Modelpublic

GRMenon/mental-health-mistral-7b-instructv0.2-finetuned-V2

sourceHugging Faceapache-2.0updated 3y agoView on Hugging Face

40likes13downloads

Model Card

mental-health-mistral-7b-instructv0.2-finetuned-V2

This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.2 on the mental_health_counseling_conversations dataset. It achieves the following results on the evaluation set:

—Loss: 0.6432

Model description

A Mistral-7B-Instruct-v0.2 model finetuned on a corpus of mental health conversations between a psychologist and a user. The intention was to create a mental health assistant, "Connor", to address user questions based on responses from a psychologist.

Training and evaluation data

The model is finetuned on a corpus of mental health conversations between a psychologist and a client, in the form of context - response pairs. This dataset is a collection of questions and answers sourced from two online counseling and therapy platforms. The questions cover a wide range of mental health topics, and the answers are provided by qualified psychologists. Dataset found here :-

—Kaggle
—Huggingface

Training hyperparameters

The following hyperparameters were used during training:

—learning_rate: 0.0002
—trainbatchsize: 8
—evalbatchsize: 8
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: cosine
—lrschedulerwarmup_ratio: 0.05
—num_epochs: 3
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
1.4325	1.0	352	0.9064
1.2608	2.0	704	0.6956
1.1845	3.0	1056	0.6432

Usage

python

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftConfig, PeftModel

base_model = "mistralai/Mistral-7B-Instruct-v0.2"
adapter = "GRMenon/mental-health-mistral-7b-instructv0.2-finetuned-V2"

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(
    base_model,
    add_bos_token=True,
    trust_remote_code=True,
    padding_side='left'
)

# Create peft model using base_model and finetuned adapter
config = PeftConfig.from_pretrained(adapter)
model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path,
                                             load_in_4bit=True,
                                             device_map='auto',
                                             torch_dtype='auto')
model = PeftModel.from_pretrained(model, adapter)

device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)
model.eval()

# Prompt content:
messages = [
    {"role": "user", "content": "Hey Connor! I have been feeling a bit down lately.I could really use some advice on how to feel better?"}
]

input_ids = tokenizer.apply_chat_template(conversation=messages,
                                          tokenize=True,
                                          add_generation_prompt=True,
                                          return_tensors='pt').to(device)
output_ids = model.generate(input_ids=input_ids, max_new_tokens=512, do_sample=True, pad_token_id=2)
response = tokenizer.batch_decode(output_ids.detach().cpu().numpy(), skip_special_tokens = True)

# Model response: 
print(response[0])

Framework versions

—PEFT 0.7.1
—Transformers 4.36.1
—Pytorch 2.0.0
—Datasets 2.1.0
—Tokenizers 0.15.0