Upload readme
Browse files
README.md
ADDED
|
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- unsloth
|
| 5 |
+
- trl
|
| 6 |
+
- sft
|
| 7 |
+
- code
|
| 8 |
+
- reasoning
|
| 9 |
+
- abliterated
|
| 10 |
+
- baukit-abliterated
|
| 11 |
+
datasets:
|
| 12 |
+
- nvidia/OpenCodeReasoning
|
| 13 |
+
language:
|
| 14 |
+
- en
|
| 15 |
+
base_model:
|
| 16 |
+
- suayptalha/Qwen3-0.6B-Code-Expert
|
| 17 |
+
pipeline_tag: text-generation
|
| 18 |
+
library_name: transformers
|
| 19 |
+
---
|
| 20 |
+
|
| 21 |
+
# Qwen3-0.6B-Code-Expert (Abliterated)
|
| 22 |
+
|
| 23 |
+
This project performs full fine-tuning on the **Qwen3-0.6B** language model to enhance its code reasoning and generation capabilities. Training was conducted exclusively on the `nvidia/OpenCodeReasoning` dataset, and the model was optimized using the bfloat16 (bf16) data type.
|
| 24 |
+
Additionally, it has been abliterated to make it steer away from censorship.
|
| 25 |
+
|
| 26 |
+
## Training Procedure
|
| 27 |
+
|
| 28 |
+
1. **Dataset Preparation**
|
| 29 |
+
|
| 30 |
+
* `nvidia/OpenCodeReasoning` dataset was used.
|
| 31 |
+
* Each example consists of code snippets paired with detailed step-by-step reasoning in Chain-of-Thought (CoT) style.
|
| 32 |
+
|
| 33 |
+
2. **Model Loading and Configuration**
|
| 34 |
+
|
| 35 |
+
* Qwen3-0.6B base model weights were loaded via the `unsloth` library in bf16 precision.
|
| 36 |
+
* Full fine-tuning (`full_finetuning=True`) was applied to all layers for optimal adaptation to code reasoning.
|
| 37 |
+
|
| 38 |
+
3. **Supervised Fine-Tuning**
|
| 39 |
+
|
| 40 |
+
* Employed the Hugging Face TRL library with the Supervised Fine-Tuning (SFT) approach.
|
| 41 |
+
* The model was trained to generate correct code solutions along with the corresponding reasoning chains.
|
| 42 |
+
|
| 43 |
+
## Purpose and Outcome
|
| 44 |
+
|
| 45 |
+
* The model’s capacity for understanding, reasoning about, and generating code was significantly improved through specialized, single-dataset training in bf16 precision.
|
| 46 |
+
* Outputs include both intermediate reasoning steps and final code solutions, enabling transparent and interpretable code generation.
|
| 47 |
+
|
| 48 |
+
## License
|
| 49 |
+
|
| 50 |
+
This project is licensed under the Apache License 2.0. See the [LICENSE](./LICENSE) file for details.
|
| 51 |
+
|
| 52 |
+
## Support
|
| 53 |
+
|
| 54 |
+
<a href="https://www.buymeacoffee.com/suayptalha" target="_blank"><img src="https://cdn.buymeacoffee.com/buttons/v2/default-yellow.png" alt="Buy Me A Coffee" style="height: 60px !important;width: 217px !important;" ></a>
|