Default template and configuration for local run with GPU

#33
by brunoedcf - opened

I have a template like this one:

  • Is it possible to run mistral-7b using a similar template?
  • What pytorch + cuda do I need to install?
  • What dependencies are necessary?

from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained( )

pipeline = pipeline(
"text-generation",
model= ,
torch_dtype=torch.float16,
device_map="auto",
eos_token_id=tokenizer.eos_token_id,
pad_token_id=tokenizer.eos_token_id,
)

prompt = "prompt"

sequences = []

sequences.append(
pipeline(
prompt,
top_k=10,
max_new_tokens=150,
num_return_sequences=1,
)
)

for sequence in sequences:
generated_text = sequence[0]['generated_text']

print(generated_text)

Sign up or log in to comment