Skip to content

GitLab

"examples/vscode:/vscode.git/clone" did not exist on "ec0866804c3f028f73d4e4d0bc1f3309362c4e89"
Unverified Commit cb7ec714 authored by Frank Lee's avatar Frank Lee Committed by GitHub
Browse files

[tutorial] removed duplicated tutorials (#1904)

parent 351f0f64
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 0 additions and 1488 deletions
examples/tutorial/handson5/inference/opt_server.py deleted 100644 → 0
View file @ 351f0f64
import logging
import argparse
import random
from torch import Tensor
from pydantic import BaseModel, Field
from typing import Optional
from energonai.model import opt_125M, opt_30B, opt_175B, opt_6B
from transformers import GPT2Tokenizer
from energonai import launch_engine, QueueFullError
from sanic import Sanic
from sanic.request import Request
from sanic.response import json
from sanic_ext import validate, openapi
from batch import BatchManagerForGeneration
from cache import ListCache, MissCacheError
class GenerationTaskReq(BaseModel):
max_tokens: int = Field(gt=0, le=256, example=64)
prompt: str = Field(
min_length=1, example='Question: Where were the 2004 Olympics held?\nAnswer: Athens, Greece\n\nQuestion: What is the longest river on the earth?\nAnswer:')
top_k: Optional[int] = Field(default=None, gt=0, example=50)
top_p: Optional[float] = Field(default=None, gt=0.0, lt=1.0, example=0.5)
temperature: Optional[float] = Field(default=None, gt=0.0, lt=1.0, example=0.7)
app = Sanic('opt')
@app.post('/generation')
@openapi.body(GenerationTaskReq)
@validate(json=GenerationTaskReq)
async def generate(request: Request, body: GenerationTaskReq):
logger.info(f'{request.ip}:{request.port} - "{request.method} {request.path}" - {body}')
key = (body.prompt, body.max_tokens)
try:
if cache is None:
raise MissCacheError()
outputs = cache.get(key)
output = random.choice(outputs)
logger.info('Cache hit')
except MissCacheError:
inputs = tokenizer(body.prompt, truncation=True, max_length=512)
inputs['max_tokens'] = body.max_tokens
inputs['top_k'] = body.top_k
inputs['top_p'] = body.top_p
inputs['temperature'] = body.temperature
try:
uid = id(body)
engine.submit(uid, inputs)
output = await engine.wait(uid)
assert isinstance(output, Tensor)
output = tokenizer.decode(output, skip_special_tokens=True)
if cache is not None:
cache.add(key, output)
except QueueFullError as e:
return json({'detail': e.args[0]}, status=406)
return json({'text': output})
@app.after_server_stop
def shutdown(*_):
engine.shutdown()
def get_model_fn(model_name: str):
model_map = {
'opt-125m': opt_125M,
'opt-6.7b': opt_6B,
'opt-30b': opt_30B,
'opt-175b': opt_175B
}
return model_map[model_name]
def print_args(args: argparse.Namespace):
print('\n==> Args:')
for k, v in args.__dict__.items():
print(f'{k} = {v}')
FIXED_CACHE_KEYS = [
('Question: What is the name of the largest continent on earth?\nAnswer: Asia\n\nQuestion: What is at the center of the solar system?\nAnswer:', 64),
('A chat between a salesman and a student.\n\nSalesman: Hi boy, are you looking for a new phone?\nStudent: Yes, my phone is not functioning well.\nSalesman: What is your budget? \nStudent: I have received my scholarship so I am fine with any phone.\nSalesman: Great, then perhaps this latest flagship phone is just right for you.', 64),
