lyra_xverse/model.py

# Copyright (c) 2021-2022, 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.

from __future__ import print_function

import copy
import os
import pathlib
import typing

import numpy as np
import torch
import torch.distributed as dist
import torch.nn as nn

str_type_map = {"fp32": torch.float32, "fp16": torch.float16, "bf16": torch.bfloat16}

class XVERSEModel(nn.Module):
    def __init__(self,
                 head_num, 
                 size_per_head, 
                 inter_size,
                 vocab_size, 
                 rotary_embedding_dim,
                 start_id, end_id, layer_num,
                 max_seq_len: int, 
                 layernorm_eps,
                 tensor_para_size: int, 
                 pipeline_para_size: int,
                 use_gptj_residual,
                 lib_path: typing.Union[str, pathlib.Path],
                 model_path,
                 memopt_mode: int = 1,
                 inference_data_type: str = "fp16",
                 weights_data_type: typing.Union[str, np.dtype] = np.float32):
        super().__init__()
        self.head_num = head_num
        self.size_per_head = size_per_head
        self.inter_size = inter_size
        self.vocab_size = vocab_size
        self.rotary_embedding_dim = rotary_embedding_dim
        self.start_id = start_id
        self.end_id = end_id
        self.max_seq_len = max_seq_len
        self.layer_num = layer_num
        self.use_gptj_residual = use_gptj_residual
        self.layernorm_eps = layernorm_eps
        self.memopt_mode = memopt_mode

        # multi-gpu params
        self.tensor_para_size = tensor_para_size
        self.pipeline_para_size = pipeline_para_size
        self.build_model = False
        self.weights_data_type = weights_data_type
        self.inference_data_type = inference_data_type

        assert torch.cuda.is_available(), "CUDA is required for this model."

        assert head_num % tensor_para_size == 0, "head_num must be a multiple of tensor_para_size."
        assert layer_num % pipeline_para_size == 0, "layer_num must be a multiple of pipeline_para_size."

        # Load the C++ model into Pytorch model.
        torch.classes.load_library(os.path.abspath(lib_path))

        # Prepare for tensor/pipeline parallel
        try:
            dist.init_process_group(backend='mpi')
        except:
            print("[INFO] WARNING: Have initialized the process group")
        self.rank = dist.get_rank()
        self.device_count = torch.cuda.device_count()
        self.device = self.rank % self.device_count
        torch.cuda.set_device(self.device)

        world_size = dist.get_world_size()
        # print(tensor_para_size * pipeline_para_size)
        assert world_size == tensor_para_size * pipeline_para_size, "tensor_para_size * pipeline_para_size must be equal to world_size."

        self.tensor_para_rank = self.rank % self.tensor_para_size
        self.pipeline_para_rank = self.rank // self.tensor_para_size

        self.model = torch.classes.FasterTransformer.LlamaOp(
            self.head_num, self.size_per_head, self.inter_size,
            self.layer_num,
            self.vocab_size,
            self.rotary_embedding_dim,
            self.layernorm_eps,
            self.start_id, self.end_id,
            self.tensor_para_size, self.pipeline_para_size,
            self.max_seq_len,
            self.use_gptj_residual,
            self.memopt_mode,
            model_path,
            self.weights_data_type,
            self.inference_data_type)
        
        self.build_model = True
        torch.cuda.empty_cache()

    def forward(self,
                start_ids: torch.Tensor,
                start_lengths: torch.Tensor,
                output_len,
                beam_width=1,
                top_k: torch.Tensor = None,
                top_p: torch.Tensor = None,
                beam_search_diversity_rate: torch.Tensor = None,
                temperature: torch.Tensor = None,
                len_penalty: torch.Tensor = None,
                repetition_penalty: torch.Tensor = None,
                random_seed: torch.Tensor = None,
                return_output_length=False,
                return_cum_log_probs=0):

        input_len = start_ids.size(1)
        assert input_len > 0, "input len must be larger than zero. For an unconditional case, use start_id as the first token."

        # Inputs to device
        input_ids = start_ids.cuda(self.device)
        input_lengths = start_lengths.cuda(self.device)
        # outputs: output_ids, output_lengths, output_cum_log_probs (optional)
        outputs = self.model.forward(input_ids,
                                     input_lengths,
                                     output_len,
                                     beam_width,  # optional, can be None
                                     top_k,  # optional, can be None
                                     top_p,  # optional, can be None
                                     beam_search_diversity_rate,  # optional, can be None
                                     temperature,  # optional, can be None
                                     len_penalty,  # optional, can be None
                                     repetition_penalty,  # optional, can be None
                                     random_seed,  # optional, can be None
                                     return_cum_log_probs)  # optional, can be None

        if return_cum_log_probs == 0:
            output_ids, output_lengths = outputs
        else:
            output_ids, output_lengths, output_cum_log_probs = outputs
        if return_output_length:
            if return_cum_log_probs > 0:
                return output_ids, output_lengths, output_cum_log_probs
            else:
                return output_ids, output_lengths
        else:
            return output_ids

    def set_input_tensor(self, input_tensor):
        """Set input tensor to be used instead of forward()'s input.

        When doing pipeline parallelism the input from the previous
        stage comes from communication, not from the input, so the
        model's forward_step_func won't have it. This function is thus
        used by internal code to bypass the input provided by the
        forward_step_func"""
        self.input_tensor = input_tensor