|
|
--- |
|
|
license: apache-2.0 |
|
|
base_model: Qwen/Qwen2.5-32B-Instruct |
|
|
tags: |
|
|
- lora |
|
|
- qwen |
|
|
- fine-tuned |
|
|
library_name: peft |
|
|
--- |
|
|
|
|
|
# LoRA Fine-tuned Qwen3-32B |
|
|
|
|
|
This model is a LoRA fine-tuned version of Qwen2.5-32B-Instruct. |
|
|
|
|
|
## Training Details |
|
|
- Training runtime: 13,587 seconds (3h 46m) |
|
|
- Training samples per second: 2.208 |
|
|
- Final training loss: 0.664 |
|
|
- Epochs: 1.0 |
|
|
- Training steps: 1,161 |
|
|
|
|
|
## Usage |
|
|
|
|
|
```python |
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
|
from peft import PeftModel |
|
|
|
|
|
base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-32B-Instruct") |
|
|
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-32B-Instruct") |
|
|
model = PeftModel.from_pretrained(base_model, "abhayesian/lora-qwen3-32b-docs") |
|
|
``` |
|
|
|
