File size: 2,846 Bytes
01b0dd4
 
 
e7a58b9
 
 
 
 
 
01b0dd4
e7a58b9
01b0dd4
e7a58b9
01b0dd4
e7a58b9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
01b0dd4
 
668fa97
 
 
 
 
 
 
 
 
 
 
 
 
 
01b0dd4
 
 
 
 
 
 
8a17009
 
01b0dd4
ea33e69
01b0dd4
 
941c84d
 
e7a58b9
668fa97
 
 
 
 
e7a58b9
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
---
license: cc-by-4.0
tags:
- health
- physical-activity
- behavior-change
- stage-of-change
- llm
- transformers
datasets:
- SriyaM/MHC_LLM_Preference_Data
language:
- en
model-index:
- name: MHC-Coach
  results:
  - task:
      name: Health Coaching
      type: health-coaching
    dataset:
      name: MyHeartCounts Coaching Data
      type: SriyaM/MHC_LLM_Preference_Data
    metrics:
    - name: Human Preference Rate
      type: preference
      value: 68
    - name: Expert Effectiveness Score
      type: likert
      value: 4.4
    - name: Stage Match Score
      type: likert
      value: 4.1
base_model:
- meta-llama/Meta-Llama-3-70B
pipeline_tag: text-generation
---

# MHC-Coach: A Behaviorally Informed Health Coaching Language Model

**MHC-Coach** is a large language model fine-tuned on behavioral science principles to deliver stage-specific health coaching messages aimed at increasing physical activity. Built on LLaMA 3-70B, it integrates the Transtheoretical Model of Change and cardiovascular health content to provide motivational, personalized coaching aligned with each user’s behavioral readiness.

## Highlights
- Fine-tuned on 3,000+ human-expert health coaching messages
- Embeds psychological theory (Transtheoretical Model) for personalized messaging
- Evaluated on 632 users in the MyHeartCounts app
- Preferred over expert-written messages in 68% of matched-stage comparisons

## Reference
For more details, see the preprint:
> [Fine-tuning Large Language Models in Behavioral Psychology for Scalable Physical Activity Coaching](https://doi.org/10.1101/2025.02.19.25322559)

## Usage

You can load the model using Hugging Face's `transformers` library:

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("SriyaM/MHC-Coach")
model = AutoModelForCausalLM.from_pretrained("SriyaM/MHC-Coach")

prompt = "Write a 20-word notification to motivate someone in the Precontemplation stage of change to increase their exercise levels. In this stage, people do not intend to take action in the foreseeable future (defined as within the next 6 months). People are often unaware that their behavior is problematic or produces negative consequences."
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

## Citation
If you use this model, please cite:
@article{mantena2025mhccoach, title={Fine-tuning Large Language Models in Behavioral Psychology for Scalable Physical Activity Coaching}, author={Mantena, Sriya and Johnson, Anders and Oppezzo, Marily and Schuetz, Narayan and Tolas, Alexander and others}, journal={medRxiv}, year={2025}, doi={10.1101/2025.02.19.25322559} }

## License
[CC-BY 4.0](http://creativecommons.org/licenses/by/4.0/)