Llama-2-7b-FitnessAssistant

Model Card for Model ID

<!-- Provide a quick summary of what the model is/does. -->

Model Details

Model Description

This model is a fine-tuned version of Llama-2-7b-chat, optimized for tasks related to fitness assistance. It has been trained to provide recommendations, answer questions, and perform related language-based tasks within the fitness and exercise domain.

• Developed by: Soorya03
• Finetuned from model: meta-llama/Llama-2-7b-chat-hf
• Model Type: Causal Language Model with LoRA fine-tuning
• Language(s): English
• License: Refer to the original model’s license
• Model Repository: Soorya03/Llama-2-7b-chat-finetune

Uses

Direct Use

This model is intended for interactive fitness and exercise assistance, such as providing exercise recommendations, suggesting workout routines, and answering general fitness-related questions.

Downstream Use

May be adapted to various other fitness or health-oriented conversational applications.

Out-of-Scope Use

Not suitable for medical or professional health advice. Avoid use cases where specialized knowledge or regulated health guidelines are required.

Bias, Risks, and Limitations

• Potential Bias: The model was fine-tuned on a limited dataset and might not cover all fitness-related questions with cultural or demographic sensitivity.
• Limitations: Not a replacement for professional medical advice.

Recommendations

Users should be aware that the model's responses are based on general fitness knowledge and are not specialized medical guidance.

How to Get Started with the Model

! pip install accelerate peft bitsandbytes transformers trl

import torch from transformers import BitsAndBytesConfig, AutoModelForCausalLM, AutoTokenizer, pipeline

device_map = {"": 0}

compute_dtype = getattr(torch, "float16")

bnb_config = BitsAndBytesConfig( load_in_4bit=True, bnb_4bit_quant_type="nf4", bnb_4bit_compute_dtype="float16", bnb_4bit_use_double_quant=False, )

model = AutoModelForCausalLM.from_pretrained( "Soorya03/Llama-2-7b-FitnessAssistant", quantization_config=bnb_config, device_map=device_map ) model.config.use_cache = False model.config.pretraining_tp = 1

tokenizer = AutoTokenizer.from_pretrained("Soorya03/Llama-2-7b-FitnessAssistant", trust_remote_code=True)

prompt = "Is it possible to build muscle while losing weight?" pipe = pipeline(task="text-generation", model=model, tokenizer=tokenizer, max_length=200) result = pipe(f"[INST] {prompt} [/INST]") print(result[0]['generated_text'])

Training Details

Training Data

The model was fine-tuned on a fitness and exercise dataset (onurSakar/GYM-Exercise) to improve its domain knowledge in providing fitness-related responses.

<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->

Training Procedure

<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->

• Method: LoRA fine-tuning on top of Llama-2-7b-chat.
• Hyperparameters: Adjusted learning rate, FP16 precision for efficiency.
• Compute: Training was performed on Google Colab with a single GPU.

Evaluation

<!-- This section describes the evaluation protocols and provides the results. -->

Testing Data & Metrics

Testing Data

Sample fitness-related prompts were used for evaluation, but a formal benchmarking dataset was not utilized.

Metrics

Manual qualitative assessments showed the model’s suitability for fitness Q&A and general suggestions.

Results

The model effectively generates coherent responses related to fitness, workouts, and exercise routines, with accurate language comprehension.

Environmental Impact

Compute Infrastructure

• Hardware Type: Google Colab (NVIDIA GPU)

Model Architecture and Objective

This model is based on the Llama-2-7b-chat architecture, adapted to provide conversational responses within a specific fitness domain.