Text Generation
PEFT
Safetensors
code
code-review
bug-fixing
qwen
qwen2.5-coder
qlora
trl
static-analysis
conversational
Eval Results (legacy)
Instructions to use devanshty/Code-Autopsy with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use devanshty/Code-Autopsy with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-Coder-7B-Instruct") model = PeftModel.from_pretrained(base_model, "devanshty/Code-Autopsy") - Notebooks
- Google Colab
- Kaggle
| { | |
| "experiment_name": "Code-Autopsy QLoRA Bug-Fixing Fine-Tuning v2", | |
| "timestamp": "2026-08-23T04:45:00+05:30", | |
| "base_model": { | |
| "name": "Qwen/Qwen2.5-Coder-7B-Instruct", | |
| "quantization": "4-bit NF4", | |
| "initial_train_loss": 2.162, | |
| "initial_eval_loss": 1.397, | |
| "initial_token_accuracy": "60.29%" | |
| }, | |
| "finetuned_model": { | |
| "name": "Code-Autopsy QLoRA Adapter v2", | |
| "training_framework": "PEFT / TRL SFTTrainer", | |
| "final_train_loss": 0.5211, | |
| "final_eval_loss": 0.2442, | |
| "final_token_accuracy": "93.20%", | |
| "best_step_loss": 0.2472 | |
| }, | |
| "training_metrics": { | |
| "epochs": 3, | |
| "total_steps": 246, | |
| "eval_loss_reduction": "-82.5%", | |
| "accuracy_gain": "+32.91%", | |
| "effective_batch_size": 8, | |
| "learning_rate": 0.0002, | |
| "lr_scheduler": "cosine" | |
| }, | |
| "cloud_logging": { | |
| "platform": "Weights & Biases", | |
| "wandb_run_url": "https://wandb.ai/devanshtyagi1903-innothoughts/code-autopsy/runs/gc70q2q2", | |
| "wandb_project_url": "https://wandb.ai/devanshtyagi1903-innothoughts/code-autopsy" | |
| }, | |
| "hardware_specs": { | |
| "gpu": "NVIDIA GeForce RTX 5060 (8GB VRAM)", | |
| "quantization": "4-bit NF4 (bitsandbytes)", | |
| "precision": "bfloat16", | |
| "optimizer": "adamw_8bit" | |
| } | |
| } |