Upload fine-tuned model
Browse files
README.md
CHANGED
|
@@ -7,20 +7,19 @@ tags:
|
|
| 7 |
- commit-message
|
| 8 |
- qwen2
|
| 9 |
- lora
|
| 10 |
-
base_model: muellerzr/qwen-0.5-git-commit-message-generation
|
| 11 |
datasets:
|
| 12 |
- bigcode/commitpackft
|
| 13 |
---
|
| 14 |
|
| 15 |
# Git Commit Message Generator
|
| 16 |
|
| 17 |
-
Fine-tuned
|
| 18 |
|
| 19 |
## Model Description
|
| 20 |
|
| 21 |
This model was fine-tuned using LoRA (Low-Rank Adaptation) on the CommitPackFT dataset to generate concise, professional commit messages from git diffs.
|
| 22 |
|
| 23 |
-
**Base Model**:
|
| 24 |
**Fine-tuning Method**: LoRA (r=16, alpha=32)
|
| 25 |
**Training Data**: 55K filtered commits from CommitPackFT
|
| 26 |
**Languages**: Python, JavaScript, TypeScript, Java, C++, Go, Rust, and more
|
|
@@ -68,28 +67,25 @@ inputs = tokenizer(prompt, return_tensors="pt")
|
|
| 68 |
with torch.no_grad():
|
| 69 |
outputs = model.generate(
|
| 70 |
**inputs,
|
| 71 |
-
max_new_tokens=
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
top_p=0.9,
|
| 75 |
)
|
| 76 |
|
| 77 |
message = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 78 |
print(message.split("Commit message:")[-1].strip())
|
| 79 |
-
# Output: "
|
| 80 |
```
|
| 81 |
|
| 82 |
### CLI Tool
|
| 83 |
|
| 84 |
-
For easier usage, install the companion CLI tool:
|
| 85 |
|
| 86 |
```bash
|
| 87 |
-
pip install
|
| 88 |
commit-gen generate --commit
|
| 89 |
```
|
| 90 |
|
| 91 |
-
See the [GitHub repository](https://github.com/YOUR_USERNAME/GitCommitGenerator) for more details.
|
| 92 |
-
|
| 93 |
## Training Details
|
| 94 |
|
| 95 |
### Training Data
|
|
@@ -144,7 +140,7 @@ This model is intended to assist developers in writing commit messages, not repl
|
|
| 144 |
|
| 145 |
```bibtex
|
| 146 |
@misc{git-commit-generator,
|
| 147 |
-
author = {
|
| 148 |
title = {Git Commit Message Generator},
|
| 149 |
year = {2024},
|
| 150 |
publisher = {Hugging Face},
|
|
|
|
| 7 |
- commit-message
|
| 8 |
- qwen2
|
| 9 |
- lora
|
|
|
|
| 10 |
datasets:
|
| 11 |
- bigcode/commitpackft
|
| 12 |
---
|
| 13 |
|
| 14 |
# Git Commit Message Generator
|
| 15 |
|
| 16 |
+
Fine-tuned Qwen-0.5B model for generating professional Git commit messages from code diffs.
|
| 17 |
|
| 18 |
## Model Description
|
| 19 |
|
| 20 |
This model was fine-tuned using LoRA (Low-Rank Adaptation) on the CommitPackFT dataset to generate concise, professional commit messages from git diffs.
|
| 21 |
|
| 22 |
+
**Base Model**: Qwen-0.5B
|
| 23 |
**Fine-tuning Method**: LoRA (r=16, alpha=32)
|
| 24 |
**Training Data**: 55K filtered commits from CommitPackFT
|
| 25 |
**Languages**: Python, JavaScript, TypeScript, Java, C++, Go, Rust, and more
|
|
|
|
| 67 |
with torch.no_grad():
|
| 68 |
outputs = model.generate(
|
| 69 |
**inputs,
|
| 70 |
+
max_new_tokens=30,
|
| 71 |
+
do_sample=False, # Deterministic
|
| 72 |
+
pad_token_id=tokenizer.eos_token_id,
|
|
|
|
| 73 |
)
|
| 74 |
|
| 75 |
message = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 76 |
print(message.split("Commit message:")[-1].strip())
|
| 77 |
+
# Output: "Check for user existence before accessing password"
|
| 78 |
```
|
| 79 |
|
| 80 |
### CLI Tool
|
| 81 |
|
| 82 |
+
For easier usage, install the companion CLI tool from the [GitHub repository](https://github.com/rajtiwariee/GitCommitGenerator):
|
| 83 |
|
| 84 |
```bash
|
| 85 |
+
pip install -e .
|
| 86 |
commit-gen generate --commit
|
| 87 |
```
|
| 88 |
|
|
|
|
|
|
|
| 89 |
## Training Details
|
| 90 |
|
| 91 |
### Training Data
|
|
|
|
| 140 |
|
| 141 |
```bibtex
|
| 142 |
@misc{git-commit-generator,
|
| 143 |
+
author = {Raj Tiwari},
|
| 144 |
title = {Git Commit Message Generator},
|
| 145 |
year = {2024},
|
| 146 |
publisher = {Hugging Face},
|