generate_text.py

# coding=utf-8
# Copyright (c) 2020, NVIDIA CORPORATION.  All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

"""Sample Generate GPT"""
import os
import sys
sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__),
                                             os.path.pardir)))
from megatron import get_args
from megatron import print_rank_0
from megatron import mpu
from megatron.checkpointing import load_checkpoint
from megatron.initialize import initialize_megatron
from megatron.model import GPTModel
from megatron.training import get_model
from megatron.text_generation import generate_and_post_process

import numpy as np
import time
import torch

def model_provider(pre_process=True, post_process=True):
    """Build the model."""

    print_rank_0('building GPT model ...')
    model = GPTModel(num_tokentypes=0, parallel_output=False, pre_process=pre_process, post_process=post_process)

    return model

def add_text_generate_args(parser):
    group = parser.add_argument_group(title='text generation')

    group.add_argument("--prompts-file", type=str, required=True,
                       help='File with prompt (that can then be truncated)')
    group.add_argument("--all-num-input-tokens", type=int, nargs='+',
                       help='Number of input tokens (must be > 0)')
    group.add_argument("--all-num-output-tokens", type=int, nargs='+',
                       help='Number of tokens to generate')
    group.add_argument("--temperature", type=float, default=1.0,
                       help='Sampling temperature.')
    group.add_argument("--top-p", type=float, default=0.0,
                       help='Top p sampling.')
    group.add_argument("--top-k", type=int, default=0,
                       help='Top k sampling.')
    return parser


if __name__ == "__main__":
    initialize_megatron(extra_args_provider=add_text_generate_args,
                        args_defaults={'tokenizer_type': 'GPT2BPETokenizer',
                                       'no_load_rng': True,
                                       'no_load_optim': True})

    args = get_args()
    if args.num_layers_per_virtual_pipeline_stage is not None:
        print("Interleaved pipeline schedule is not yet supported for text generation.")
        exit()
    # Set up model and load checkpoint
    model = get_model(model_provider, wrap_with_ddp=False)

    if args.load is not None:
        _ = load_checkpoint(model, None, None)

    assert len(model) == 1, "Above condition should have caught this"
    model = model[0]

    num_warmup_iterations = 10
    num_iterations = 50

    for num_input_tokens in args.all_num_input_tokens:
        with open(args.prompts_file, 'r') as f:
            prompts = [f.read()[:10*num_input_tokens]]  # Hack to prevent tokenization from taking too long.
        for num_output_tokens in args.all_num_output_tokens:
            times = []
            last_start_time = time.time()
            for i in range(num_warmup_iterations + num_iterations):
                if i == num_warmup_iterations:
                    start_time = time.time()
                output = generate_and_post_process(model, prompts=prompts,
                                                   num_input_tokens=num_input_tokens,
                                                   tokens_to_generate=num_output_tokens,
                                                   temperature=args.temperature,
                                                   top_k_sampling=args.top_k,
                                                   top_p_sampling=args.top_p,
                                                   use_eod_token_for_early_termination=False)
                times.append(time.time() - last_start_time)
                last_start_time = time.time()
            runtime = (time.time() - start_time) / num_iterations

            if torch.distributed.get_rank() == 0:
                label = {'num_input_tokens': num_input_tokens, 'num_output_tokens': num_output_tokens}
                print(f"Runtime for {label}: {runtime:.3f} seconds +- {np.std(times[num_warmup_iterations:]):.3f}")