("English: I am happy today.\nChinese: 我今天很开心。\n\nEnglish: I am going to play basketball.\nChinese: 我一会去打篮球。\n\nEnglish: Let's celebrate our anniversary.\nChinese:", 64)
]
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('model', choices=['opt-125m', 'opt-6.7b', 'opt-30b', 'opt-175b'])
parser.add_argument('--tp', type=int, default=1)
parser.add_argument('--master_host', default='localhost')
parser.add_argument('--master_port', type=int, default=19990)
parser.add_argument('--rpc_port', type=int, default=19980)
parser.add_argument('--max_batch_size', type=int, default=8)
parser.add_argument('--pipe_size', type=int, default=1)
parser.add_argument('--queue_size', type=int, default=0)
parser.add_argument('--http_host', default='0.0.0.0')
parser.add_argument('--http_port', type=int, default=7070)
parser.add_argument('--checkpoint', default=None)
parser.add_argument('--cache_size', type=int, default=0)
parser.add_argument('--cache_list_size', type=int, default=1)
args = parser.parse_args()
print_args(args)
model_kwargs = {}
if args.checkpoint is not None:
model_kwargs['checkpoint'] = args.checkpoint
logger = logging.getLogger(__name__)
tokenizer = GPT2Tokenizer.from_pretrained('facebook/opt-30b')
if args.cache_size > 0:
cache = ListCache(args.cache_size, args.cache_list_size, fixed_keys=FIXED_CACHE_KEYS)
else:
cache = None
engine = launch_engine(args.tp, 1, args.master_host, args.master_port, args.rpc_port, get_model_fn(args.model),
batch_manager=BatchManagerForGeneration(max_batch_size=args.max_batch_size,
pad_token_id=tokenizer.pad_token_id),
pipe_size=args.pipe_size,
queue_size=args.queue_size,
**model_kwargs)
app.run(args.http_host, args.http_port)
examples/tutorial/handson5/inference/requirements.txt deleted 100644 → 0
View file @ 351f0f64
fastapi==0.85.1
locust==2.11.0
pydantic==1.10.2
sanic==22.9.0
sanic_ext==22.9.0
torch>=1.10.0
transformers==4.23.1
uvicorn==0.19.0
examples/tutorial/handson5/inference/script/process-opt-175b/README.md deleted 100644 → 0
View file @ 351f0f64
# Process OPT-175B weights
You should download the pre-trained weights following the [doc](https://github.com/facebookresearch/metaseq/tree/main/projects/OPT) before reading this.
First, install `metaseq` and `git clone https://github.com/facebookresearch/metaseq.git`.
Then, `cd metaseq`.
To consolidate checkpoints to eliminate FSDP:
```shell
bash metaseq/scripts/reshard_mp_launch_no_slurm.sh <directory_where_all_the_shards_are>/checkpoint_last <output_dir>/ 8 1
```
You will get 8 files in `<output_dir>`, and you should have the following checksums:
```
7e71cb65c4be784aa0b2889ac6039ee8 reshard-model_part-0-shard0.pt
c8123da04f2c25a9026ea3224d5d5022 reshard-model_part-1-shard0.pt
45e5d10896382e5bc4a7064fcafd2b1e reshard-model_part-2-shard0.pt
abb7296c4d2fc17420b84ca74fc3ce64 reshard-model_part-3-shard0.pt
05dcc7ac6046f4d3f90b3d1068e6da15 reshard-model_part-4-shard0.pt
d24dd334019060ce1ee7e625fcf6b4bd reshard-model_part-5-shard0.pt
fb1615ce0bbe89cc717f3e5079ee2655 reshard-model_part-6-shard0.pt
2f3124432d2dbc6aebfca06be4b791c2 reshard-model_part-7-shard0.pt
```
Copy `flat-meta.json` to `<output_dir>`.
Then cd to this dir, and we unflatten parameters.
