infer.py

import torch
from optimum.neuron import NeuronModelForCausalLM
from transformers import AutoTokenizer
model_id="deepseek-ai/DeepSeek-R1-Distill-Llama-70B"
prompt="Who are you? what is the model that powers you?"
tokenizer = AutoTokenizer.from_pretrained(model_id)
tokenizer.pad_token_id = tokenizer.eos_token_id if tokenizer.pad_token_id is None else tokenizer.pad_token_id
inputs = tokenizer(prompt, return_tensors="pt")
model = NeuronModelForCausalLM.from_pretrained("yahavb/DeepSeek-R1-Distill-Llama-70B-Neuron")
for i in range(10):
  outputs = model.generate(**inputs,max_new_tokens=512,do_sample=True,use_cache=True,temperature=0.7,top_k=50,top_p=0.9)
  outputs=outputs[0, inputs.input_ids.size(-1):]
  response=tokenizer.decode(outputs, skip_special_tokens=True)
  print(response)