Instructions to use goodarzilab/decodon-200M-euk with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use goodarzilab/decodon-200M-euk with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="goodarzilab/decodon-200M-euk", trust_remote_code=True)

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("goodarzilab/decodon-200M-euk", trust_remote_code=True, dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use goodarzilab/decodon-200M-euk with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "goodarzilab/decodon-200M-euk"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "goodarzilab/decodon-200M-euk",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/goodarzilab/decodon-200M-euk

SGLang

How to use goodarzilab/decodon-200M-euk with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "goodarzilab/decodon-200M-euk" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "goodarzilab/decodon-200M-euk",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "goodarzilab/decodon-200M-euk" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "goodarzilab/decodon-200M-euk",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use goodarzilab/decodon-200M-euk with Docker Model Runner:
```
docker model run hf.co/goodarzilab/decodon-200M-euk
```

Get started 🚀

Installation

From source

Currently, this is the only way to install the package but will push a pip installable version soon. To install the package from source, run the following command:

pip install git+https://github.com/goodarzilab/cdsFM.git

Applications

Now that you have cdsFM installed, you can use AutoEnCodon and AutoDeCodon classes which serve as wrappers around the pre-trained models. Here are some examples on how to use them:

Sequence Embedding Extraction with EnCodon

Following is an example of how to use the EnCodon model to extract sequence embeddings:

from cdsFM import AutoEnCodon

# Load your dataframe containing sequences
seqs = ...

# Load a pre-trained EnCodon model
model = AutoEnCodon.from_pretrained("goodarzilab/encodon-620M")

# Extract embeddings
embeddings = model.get_embeddings(seqs, batch_size=32)

Sequence Generation with DeCodon

You can generate organism-specific coding sequences with DeCodon simply by:

from cdsFM import AutoDeCodon

# Load a pre-trained DeCodon model
model = AutoDeCodon.from_pretrained("goodarzilab/DeCodon-200M")

# Generate!
gen_seqs = model.generate(
    taxid=9606, # NCBI Taxonomy ID for Homo sapiens
    num_return_sequences=32, # Number of sequences to return
    max_length=1024, # Maximum length of the generated sequence
    batch_size=8, # Batch size for generation
)

Tokenization

EnCodon and DeCodon are pre-trained on coding sequences of length up to 2048 codons (i.e. 6144 nucleotides), including the <CLS> token prepended automatically to the beginning of the sequence and the <SEP> token appended at the end. The tokenizer's vocabulary consists of 64 codons and 5 special tokens namely <CLS>, <SEP>, <PAD>, <MASK> and <UNK>.

HuggingFace 🤗

A collection of pre-trained checkpoints of EnCodon & DeCodon models are available on HuggingFace 🤗. Following table contains the list of available models:

Model	name	num. params	description	weights
EnCodon	encodon-80M	80M	Pre-trained checkpoint	🤗
EnCodon	encodon-80M-euk	80M	Eukaryotic-expert	🤗
EnCodon	encodon-620M	620M	Pre-trained checkpoint	🤗
EnCodon	encodon-620M-euk	620M	Eukaryotic-expert	🤗
DeCodon	decodon-200M	200M	Pre-trained checkpoint	🤗
DeCodon	decodon-200M-euk	200M	Eukaryotic-expert	🤗

Citation

@article{Naghipourfar2024,
  title = {A Suite of Foundation Models Captures the Contextual Interplay Between Codons},
  url = {http://dx.doi.org/10.1101/2024.10.10.617568},
  DOI = {10.1101/2024.10.10.617568},
  publisher = {Cold Spring Harbor Laboratory},
  author = {Naghipourfar,  Mohsen and Chen,  Siyu and Howard,  Mathew and Macdonald,  Christian and Saberi,  Ali and Hagen,  Timo and Mofrad,  Mohammad and Coyote-Maestas,  Willow and Goodarzi,  Hani},
  year = {2024},
  month = oct 
}

Downloads last month: 9

Safetensors

Model size

0.2B params

Tensor type

F32

Collection including goodarzilab/decodon-200M-euk

cdsFM

Collection

Suite of codon foundation models • 6 items • Updated Nov 14, 2024