File size: 1,645 Bytes
9711fd0
 
0d2677e
 
 
 
 
 
 
0fb9a51
 
9711fd0
 
 
0fb9a51
9711fd0
 
 
0fb9a51
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9711fd0
0fb9a51
 
 
9711fd0
 
 
0fb9a51
 
 
 
 
9711fd0
0fb9a51
 
9711fd0
0fb9a51
9711fd0
0fb9a51
 
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
---
library_name: transformers
tags:
- medical
license: apache-2.0
language:
- fr
- en
base_model:
- ik-ram28/MedMistralInstruct-CPT-7B
- mistralai/Mistral-7B-Instruct-v0.1
---


## MedMistralInstruct-CPT-SFT-7B

### Model Description

MedMistralInstruct-CPT-SFT-7B is a French medical language model based on Mistral-7B-Instruct-v0.1, adapted through Continual Pre-Training followed by Supervised Fine-Tuning. 

### Model Details

- **Model Type**: Causal Language Model
- **Base Model**: Mistral-7B-Instruct-v0.1
- **Language**: French
- **Domain**: Medical/Healthcare
- **Parameters**: 7 billion
- **License**: Apache 2.0

### Training Details

**Continual Pre-Training (CPT)**
- **Dataset**: NACHOS corpus (7.4 GB French medical texts)
- **Training Duration**: 2.8 epochs
- **Hardware**: 32 NVIDIA A100 80GB GPUs
- **Training Time**: ~40 hours

**Supervised Fine-Tuning (SFT)**
- **Dataset**: 30K French medical question-answer pairs
- **Method**: DoRA (Weight-Decomposed Low-Rank Adaptation)
- **Training Duration**: 10 epochs
- **Hardware**: 1 NVIDIA H100 80GB GPU
- **Training Time**: ~42 hours


### Computational Requirements
- **Carbon Emissions**: 33.96 kgCO2e (CPT+SFT)
- **Training Time**: 82 hours total (CPT+SFT)



### Ethical Considerations
- **Medical Accuracy**: For research and educational purposes only
- **Professional Oversight**: Requires verification by qualified medical professionals
- **Bias Awareness**: May contain biases from training data
- **Privacy**: Do not input private health information

### Citation
```bibtex

```

### Contact
For questions about these models, please contact: ikram.belmadani@lis-lab.fr