```shell
bash unflat.sh <output_dir>/ <new_output_dir>/
```
Finally, you will get 8 files in `<new_output_dir>` with following checksums:
```
6169c59d014be95553c89ec01b8abb62 reshard-model_part-0.pt
58868105da3d74a528a548fdb3a8cff6 reshard-model_part-1.pt
69b255dc5a49d0eba9e4b60432cda90b reshard-model_part-2.pt
002c052461ff9ffb0cdac3d5906f41f2 reshard-model_part-3.pt
6d57f72909320d511ffd5f1c668b2beb reshard-model_part-4.pt
93c8c4041cdc0c7907cc7afcf15cec2a reshard-model_part-5.pt
5d63b8750d827a1aa7c8ae5b02a3a2ca reshard-model_part-6.pt
f888bd41e009096804fe9a4b48c7ffe8 reshard-model_part-7.pt
```
examples/tutorial/handson5/inference/script/process-opt-175b/convert_ckpt.py deleted 100644 → 0
View file @ 351f0f64
import argparse
import json
import os
import re
from collections import defaultdict
import numpy as np
import torch
def load_json(path: str):
with open(path) as f:
return json.load(f)
def parse_shape_info(flat_dir: str):
data = load_json(os.path.join(flat_dir, 'shape.json'))
flat_info = defaultdict(lambda: defaultdict(list))
for k, shape in data.items():
matched = re.match(r'decoder.layers.\d+', k)
if matched is None:
flat_key = 'flat_param_0'
else:
flat_key = f'{matched[0]}.flat_param_0'
flat_info[flat_key]['names'].append(k)
flat_info[flat_key]['shapes'].append(shape)
flat_info[flat_key]['numels'].append(int(np.prod(shape)))
return flat_info
def convert(flat_dir: str, output_dir: str, part: int):
flat_path = os.path.join(flat_dir, f'reshard-model_part-{part}-shard0.pt')
output_path = os.path.join(output_dir, f'reshard-model_part-{part}.pt')
flat_meta = load_json(os.path.join(flat_dir, 'flat-meta.json'))
flat_sd = torch.load(flat_path)
print(f'Loaded flat state dict from {flat_path}')
output_sd = {}
for flat_key, param_meta in flat_meta.items():
flat_param = flat_sd['model'][flat_key]
assert sum(param_meta['numels']) == flat_param.numel(
), f'flat {flat_key} {flat_param.numel()} vs {sum(param_meta["numels"])}'
for name, shape, param in zip(param_meta['names'], param_meta['shapes'], flat_param.split(param_meta['numels'])):
output_sd[name] = param.view(shape)
torch.save(output_sd, output_path)
print(f'Saved unflat state dict to {output_path}')
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('flat_dir')
parser.add_argument('output_dir')
parser.add_argument('part', type=int)
args = parser.parse_args()
convert(args.flat_dir, args.output_dir, args.part)
examples/tutorial/handson5/inference/script/process-opt-175b/unflat.sh deleted 100644 → 0
View file @ 351f0f64
#!/usr/bin/env sh
for i in $(seq 0 7); do
python convert_ckpt.py $1 $2 ${i} &
done
wait $(jobs -p)
examples/tutorial/handson5/inference/script/processing_ckpt_66b.py deleted 100644 → 0
View file @ 351f0f64
import os
import torch
from multiprocessing import Pool
# download pytorch model ckpt in https://huggingface.co/facebook/opt-66b/tree/main
# you can use whether wget or git lfs
path = "/path/to/your/ckpt"
new_path = "/path/to/the/processed/ckpt/"
assert os.path.isdir(path)
files = []
for filename in os.listdir(path):
filepath = os.path.join(path, filename)
if os.path.isfile(filepath):
files.append(filepath)
with Pool(14) as pool:
ckpts = pool.map(torch.load, files)
restored = {}
for ckpt in ckpts:
for k,v in ckpt.items():
if(k[0] == 'm'):
k = k[6:]
if(k == "lm_head.weight"):
k = "head.dense.weight"
if(k == "decoder.final_layer_norm.weight"):
k = "decoder.layer_norm.weight"
if(k == "decoder.final_layer_norm.bias"):
k = "decoder.layer_norm.bias"
restored[k] = v
restored["decoder.version"] = "0.0"
split_num = len(restored.keys()) // 60
count = 0
file_count = 1
tmp = {}
for k,v in restored.items():
print(k)
tmp[k] = v
count = count + 1
if(count == split_num):
filename = str(file_count) + "-restored.pt"
torch.save(tmp, os.path.join(new_path, filename))
file_count = file_count + 1
count = 0
tmp = {}
filename = str(file_count) + "-restored.pt"
torch.save(tmp, os.path.join(new_path, filename))
examples/tutorial/handson5/opt/README.md deleted 100644 → 0
View file @ 351f0f64
<!---
Copyright 2020 The HuggingFace Team. 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.
