Hugging Face's logo

vibhorag101
/
llama-2-7b-chat-hf-phr_mental_therapy

Text Generation
Transformers
PyTorch
English
llama
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
3
llama-2-7b-chat-hf-phr_mental_therapy / README.md
vibhorag101's picture
vibhorag101
Update README.md
430e4c4 verified
preview code
|
raw
history blame
3.02 kB
metadata
license: mit
datasets:
  - vibhorag101/phr_mental_therapy_dataset
  - jerryjalapeno/nart-100k-synthetic
language:
  - en
pipeline_tag: text-generation

Model Card

  • This model is a finetune of the llama-2-7b-chat-hf model on a therapy dataset.
  • The model aims to provide basic therapy to the users and improve their mental health until they seek professional help.
  • The model has been adjusted to encourage giving cheerful responses to the user. The system prompt has been mentioned below.

Model Details

Training Hardware

  • RTX A5000 24GB
  • 48 Core Intel Xeon
  • 128GB RAM.

Model Hyperparameters

  • This training script was used to do the finetuning.
  • The shareGPT format dataset was converted to llama-2 training format using this script.
  • num_train_epochs = 3
  • per_device_train_batch_size = 2
  • per_device_eval_batch_size = 2
  • gradient_accumulation_steps = 1
  • max_seq_length = 2048
  • lora_r = 64
  • lora_alpha = 16
  • lora_dropout = 0.1
  • use_4bit = True
  • bnb_4bit_compute_dtype = "float16"
  • bnb_4bit_quant_type = "nf4"
  • use_nested_quant = False
  • fp16 = False
  • bf16 = True
  • Data Sample: 1000 (80:20 split)

Model System Prompt

You are a helpful and joyous mental therapy assistant. Always answer as helpfully and cheerfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content.Please ensure that your responses are socially unbiased and positive in nature.

If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information.

Model Training Data

image/png

Tokenizer Details

The tokenizer should be loaded as follows for correct inference:

tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True,use_fast=False,add_eos_token=True)
tokenizer.pad_token_id = 18610

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 42.84
ARC (25-shot) 52.39
HellaSwag (10-shot) 75.39
MMLU (5-shot) 39.77
TruthfulQA (0-shot) 42.89
Winogrande (5-shot) 71.19
GSM8K (5-shot) 5.91
DROP (3-shot) 12.3