FinTune: LoRA-adapted Qwen2.5-1.5B-Instruct for Financial Sentiment
LoRA (r=16, alpha=32) adapter fine-tuned on the Financial PhraseBank dataset (sentences_50agree, 4,846 sentences, 80/10/10 split) for 3-class financial sentiment classification (positive / negative / neutral).
Trained with a custom PyTorch training loop (manual forward pass, loss
computation, backward pass, and optimizer step — no Trainer class) on a
free-tier Colab T4 GPU using 4-bit QLoRA.
Benchmark (held-out test set, N=486)
| Model | Accuracy | Macro F1 |
|---|---|---|
| Base Qwen2.5-1.5B-Instruct | 50.6% | 0.540 |
| Fine-tuned (this adapter) | 88.9% | 0.887 |
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-1.5B-Instruct")
model = PeftModel.from_pretrained(base, "wrtdevcod/fintune-qwen2.5-1.5b-lora")
- Downloads last month
- 19
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support