from optimum.intel.openvino import OVModelForCausalLM
from transformers import pipeline
from huggingface_hub import create_repo, upload_folder
import torch
import transformers
import os

model_id = 'mistralai/Mixtral-8x7B-v0.1'
save_path = '/tmp/yujiepan/mixtral-8xtiny-random'
repo_id = 'yujiepan/mixtral-8xtiny-random'

config = transformers.AutoConfig.from_pretrained(model_id)
config.hidden_size = 8
config.intermediate_size = 32
config.num_attention_heads = 4
config.num_experts_per_tok = 2
config.num_hidden_layers = 2
config.num_key_value_heads = 2
config.num_local_experts = 8
print(config)

tokenizer = transformers.AutoTokenizer.from_pretrained(model_id)
tokenizer.save_pretrained(save_path)

model = transformers.AutoModelForCausalLM.from_config(config, torch_dtype=torch.float16)
model = model.half()

pipe = pipeline('text-generation', model=model, tokenizer=tokenizer, do_sample=False, device='cuda')
print(pipe('Hello World!'))

model.save_pretrained(save_path)

# ovmodel = OVModelForCausalLM.from_pretrained(save_path, export=True)
# ovmodel = ovmodel.half()
# ovmodel.save_pretrained(save_path)

os.system(f'ls -alh /tmp/yujiepan/mixtral-8xtiny-random')
create_repo(repo_id, exist_ok=True)
upload_folder(repo_id=repo_id, folder_path=save_path)