compile.py

from huggingface_hub.hf_api import HfFolder
from huggingface_hub import login
from optimum.neuron import NeuronModelForCausalLM
import os

hf_token="hf_TBD"
sequence_length=256
auto_cast_type="bf16"
batch_size=8
num_cores=16
hf_repo="yahavb/DeepSeek-R1-Distill-Llama-70B-Neuron"
model_dir="/deepseek-ai/DeepSeek-R1-Distill-Llama-70B"
model_id="deepseek-ai/DeepSeek-R1-Distill-Llama-70B"

from huggingface_hub.hf_api import HfFolder
from huggingface_hub import login
from optimum.neuron import NeuronModelForCausalLM

login(hf_token,add_to_git_credential=True)

compiler_args = {"num_cores": num_cores, "auto_cast_type": auto_cast_type}
input_shapes = {"batch_size": batch_size, "sequence_length": sequence_length}
model = NeuronModelForCausalLM.from_pretrained(
        model_id,
        export=True,
        **compiler_args,
        **input_shapes)
model.save_pretrained(model_dir)
model.push_to_hub(model_dir,repository_id=hf_repo)