File size: 1,491 Bytes
27fc147 d853c93 17a1aea d853c93 c15613f f907d22 c15613f d853c93 27fc147 |
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 |
---
library_name: peft
base_model: facebook/esm2_t33_650M_UR50D
---
# Model Card for Model ID
This model builds upon [PepMLM](https://github.com/programmablebio/pepmlm/tree/main), aimed at generating peptides from receptor sequences. It incorporates the [ESM model](https://huggingface.co/docs/transformers/model_doc/esm) framework from HuggingFace for its core architecture. The key enhancement in this model is the adoption of LoRA for training, distinguishing it from its predecessor.
Usage:
```
from transformers import AutoTokenizer, AutoModelForMaskedLM
from peft import PeftConfig
import torch
model_name = "littleworth/esm2_t33_650M_UR50D_pepmlm_lora_adapter_merged"
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForMaskedLM.from_pretrained(model_name).to(device)
```
Packages version:
```
{'transformers': '4.36.0', 'peft': '0.9.0', 'torch': '2.0.0'}
```
Training summary:
```
{
"train/loss": 1.5091,
"train/grad_norm": 3.6427412033081055,
"train/learning_rate": 6.773224309612687e-7,
"train/epoch": 5,
"train/global_step": 6395,
"_timestamp": 1709229361.5373268,
"_runtime": 25556.57973074913,
"_step": 639,
"train/train_runtime": 25557.6176,
"train/train_samples_per_second": 4.003,
"train/train_steps_per_second": 0.25,
"train/total_flos": 220903283526564960,
"train/train_loss": 1.8436848362317955,
"_wandb": {
"runtime": 25556
}
}
```
|