|
import torch |
|
import torch.distributed as dist |
|
import torch.multiprocessing as mp |
|
from transformers import AutoTokenizer, LlamaForCausalLM |
|
from torch.nn.parallel import DistributedDataParallel as DDP |
|
from evalplus.data import get_human_eval_plus, write_jsonl |
|
import os |
|
from tqdm import tqdm |
|
|
|
def setup(rank, world_size): |
|
os.environ['MASTER_ADDR'] = 'localhost' |
|
os.environ['MASTER_PORT'] = '12355' |
|
dist.init_process_group("gloo", rank=rank, world_size=world_size) |
|
|
|
def cleanup(): |
|
dist.destroy_process_group() |
|
|
|
def generate_one_completion(ddp_model, tokenizer, prompt: str): |
|
tokenizer.pad_token = tokenizer.eos_token |
|
inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=4096) |
|
|
|
|
|
generate_ids = ddp_model.module.generate(inputs.input_ids.to("cuda"), max_new_tokens=384, do_sample=True, top_p=0.75, top_k=40, temperature=0.1, pad_token_id=tokenizer.eos_token_id) |
|
completion = tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0] |
|
completion = completion.replace(prompt, "").split("\n\n\n")[0] |
|
|
|
print("-------------------") |
|
print(completion) |
|
return completion |
|
|
|
def run(rank, world_size): |
|
setup(rank, world_size) |
|
|
|
model_path = "Nondzu/Mistral-7B-codealpaca-lora" |
|
model = LlamaForCausalLM.from_pretrained(model_path,load_in_8bit=True) |
|
ddp_model = DDP(model, device_ids=[rank]) |
|
tokenizer = AutoTokenizer.from_pretrained(model_path) |
|
|
|
problems = get_human_eval_plus() |
|
num_samples_per_task = 1 |
|
|
|
samples = [ |
|
dict(task_id=task_id, completion=generate_one_completion(ddp_model, tokenizer, problems[task_id]["prompt"])) |
|
for task_id in tqdm(problems) |
|
for _ in range(num_samples_per_task) |
|
] |
|
write_jsonl(f"samples-Nondzu-Mistral-7B-codealpaca-lora-rank{rank}.jsonl", samples) |
|
|
|
cleanup() |
|
|
|
def main(): |
|
world_size = 1 |
|
mp.spawn(run, args=(world_size,), nprocs=world_size, join=True) |
|
|
|
if __name__=="__main__": |
|
main() |
|
|