RaphaelMourad
commited on
Commit
•
a638111
1
Parent(s):
474f148
Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,61 @@
|
|
1 |
---
|
2 |
-
license:
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
license: apache-2.0
|
3 |
+
tags:
|
4 |
+
- pretrained
|
5 |
+
- mistral
|
6 |
+
- DNA
|
7 |
+
- biology
|
8 |
+
- genomics
|
9 |
---
|
10 |
+
|
11 |
+
# Model Card for mixtral-dna-yeast-v0.2 (mistral for DNA)
|
12 |
+
|
13 |
+
The mixtral-dna-yeast-v0.2 Large Language Model (LLM) is a pretrained generative DNA text model with 17.31M parameters x 8 experts = 138.5M parameters.
|
14 |
+
It is derived from Mistral-7B-v0.1 model, which was simplified for DNA: the number of layers and the hidden size were reduced.
|
15 |
+
The model was pretrained using around 1000 yeast genomes with 10kb DNA sequences.
|
16 |
+
|
17 |
+
The yeast genomes are from: https://www.nature.com/articles/s41586-018-0030-5
|
18 |
+
|
19 |
+
For full details of this model please read our [github repo](https://github.com/raphaelmourad/Mistral-DNA).
|
20 |
+
|
21 |
+
## Model Architecture
|
22 |
+
|
23 |
+
Like Mistral-7B-v0.1, it is a transformer model, with the following architecture choices:
|
24 |
+
- Grouped-Query Attention
|
25 |
+
- Sliding-Window Attention
|
26 |
+
- Byte-fallback BPE tokenizer
|
27 |
+
|
28 |
+
## Load the model from huggingface:
|
29 |
+
|
30 |
+
```
|
31 |
+
import torch
|
32 |
+
from transformers import AutoTokenizer, AutoModel
|
33 |
+
|
34 |
+
tokenizer = AutoTokenizer.from_pretrained("RaphaelMourad/mixtral-dna-yeast-v0.2", trust_remote_code=True) # Same as DNABERT2
|
35 |
+
model = AutoModel.from_pretrained("RaphaelMourad/mixtral-dna-yeast-v0.2", trust_remote_code=True)
|
36 |
+
```
|
37 |
+
|
38 |
+
## Calculate the embedding of a DNA sequence
|
39 |
+
|
40 |
+
```
|
41 |
+
dna = "TGATGATTGGCGCGGCTAGGATCGGCT"
|
42 |
+
inputs = tokenizer(dna, return_tensors = 'pt')["input_ids"]
|
43 |
+
hidden_states = model(inputs)[0] # [1, sequence_length, 256]
|
44 |
+
|
45 |
+
# embedding with max pooling
|
46 |
+
embedding_max = torch.max(hidden_states[0], dim=0)[0]
|
47 |
+
print(embedding_max.shape) # expect to be 256
|
48 |
+
```
|
49 |
+
|
50 |
+
## Troubleshooting
|
51 |
+
|
52 |
+
Ensure you are utilizing a stable version of Transformers, 4.34.0 or newer.
|
53 |
+
|
54 |
+
## Notice
|
55 |
+
|
56 |
+
Mistral-DNA is a pretrained base model for DNA.
|
57 |
+
|
58 |
+
## Contact
|
59 |
+
|
60 |
+
Raphaël Mourad. [email protected]
|
61 |
+
|