README.md

metadata

license: apache-2.0

NeuroBLAST 1.9B Instruct Early Preview

WIP

Inference code

import torch 
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline 

torch.random.manual_seed(0)

model = AutoModelForCausalLM.from_pretrained( 
    "meditsolutions/NeuroBLAST-1.9B-Instruct-Early-Preview",  
    device_map="cuda",  
    torch_dtype=torch.bfloat16,  
    trust_remote_code=True,  
) 

tokenizer = AutoTokenizer.from_pretrained("meditsolutions/NeuroBLAST-1.9B-Instruct-Early-Preview") 

messages = [ 
    {"role": "user", "content": "What is enalapril?"}, 
] 

pipe = pipeline( 
    "text-generation", 
    model=model, 
    tokenizer=tokenizer, 
) 

generation_args = { 
    "max_new_tokens": 500, 
    "return_full_text": False, 
    "temperature": 0.0, 
    "do_sample": False, 
} 

output = pipe(messages, **generation_args) 
print(output[0]['generated_text'])