Model Card for Llama-3.2-1B-Instruct Fine-Tuned with LoRA Weights
This model is a fine-tuned version of the "meta-llama/Llama-3.2-1B-Instruct" using LoRA (Low-Rank Adaptation) weights. It was trained to assist in answering questions and providing information on a range of topics. The model is designed to be used with the 🤗 Hugging Face transformers library
Model Details
Model Description
This model is based on the Llama-3.2-1B-Instruct architecture and has been fine-tuned with LoRA weights to improve its performance on specific downstream tasks. It was trained on a carefully selected dataset to enable more focused and contextual responses. The model is designed to perform well in environments where GPU resources may be limited, using optimizations like FP16 and device mapping.
- Developed by: Soorya R
- Model type: Causal Language Model with LoRA fine-tuning
- Language(s) (NLP): Primarily English
- License: Model card does not specify a particular license; check the base model's license on Hugging Face for usage guidelines.
- Finetuned from model: meta-llama/Llama-3.2-1B-Instruct
Model Sources [optional]
Uses
Direct Use
This model can be directly used for general-purpose question-answering and information retrieval tasks in English. It is suitable for chatbots and virtual assistants and performs well in scenarios where contextual responses are important.
Downstream Use
The model may also be further fine-tuned for specific tasks that require conversational understanding and natural language generation.
Out-of-Scope Use
This model is not suitable for tasks outside of general-purpose NLP. It should not be used for high-stakes decision-making, tasks requiring detailed scientific or legal knowledge, or applications that could impact user safety or privacy.
Bias, Risks, and Limitations
This model was fine-tuned on a curated dataset but still inherits biases from the underlying Llama model. Users should be cautious about using it in sensitive or biased contexts, as the model may inadvertently produce outputs that reflect biases present in the training data.
Recommendations
Users (both direct and downstream) should be made aware of the potential risks and limitations of the model, including biases in language or domain limitations. More robust evaluation is recommended before deployment in critical applications.
How to Get Started with the Model
Use the code below to get started with the model.
from transformers import AutoModelForCausalLM, AutoTokenizer
Load the model
model = AutoModelForCausalLM.from_pretrained("Soorya03/Llama-3.2-1B-Instruct-LoRA")
Load the tokenizer
tokenizer = AutoTokenizer.from_pretrained("Soorya03/Llama-3.2-1B-Instruct-LoRA")
Generate text
inputs = tokenizer("Your input text here", return_tensors="pt") outputs = model.generate(**inputs) print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Training Details
Training Data
The model was fine-tuned on a custom dataset, optimized for contextual question-answering tasks and general-purpose conversational use. The dataset was split into training and validation sets to enhance model generalization.
Training Procedure
Training Hyperparameters
Precision: FP16 mixed precision Epochs: 10 Batch size: 4 Learning rate: 2e-4
Times
Training time: Approximately 1 hour on Google Colab's T4 GPU.
Model Examination
For interpretability, tools like transformers's pipeline can help visualize the model's attention mechanisms and interpret its outputs. However, users should be aware that this is a black-box model.
Environmental Impact
Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).
- Hardware Type: Google Colab T4 GPU
- Hours used: 1
- Cloud Provider: Google Colab
Technical Specifications
Model Architecture and Objective
The model follows the Llama architecture, which is a transformer-based model designed for NLP tasks. The objective of fine-tuning with LoRA weights was to enhance contextual understanding and response accuracy.
Compute Infrastructure
Hardware
Google Colab T4 GPU with FP16 precision enabled
Software
Library: 🤗 Hugging Face transformers Framework: PyTorch Other dependencies: PEFT library for LoRA weights integration
Citation [optional]
@misc{soorya2024llama, author = {Soorya R}, title = {Llama-3.2-1B-Instruct Fine-Tuned with LoRA Weights}, year = {2024}, url = {https://huggingface.co/Soorya03/Llama-3.2-1B-Instruct-LoRA}, }
Glossary
FP16: 16-bit floating point precision, used to reduce memory usage and speed up computation. LoRA: Low-Rank Adaptation, a method for parameter-efficient fine-tuning.
More Information [optional]
For more details, please visit the model repository.
Model Card Authors
Soorya R