-->
# Train OPT model with Colossal-AI
## OPT
Meta recently released [Open Pretrained Transformer (OPT)](https://github.com/facebookresearch/metaseq), a 175-Billion parameter AI language model, which stimulates AI programmers to perform various downstream tasks and application deployments.
The following example of [Colossal-AI](https://github.com/hpcaitech/ColossalAI) demonstrates fine-tuning Casual Language Modelling at low cost.
We are using the pre-training weights of the OPT model provided by Hugging Face Hub on the raw WikiText-2 (no tokens were replaced before
the tokenization). This training script is adapted from the [HuggingFace Language Modelling examples](https://github.com/huggingface/transformers/tree/main/examples/pytorch/language-modeling).
## Our Modifications
We adapt the OPT training code to ColossalAI by leveraging Gemini and ZeRO DDP.
## Quick Start
You can launch training by using the following bash script
```bash
bash ./run_clm.sh <batch-size-per-gpu> <mem-cap> <model> <gpu-num>
```
- batch-size-per-gpu: number of samples fed to each GPU, default is 16
- mem-cap: limit memory usage within a value in GB, default is 0 (no limit)
- model: the size of the OPT model, default is `6.7b`. Acceptable values include `125m`, `350m`, `1.3b`, `2.7b`, `6.7`, `13b`, `30b`, `66b`. For `175b`, you can request
the pretrained weights from [OPT weight downloading page](https://github.com/facebookresearch/metaseq/tree/main/projects/OPT).
- gpu-num: the number of GPUs to use, default is 1.
## Remarkable Performance
On a single GPU, Colossal-AI’s automatic strategy provides remarkable performance gains from the ZeRO Offloading strategy by Microsoft DeepSpeed.
Users can experience up to a 40% speedup, at a variety of model scales. However, when using a traditional deep learning training framework like PyTorch, a single GPU can no longer support the training of models at such a scale.
<p align="center">
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/OPT.png" width=1000/>
</p>
Adopting the distributed training strategy with 8 GPUs is as simple as adding a `-nprocs 8` to the training command of Colossal-AI!
More details about behind the scenes can be found on the corresponding [blog](https://medium.com/@yangyou_berkeley/colossal-ai-seamlessly-accelerates-large-models-at-low-costs-with-hugging-face-4d1a887e500d),
and a detailed tutorial will be added in [Documentation](https://www.colossalai.org/docs/get_started/installation) very soon.
examples/tutorial/handson5/opt/benchmark.sh deleted 100644 → 0
View file @ 351f0f64
export BS=16
export MEMCAP=0
export MODEL="6.7b"
export GPUNUM=1
for MODEL in "6.7b" "13b" "1.3b"
do
for GPUNUM in 8 1
do
for BS in 16 24 32 8
do
for MEMCAP in 0 40
do
pkill -9 torchrun
pkill -9 python
bash ./run_clm.sh $BS $MEMCAP $MODEL $GPUNUM
done
done
done
done
examples/tutorial/handson5/opt/colossalai_zero.py deleted 100644 → 0
View file @ 351f0f64
from colossalai.zero.shard_utils import TensorShardStrategy
zero = dict(model_config=dict(shard_strategy=TensorShardStrategy(),
tensor_placement_policy="auto",
reuse_fp16_shard=True),
optimizer_config=dict(gpu_margin_mem_ratio=0.8, initial_scale=16384))
examples/tutorial/handson5/opt/context.py deleted 100644 → 0
View file @ 351f0f64
import torch.distributed as dist
from colossalai.context import ParallelMode
from colossalai.core import global_context as gpc
class barrier_context():
"""
This context manager is used to allow one process to execute while blocking all
other processes in the same process group. This is often useful when downloading is required
as we only want to download in one process to prevent file corruption.
