BioGPT

BioGPT is a generative Transformer model based on GPT-2 and pretrained on 15 million PubMed abstracts. It is designed for biomedical language tasks.

You can find all the original BioGPT checkpoints under the Microsoft organization.

Click on the BioGPT models in the right sidebar for more examples of how to apply BioGPT to different language tasks.

The example below demonstrates how to generate biomedical text with [Pipeline], [AutoModel], and also from the command line.

import torch
from transformers import pipeline

generator = pipeline(
    task="text-generation",
    model="microsoft/biogpt",
    torch_dtype=torch.float16,
    device=0,
)
result = generator("Ibuprofen is best used for", truncation=True, max_length=50, do_sample=True)[0]["generated_text"]
print(result)

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("microsoft/biogpt")
model = AutoModelForCausalLM.from_pretrained(
    "microsoft/biogpt",
    torch_dtype=torch.float16,
    device_map="auto",
    attn_implementation="sdpa"
)

input_text = "Ibuprofen is best used for"
inputs = tokenizer(input_text, return_tensors="pt").to(model.device)

with torch.no_grad():
    generated_ids = model.generate(**inputs, max_length=50)
    
output = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
print(output)

echo -e "Ibuprofen is best used for" | transformers-cli run --task text-generation --model microsoft/biogpt --device 0

Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the Quantization overview for more available quantization backends.

The example below uses bitsandbytes to only quantize the weights to 4-bit precision.

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_use_double_quant=True
)

tokenizer = AutoTokenizer.from_pretrained("microsoft/BioGPT-Large")
model = AutoModelForCausalLM.from_pretrained(
    "microsoft/BioGPT-Large", 
    quantization_config=bnb_config,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

input_text = "Ibuprofen is best used for"
inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
with torch.no_grad():
    generated_ids = model.generate(**inputs, max_length=50)    
output = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
print(output)

Notes

Pad inputs on the right because BioGPT uses absolute position embeddings.
BioGPT can reuse previously computed key-value attention pairs. Access this feature with the past_key_values parameter in [BioGPTModel.forward].

The head_mask argument is ignored when using an attention implementation other than "eager". If you want to use head_mask, make sure attn_implementation="eager").

from transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(
   "microsoft/biogpt",
   attn_implementation="eager"
)

BioGptConfig

[[autodoc]] BioGptConfig

BioGptTokenizer

[[autodoc]] BioGptTokenizer - save_vocabulary

BioGptModel

[[autodoc]] BioGptModel - forward

BioGptForCausalLM

[[autodoc]] BioGptForCausalLM - forward

BioGptForTokenClassification

[[autodoc]] BioGptForTokenClassification - forward

BioGptForSequenceClassification

[[autodoc]] BioGptForSequenceClassification - forward