Commit
·
b2175f4
1
Parent(s):
8158986
readme
Browse files
README.md
CHANGED
@@ -1,9 +1,47 @@
|
|
1 |
---
|
2 |
library_name: transformers
|
3 |
-
tags:
|
|
|
|
|
|
|
|
|
4 |
---
|
5 |
|
6 |
# DNA and Block Diffusion
|
7 |
|
8 |
-
Untrained architecture test using the Block Diffusion architecture and
|
9 |
-
AgroNT's six-nucleotide-length tokens.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
library_name: transformers
|
3 |
+
tags:
|
4 |
+
- biology
|
5 |
+
- bd3lm
|
6 |
+
license: apache-2.0
|
7 |
+
base_model: kuleshov-group/bd3lm-owt-block_size1024-pretrain
|
8 |
---
|
9 |
|
10 |
# DNA and Block Diffusion
|
11 |
|
12 |
+
Untrained architecture test using the [Block Diffusion](https://github.com/kuleshov-group/bd3lms) architecture and
|
13 |
+
[AgroNT](https://huggingface.co/InstaDeepAI/agro-nucleotide-transformer-1b)'s six-nucleotide-length tokens.
|
14 |
+
|
15 |
+
### Loading model
|
16 |
+
|
17 |
+
```python
|
18 |
+
from transformers import AutoModelForMaskedLM
|
19 |
+
m = AutoModelForMaskedLM.from_pretrained(
|
20 |
+
"monsoon-nlp/dna-blockdiff",
|
21 |
+
trust_remote_code=True,
|
22 |
+
)
|
23 |
+
```
|
24 |
+
|
25 |
+
### Generating text
|
26 |
+
|
27 |
+
```bash
|
28 |
+
cd bd3lms
|
29 |
+
python -u main.py \
|
30 |
+
loader.eval_batch_size=1 \
|
31 |
+
model=small \
|
32 |
+
algo=bd3lm \
|
33 |
+
algo.T=900 \
|
34 |
+
algo.backbone=hf_dit \
|
35 |
+
algo.sampler=analytic \
|
36 |
+
data=openwebtext-split \
|
37 |
+
model.length=2048 \
|
38 |
+
block_size=4 \
|
39 |
+
wandb=null \
|
40 |
+
mode=sample_eval \
|
41 |
+
eval.checkpoint_path="monsoon-nlp/dna-blockdiff" \
|
42 |
+
model.attn_backend=sdpa \
|
43 |
+
sampling.nucleus_p=0.9 \
|
44 |
+
sampling.kv_cache=true \
|
45 |
+
sampling.logdir=$PWD/sample_logs/samples_genlen_bd3lm_blocksize4 \
|
46 |
+
data.tokenizer_name_or_path="monsoon-nlp/dna-blockdiff"
|
47 |
+
```
|