gandhiraketla277/gpt-neo-1.3b-dolly15k-lora
011
GPT-Neo 1.3B LoRA Fine-tuned on Dolly-15k
This is a LoRA (Low-Rank Adaptation) fine-tuned version of EleutherAI/gpt-neo-1.3B on the Dolly-15k instruction-following dataset.
Model Details
- Base Model: EleutherAI/gpt-neo-1.3B
- Fine-tuning Method: LoRA (Low-Rank Adaptation)
- Dataset: Databricks Dolly-15k
- Training Date: 2025-08-10
- Trainable Parameters: 3,145,728 (0.24% of total parameters)
Training Results
- Initial Loss: 2.32
- Final Loss: 2.20
- Training: Stable convergence with no loss explosions
- Data Processing: Fixed padding issues for 4x efficiency gain
LoRA Configuration
LoraConfig(
r=16,
lora_alpha=32,
target_modules=["q_proj", "v_proj"],
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM"
)Usage
Loading the Model
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch
# Load base model
base_model = AutoModelForCausalLM.from_pretrained(
"EleutherAI/gpt-neo-1.3B",
torch_dtype=torch.float16,
device_map="auto"
)
# Load LoRA adapters
model = PeftModel.from_pretrained(base_model, "YOUR_USERNAME/gpt-neo-1.3b-dolly15k-lora")
# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-neo-1.3B")
if tokenizer.pad_token is None:
tokenizer.pad_token = tokenizer.eos_tokenGenerating Text
# For best results, use the instruction format:
prompt = "Instruction: Explain machine learning in simple terms. Input: Answer:"
inputs = tokenizer(prompt, return_tensors="pt")
with torch.inference_mode():
outputs = model.generate(
**inputs,
max_new_tokens=100,
temperature=0.7,
do_sample=True,
repetition_penalty=1.3,
no_repeat_ngram_size=3
)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response[len(prompt):].strip())Training Format
This model works best with the Dolly-15k instruction format:
Instruction: [Your instruction here]
Input: [Optional context]
Answer: [Model generates response here]Examples
Prompt: Instruction: List three benefits of cloud computing. Input: Answer:
Output: The model will provide a structured response listing three benefits of cloud computing.
Limitations
- Works best with instruction-formatted prompts
- May require specific generation parameters to avoid repetition
- Optimized for English instruction-following tasks
Training Details
- Framework: Transformers + PEFT
- Optimizer: AdamW
- Learning Rate: 1e-4
- Batch Size: 16 (effective)
- Epochs: 2
- Hardware: NVIDIA A100
Citation
@misc{gpt-neo-dolly15k-lora-2025,
title={GPT-Neo 1.3B LoRA Fine-tuned on Dolly-15k},
author={Your Name},
year={2025},
publisher={Hugging Face},
url={https://huggingface.co/YOUR_USERNAME/gpt-neo-1.3b-dolly15k-lora}
}License
This model is released under the Apache 2.0 License, following the base model's license.
