|
|
--- |
|
|
base_model: Qwen/Qwen2.5-Coder-14B |
|
|
library_name: peft |
|
|
pipeline_tag: text-generation |
|
|
license: apache-2.0 |
|
|
tags: |
|
|
- code |
|
|
- coding-assistant |
|
|
- lora |
|
|
- refactor |
|
|
- bug-fix |
|
|
- optimization |
|
|
- async |
|
|
- concurrency |
|
|
- security |
|
|
- logging |
|
|
- networking |
|
|
- transformers |
|
|
--- |
|
|
|
|
|
# EdgePulse Coder 14B (LoRA) |
|
|
|
|
|
**EdgePulse Coder 14B** is a production-grade coding assistant fine-tuned using LoRA on top of **Qwen2.5-Coder-14B**. |
|
|
It is designed to handle real-world software engineering workflows with high reliability and correctness. |
|
|
|
|
|
--- |
|
|
|
|
|
## Model Details |
|
|
|
|
|
### Model Description |
|
|
|
|
|
EdgePulse Coder 14B focuses on **practical developer tasks**, trained on a large, strictly validated dataset covering: |
|
|
|
|
|
- Bug fixing |
|
|
- Code explanation |
|
|
- Refactoring |
|
|
- Optimization |
|
|
- Async & concurrency correction |
|
|
- Logging & observability |
|
|
- Security & defensive coding |
|
|
- Networking & I/O handling |
|
|
- Multi-file context reasoning |
|
|
- Test generation and impact analysis |
|
|
|
|
|
The model is optimized for **IDE usage**, **CLI workflows**, and **Cursor-like streaming environments**. |
|
|
|
|
|
--- |
|
|
|
|
|
- **Developed by:** EdgePulseAI |
|
|
- **Shared by:** EdgePulseAI |
|
|
- **Model type:** Large Language Model (Code-focused) |
|
|
- **Language(s):** Python, JavaScript, TypeScript, Bash (primary), general programming concepts |
|
|
- **License:** Apache-2.0 |
|
|
- **Finetuned from:** Qwen/Qwen2.5-Coder-14B |
|
|
|
|
|
--- |
|
|
|
|
|
## Model Sources |
|
|
|
|
|
- **Base Model:** https://huggingface.co/Qwen/Qwen2.5-Coder-14B |
|
|
- **Website:** https://EdgePulseAi.com |
|
|
|
|
|
--- |
|
|
|
|
|
## Uses |
|
|
|
|
|
### Direct Use |
|
|
|
|
|
EdgePulse Coder 14B can be used directly for: |
|
|
|
|
|
- Code explanation |
|
|
- Bug fixing |
|
|
- Refactoring existing code |
|
|
- Generating tests |
|
|
- Improving logging and error handling |
|
|
- Fixing async / concurrency bugs |
|
|
- Secure coding suggestions |
|
|
- Network & I/O robustness |
|
|
|
|
|
### Downstream Use |
|
|
|
|
|
- IDE assistants (VS Code / Cursor-style tools) |
|
|
- CI/CD automation |
|
|
- Code review bots |
|
|
- Developer copilots |
|
|
- Internal engineering tools |
|
|
|
|
|
### Out-of-Scope Use |
|
|
|
|
|
- Medical or legal advice |
|
|
- Autonomous system control |
|
|
- High-risk decision making without human review |
|
|
|
|
|
--- |
|
|
|
|
|
## Bias, Risks, and Limitations |
|
|
|
|
|
- The model may occasionally produce syntactically correct but logically incorrect code. |
|
|
- Security-sensitive code should always be reviewed by humans. |
|
|
- Performance depends on correct prompt framing and context size. |
|
|
|
|
|
### Recommendations |
|
|
|
|
|
- Use human review for production deployments. |
|
|
- Combine with static analysis and testing tools. |
|
|
- Prefer structured prompts for multi-file tasks. |
|
|
|
|
|
--- |
|
|
|
|
|
## How to Get Started with the Model |
|
|
|
|
|
```python |
|
|
from transformers import AutoTokenizer, AutoModelForCausalLM |
|
|
from peft import PeftModel |
|
|
|
|
|
base_model = "Qwen/Qwen2.5-Coder-14B" |
|
|
adapter_model = "edgepulse-ai/EdgePulse-Coder-14B-LoRA" |
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(base_model) |
|
|
|
|
|
model = AutoModelForCausalLM.from_pretrained( |
|
|
base_model, |
|
|
device_map="auto", |
|
|
torch_dtype="auto" |
|
|
) |
|
|
|
|
|
model = PeftModel.from_pretrained(model, adapter_model) |
|
|
model.eval() |
|
|
|
|
|
prompt = "Fix this bug:\n\ndef add(a,b): return a-b" |
|
|
inputs = tokenizer(prompt, return_tensors="pt").to(model.device) |
|
|
|
|
|
output = model.generate(**inputs, max_new_tokens=128) |
|
|
print(tokenizer.decode(output[0], skip_special_tokens=True)) |