Args:
executor_rank (int): the process rank to execute without blocking, all other processes will be blocked
parallel_mode (ParallelMode): the parallel mode corresponding to a process group
Usage:
with barrier_context():
dataset = CIFAR10(root='./data', download=True)
"""
def __init__(self, executor_rank: int = 0, parallel_mode: ParallelMode = ParallelMode.GLOBAL):
# the class name is lowercase by convention
current_rank = gpc.get_local_rank(parallel_mode=parallel_mode)
self.should_block = current_rank != executor_rank
self.group = gpc.get_group(parallel_mode=parallel_mode)
def __enter__(self):
if self.should_block:
dist.barrier(group=self.group)
def __exit__(self, exc_type, exc_value, exc_traceback):
if not self.should_block:
dist.barrier(group=self.group)
examples/tutorial/handson5/opt/requirements.txt deleted 100644 → 0
View file @ 351f0f64
colossalai
torch >= 1.8.1
datasets >= 1.8.0
sentencepiece != 0.1.92
protobuf
accelerate == 0.13.2
examples/tutorial/handson5/opt/run_clm.py deleted 100755 → 0
View file @ 351f0f64
This diff is collapsed.
examples/tutorial/handson5/opt/run_clm.sh deleted 100644 → 0
View file @ 351f0f64
set -x
export BS=${1:-16}
export MEMCAP=${2:-0}
export MODEL=${3:-"125m"}
export GPUNUM=${4:-1}
# make directory for logs
mkdir -p ./logs
export MODLE_PATH="facebook/opt-${MODEL}"
# HF_DATASETS_OFFLINE=1 TRANSFORMERS_OFFLINE=1
torchrun \
--nproc_per_node ${GPUNUM} \
--master_port 19198 \
run_clm.py \
--dataset_name wikitext \
--dataset_config_name wikitext-2-raw-v1 \
--output_dir $PWD \
--mem_cap ${MEMCAP} \
--model_name_or_path ${MODLE_PATH} \
--per_device_train_batch_size ${BS} 2>&1 | tee ./logs/colo_${MODEL}_bs_${BS}_cap_${MEMCAP}_gpu_${GPUNUM}.log
examples/tutorial/handson5/zero/README.md deleted 100644 → 0
View file @ 351f0f64
## Overview
This example shows how to use ColossalAI to run huggingface GPT training with Gemini and ZeRO DDP.
## GPT
We use the huggingface transformers GPT2 model. The input data is randonly generated.
## Our Modifications
We adapt the OPT training code to ColossalAI by leveraging Gemini and ZeRO DDP.
