98 lines
2.4 KiB
Markdown
98 lines
2.4 KiB
Markdown
|
---
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
- fr
|
||
|
|
- nl
|
||
|
|
- es
|
||
|
|
- it
|
||
|
|
- pl
|
||
|
|
- ro
|
||
|
|
- de
|
||
|
|
license: apache-2.0
|
||
|
|
library_name: transformers
|
||
|
|
tags:
|
||
|
|
- mergekit
|
||
|
|
- merge
|
||
|
|
- dare
|
||
|
|
- medical
|
||
|
|
- biology
|
||
|
|
- mlx
|
||
|
|
datasets:
|
||
|
|
- health_fact
|
||
|
|
base_model:
|
||
|
|
- BioMistral/BioMistral-7B
|
||
|
|
- mistralai/Mistral-7B-Instruct-v0.1
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
---
|
||
|
|
# abhishek-ch/biomistral-7b-synthetic-ehr
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
This model was converted to MLX format from [`BioMistral/BioMistral-7B-DARE`]().
|
||
|
|
Refer to the [original model card](https://huggingface.co/BioMistral/BioMistral-7B-DARE) for more details on the model.
|
||
|
|
|
||
|
|
|
||
|
|
## Use with mlx
|
||
|
|
|
||
|
|
```bash
|
||
|
|
pip install mlx-lm
|
||
|
|
```
|
||
|
|
|
||
|
|
The model was LoRA fine-tuned on [health_facts](https://huggingface.co/datasets/health_fact) and
|
||
|
|
Synthetic EHR dataset inspired by MIMIC-IV using the format below, for 1000 steps (~1M tokens) using mlx.
|
||
|
|
|
||
|
|
```python
|
||
|
|
def format_prompt(prompt:str, question: str) -> str:
|
||
|
|
return """<s>[INST]
|
||
|
|
## Instructions
|
||
|
|
{}
|
||
|
|
## User Question
|
||
|
|
{}.
|
||
|
|
[/INST]</s>
|
||
|
|
""".format(prompt, question)
|
||
|
|
```
|
||
|
|
|
||
|
|
Example For Synthetic EHR Diagnosis System Prompt
|
||
|
|
```
|
||
|
|
You are an expert in provide diagnosis summary based on clinical notes inspired by MIMIC-IV-Note dataset.
|
||
|
|
These notes encompass Chief Complaint along with Patient Summary & medical admission details.
|
||
|
|
```
|
||
|
|
|
||
|
|
Example for Healthfacts Check System Prompt
|
||
|
|
```
|
||
|
|
You are a Public Health AI Assistant. You can do the fact-checking of public health claims. \nEach answer labelled with true, false, unproven or mixture. \nPlease provide the reason behind the answer
|
||
|
|
```
|
||
|
|
|
||
|
|
## Loading the model using `mlx`
|
||
|
|
|
||
|
|
```python
|
||
|
|
from mlx_lm import generate, load
|
||
|
|
model, tokenizer = load("abhishek-ch/biomistral-7b-synthetic-ehr")
|
||
|
|
response = generate(
|
||
|
|
fused_model,
|
||
|
|
fused_tokenizer,
|
||
|
|
prompt=format_prompt(prompt, question),
|
||
|
|
verbose=True, # Set to True to see the prompt and response
|
||
|
|
temp=0.0,
|
||
|
|
max_tokens=512,
|
||
|
|
)
|
||
|
|
```
|
||
|
|
|
||
|
|
## Loading the model using `transformers`
|
||
|
|
|
||
|
|
```python
|
||
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||
|
|
repo_id = "abhishek-ch/biomistral-7b-synthetic-ehr"
|
||
|
|
tokenizer = AutoTokenizer.from_pretrained(repo_id)
|
||
|
|
model = AutoModelForCausalLM.from_pretrained(repo_id)
|
||
|
|
model.to("mps")
|
||
|
|
input_text = format_prompt(system_prompt, question)
|
||
|
|
input_ids = tokenizer(input_text, return_tensors="pt").to("mps")
|
||
|
|
outputs = model.generate(
|
||
|
|
**input_ids,
|
||
|
|
max_new_tokens=512,
|
||
|
|
)
|
||
|
|
print(tokenizer.decode(outputs[0]))
|
||
|
|
```
|