Modelpublic

DEVCamiloSepulveda/11-DeepSeekR1SP-aptanastudio-titanium

sourceHugging Faceapache-2.0updated 1y agoView on Hugging Face

0likes6downloads

Model Card

DeepSeek R1 Qwen Story Point Estimator - aptanastudio - titanium

This model is fine-tuned on issue descriptions from aptanastudio and tested on titanium for story point estimation.

Model Details

—Base Model: DeepSeek R1 Distill Qwen 1.5B
—Training Project: aptanastudio
—Test Project: titanium
—Task: Story Point Estimation (Regression)
—Architecture: PEFT (LoRA)
—Tokenizer: DeepSeek BPE Tokenizer

—Input: Issue titles
—Output: Story point estimation (continuous value)

Usage

python

from transformers import AutoModelForSequenceClassification
from peft import PeftConfig, PeftModel
from transformers import AutoTokenizer

# Load peft config model
config = PeftConfig.from_pretrained("DEVCamiloSepulveda/11-DeepSeekR1SP-aptanastudio-titanium")

# Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("DEVCamiloSepulveda/11-DeepSeekR1SP-aptanastudio-titanium")
base_model = AutoModelForSequenceClassification.from_pretrained(
    config.base_model_name_or_path,
    num_labels=1,
    torch_dtype=torch.float16,
    device_map='auto'
)
model = PeftModel.from_pretrained(base_model, "DEVCamiloSepulveda/11-DeepSeekR1SP-aptanastudio-titanium")

# Prepare input text
text = "Your issue description here"
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=20, padding="max_length")

# Get prediction
outputs = model(**inputs)
story_points = outputs.logits.item()

Training Details

—Fine-tuning method: LoRA (Low-Rank Adaptation)
—Sequence length: 20 tokens
—Best training epoch: 1 / 20 epochs
—Batch size: 32
—Training time: 103.769 seconds
—Mean Absolute Error (MAE): 4.968
—Median Absolute Error (MdAE): 4.639

Framework versions

—PEFT 0.14.0