## Quick Start
You can launch training by using the following bash script
```bash
pip install -r requirements.txt
bash run.sh
```
examples/tutorial/handson5/zero/requirements.txt deleted 100644 → 0
View file @ 351f0f64
colossalai >= 0.1.10
torch >= 1.8.1
transformers >= 4.231
examples/tutorial/handson5/zero/run.sh deleted 100644 → 0
View file @ 351f0f64
env OMP_NUM_THREADS=16 torchrun --standalone --nproc_per_node=4 train_gpt_demo.py --tp_degree=2 --placement='cpu' 2>&1 | tee run.log
examples/tutorial/handson5/zero/train_gpt_demo.py deleted 100644 → 0
View file @ 351f0f64
from functools import partial
from time import time
import psutil
import torch
import torch.nn as nn
from packaging import version
import colossalai
from colossalai.logging import disable_existing_loggers, get_dist_logger
from colossalai.nn.optimizer import HybridAdam
from colossalai.nn.parallel import ZeroDDP
from colossalai.tensor import ColoParameter, ComputePattern, ComputeSpec, ProcessGroup, ShardSpec
from colossalai.utils import get_current_device
from colossalai.utils.model.colo_init_context import ColoInitContext
from colossalai.zero import ZeroOptimizer
from transformers import GPT2Config, GPT2LMHeadModel
def parse_args():
parser = colossalai.get_default_parser()
parser.add_argument(
"--tp_degree",
type=int,
default=1,
help="Tensor Parallelism Degree.",
)
parser.add_argument(
"--placement",
type=str,
default='cpu',
help="Placement Policy for Gemini.",
)
args = parser.parse_args()
return args
## Parameter Sharding Strategies for Tensor Parallelism
def split_param_single_dim_tp1d(dim: int, param: ColoParameter, pg: ProcessGroup):
spec = (ShardSpec([dim], [pg.tp_world_size()]), ComputeSpec(ComputePattern.TP1D))
if param.process_group.tp_world_size() == 1:
param.set_process_group(pg)
param.set_tensor_spec(*spec)
def split_param_row_tp1d(param: ColoParameter, pg: ProcessGroup):
split_param_single_dim_tp1d(0, param, pg)
def split_param_col_tp1d(param: ColoParameter, pg: ProcessGroup):
split_param_single_dim_tp1d(-1, param, pg)
## Define the Model and Loss Based on Huggingface transformers GPT2LMHeadModel
class GPTLMModel(nn.Module):
def __init__(self,
hidden_size=768,
num_layers=12,
num_attention_heads=12,
max_seq_len=1024,
vocab_size=50257,
checkpoint=False):
super().__init__()
self.checkpoint = checkpoint
self.model = GPT2LMHeadModel(
GPT2Config(n_embd=hidden_size,
n_layer=num_layers,
n_head=num_attention_heads,
n_positions=max_seq_len,
n_ctx=max_seq_len,
vocab_size=vocab_size))
if checkpoint:
self.model.gradient_checkpointing_enable()
def forward(self, input_ids, attention_mask):
# Only return lm_logits
return self.model(input_ids=input_ids, attention_mask=attention_mask, use_cache=not self.checkpoint)[0]
class GPTLMLoss(nn.Module):
def __init__(self):
super().__init__()
self.loss_fn = nn.CrossEntropyLoss()
def forward(self, logits, labels):
shift_logits = logits[..., :-1, :].contiguous()
shift_labels = labels[..., 1:].contiguous()
# Flatten the tokens
return self.loss_fn(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
## Randomly Generated Data
def get_data(batch_size, seq_len, vocab_size):
input_ids = torch.randint(0, vocab_size, (batch_size, seq_len), device=torch.cuda.current_device())
attention_mask = torch.ones_like(input_ids)
return input_ids, attention_mask
def gpt2_medium(checkpoint=False):
return GPTLMModel(hidden_size=1024, num_layers=24, num_attention_heads=16, checkpoint=checkpoint)
def gpt2_xl(checkpoint=True):
return GPTLMModel(hidden_size=1600, num_layers=48, num_attention_heads=32, checkpoint=checkpoint)
def gpt2_10b(checkpoint=True):
return GPTLMModel(hidden_size=4096, num_layers=50, num_attention_heads=16, checkpoint=checkpoint)
def get_cpu_mem():
return psutil.Process().memory_info().rss / 1024**2
def get_gpu_mem():
return torch.cuda.memory_allocated() / 1024**2
def get_mem_info(prefix=''):
return f'{prefix}GPU memory usage: {get_gpu_mem():.2f} MB, CPU memory usage: {get_cpu_mem():.2f} MB'
def get_tflops(model_numel, batch_size, seq_len, step_time):
return model_numel * batch_size * seq_len * 8 / 1e12 / (step_time + 1e-12)
# Tensor Parallel
def tensor_parallelize(model: torch.nn.Module, pg: ProcessGroup):
"""tensor_parallelize
Sharding the Model Parameters.
