.",
question_type="short_answer",
model=model,
enc=enc,
device=device,
temperature=0.0,
)
print(questions[0])
Links
Tokenizer Note
This model uses the tiktoken library with the r50k_base encoding, plus <|im_start|>, <|im_end|>, and <|pad|> tokens.
- License
- mit
- Languages
- en
- Created
- Mar 18, 2026
- Updated
- Apr 18, 2026