MindWell

MindWell is a chat assistant trained by fine-tuning lmsys/vicuna-7b-v1.5 on two symptom-based depression datasets: BDI-Sen and PsySym. To expedite and optimize the fine-tuning process, we have implemented Low Rank Adaptation (LoRA) techniques, ensuring enhanced efficiency and faster adaptation.

How to Get Started with MindWell

Use the code below to get started with the model.

from peft import PeftModel
from transformers import AutoModelForCausalLM, LlamaTokenizer, GenerationConfig

config = PeftConfig.from_pretrained("irlab-udc/MindWell")
model = AutoModelForCausalLM.from_pretrained("lmsys/vicuna-13b-v1.5", device_map="auto")
model = PeftModel.from_pretrained(model, "irlab-udc/MindWell")
tokenizer = AutoTokenizer.from_pretrained("lmsys/vicuna-13b-v1.5")

def evaluate(input):
  inputs = tokenizer(input, return_tensors="pt")
  input_ids = inputs["input_ids"].cuda()
  generation_output = model.generate(
      input_ids=input_ids,
      generation_config=GenerationConfig(do_sample=True),
      return_dict_in_generate=True,
      output_scores=True,
      max_new_tokens=512,
  )
  for s in generation_output.sequences:
    output = tokenizer.decode(s)
    print("Answer:", output)

evaluate("What can you do?")

Answer: I can analyze the user's comments to determine if they exhibit any signs of depressive symptoms based on the provided list of symptoms. I will justify my decisions by means of excerpts from the user's comments. If I don't know the answer, I will truthfully say that I don't know.

Training

Configurations and Hyperparameters

The following LoraConfig config was used during training:

r: 8
lora_alpha: 16
target_modules: ["q_proj", "v_proj"]
lora_dropout: 0.05
bias: "none"
task_type: "CAUSAL_LM"

The following TrainingArguments config was used during training:

per_device_train_batch_size: 64
gradient_accumulation_steps: 32
warmup_steps: 100
num_train_epochs: 20
learning_rate: 3e-4
fp16=True

The following bitsandbytes quantization config was used during training:

quant_method: bitsandbytes
load_in_8bit: True
load_in_4bit: False
llm_int8_threshold: 6.0
llm_int8_skip_modules: None
llm_int8_enable_fp32_cpu_offload: False
llm_int8_has_fp16_weight: False
bnb_4bit_quant_type: fp4
bnb_4bit_use_double_quant: False
bnb_4bit_compute_dtype: float32

Framework versions

PyTorch 2.1.0
PEFT 0.5.0
🤗 Transformers 4.34.0
🤗 Datasets 2.14.5
🤗 Tokenizers 0.14.0

Environmental Impact

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

Hardware Type: NVIDIA RTX 6000 Ada Generation.
Hours used: 20.
Cloud Provider: Private infrastructure.
Carbon Emitted: 2.59 Kg. CO2 eq.

irlab-udc
/

MindWell