77 lines
2.1 KiB
Markdown
77 lines
2.1 KiB
Markdown
|
---
|
||
|
|
license: apache-2.0
|
||
|
|
base_model: Qwen/Qwen3-1.7B
|
||
|
|
datasets:
|
||
|
|
- ericholam/codeforces-sft-dataset-beta
|
||
|
|
- TeichAI/claude-4.5-opus-high-reasoning-250x
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
tags:
|
||
|
|
- code
|
||
|
|
- reasoning
|
||
|
|
- competitive-programming
|
||
|
|
- sft
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
library_name: transformers
|
||
|
|
---
|
||
|
|
|
||
|
|
# Qwen3-1.7B-Sushi-Coder
|
||
|
|
|
||
|
|
A fine-tuned Qwen3-1.7B model optimized for code generation and competitive programming.
|
||
|
|
|
||
|
|
## Model Details
|
||
|
|
|
||
|
|
- **Base Model:** [Qwen/Qwen3-1.7B](https://huggingface.co/Qwen/Qwen3-1.7B)
|
||
|
|
- **Fine-tuning Method:** SFT with LoRA (merged)
|
||
|
|
- **Training Steps:** 1000
|
||
|
|
- **Context Length:** 2048
|
||
|
|
|
||
|
|
## Training
|
||
|
|
|
||
|
|
This model was fine-tuned using:
|
||
|
|
- **LoRA** (r=8, alpha=16) on attention and MLP layers
|
||
|
|
- **Liger Kernel** for memory efficiency
|
||
|
|
- **Packing** with FlashAttention-2
|
||
|
|
- **Cosine learning rate schedule** (2e-5 peak)
|
||
|
|
|
||
|
|
### Datasets
|
||
|
|
|
||
|
|
- [ericholam/codeforces-sft-dataset-beta](https://huggingface.co/datasets/ericholam/codeforces-sft-dataset-beta) - 1408 competitive programming examples
|
||
|
|
- [TeichAI/claude-4.5-opus-high-reasoning-250x](https://huggingface.co/datasets/TeichAI/claude-4.5-opus-high-reasoning-250x) - High-quality reasoning examples
|
||
|
|
|
||
|
|
## Usage
|
||
|
|
|
||
|
|
```python
|
||
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||
|
|
|
||
|
|
model = AutoModelForCausalLM.from_pretrained(
|
||
|
|
"bigatuna/Qwen3-1.7B-Sushi-Coder",
|
||
|
|
torch_dtype="auto",
|
||
|
|
device_map="auto",
|
||
|
|
trust_remote_code=True,
|
||
|
|
)
|
||
|
|
tokenizer = AutoTokenizer.from_pretrained("bigatuna/Qwen3-1.7B-Sushi-Coder")
|
||
|
|
|
||
|
|
messages = [
|
||
|
|
{"role": "user", "content": "Write a Python function to solve the two-sum problem."}
|
||
|
|
]
|
||
|
|
|
||
|
|
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
||
|
|
inputs = tokenizer(text, return_tensors="pt").to(model.device)
|
||
|
|
|
||
|
|
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.6, top_p=0.95, top_k=20)
|
||
|
|
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
|
||
|
|
```
|
||
|
|
|
||
|
|
## Sampling Parameters
|
||
|
|
|
||
|
|
For best results with Qwen3 models:
|
||
|
|
- **Temperature:** 0.6-0.7
|
||
|
|
- **Top-p:** 0.95
|
||
|
|
- **Top-k:** 20
|
||
|
|
- **Do not use greedy decoding** (temp=0 causes repetitions)
|
||
|
|
|
||
|
|
## License
|
||
|
|
|
||
|
|
Apache 2.0
|