Args:
model (torch.nn.Module): a torch module to be sharded
"""
for mn, module in model.named_modules():
for pn, param in module.named_parameters(recurse=False):
# set process group for all parameters
param.set_process_group(pg)
if 'mlp.c_fc' in mn:
if 'weight' in pn or 'bias' in pn:
split_param_col_tp1d(param, pg) # colmn slice
# keep the shape of the output from c_fc
param.compute_spec.set_output_replicate(False)
elif 'mlp.c_proj' in mn:
if 'weight' in pn:
split_param_row_tp1d(param, pg) # row slice
elif 'wte' in mn or 'wpe' in mn:
split_param_col_tp1d(param, pg) # colmn slice
elif 'c_attn' in mn or 'c_proj' in mn:
split_param_col_tp1d(param, pg) # colmn slice
# Gemini + ZeRO DDP
def gemini_zero_dpp(model: torch.nn.Module, pg: ProcessGroup, placememt_policy: str = "auto"):
cai_version = colossalai.__version__
if version.parse(cai_version) > version.parse("0.1.10"):
from colossalai.nn.parallel import GeminiDDP
model = GeminiDDP(model,
device=get_current_device(),
placement_policy=placememt_policy,
pin_memory=True,
search_range_mb=32)
elif version.parse(cai_version) <= version.parse("0.1.10") and version.parse(cai_version) >= version.parse("0.1.9"):
from colossalai.gemini import ChunkManager, GeminiManager
chunk_size = ChunkManager.search_chunk_size(model, 64 * 1024**2, 32)
gemini_manager = GeminiManager(placememt_policy, chunk_manager)
chunk_manager = ChunkManager(chunk_size,
pg,
enable_distributed_storage=True,
init_device=GeminiManager.get_default_device(placememt_policy))
model = ZeroDDP(model, gemini_manager)
else:
raise NotImplemented(f"CAI version {cai_version} is not supported")
return model
def main():
args = parse_args()
BATCH_SIZE = 8
SEQ_LEN = 1024
VOCAB_SIZE = 50257
NUM_STEPS = 10
disable_existing_loggers()
colossalai.launch_from_torch(config={})
pg = ProcessGroup(tp_degree=args.tp_degree)
logger = get_dist_logger()
logger.info(get_mem_info(), ranks=[0])
# build GPT model
with ColoInitContext(device=get_current_device()):
model = gpt2_medium(checkpoint=True)
numel = sum([p.numel() for p in model.parameters()])
logger.info(f'Model numel: {numel}', ranks=[0])
get_tflops_func = partial(get_tflops, numel, BATCH_SIZE, SEQ_LEN)
# Tensor Parallelism (TP)
tensor_parallelize(model, pg)
# Gemini + ZeRO DP, Note it must be used after TP
model = gemini_zero_dpp(model, pg, args.placement)
logger.info(get_mem_info(prefix='After init model, '), ranks=[0])
# build criterion
criterion = GPTLMLoss()
# build optimizer
optimizer = HybridAdam(model.parameters(), lr=1e-3)
optimizer = ZeroOptimizer(optimizer, model, initial_scale=2**5)
logger.info(get_mem_info(prefix='After init optim, '), ranks=[0])
torch.cuda.synchronize()
model.train()
for n in range(NUM_STEPS):
# we just use randomly generated data here
input_ids, attn_mask = get_data(BATCH_SIZE, SEQ_LEN, VOCAB_SIZE)
optimizer.zero_grad()
start = time()
outputs = model(input_ids, attn_mask)
loss = criterion(outputs, input_ids)
logger.info(get_mem_info(prefix=f'[{n+1}/{NUM_STEPS}] Forward '), ranks=[0])
optimizer.backward(loss)
logger.info(get_mem_info(prefix=f'[{n+1}/{NUM_STEPS}] Backward '), ranks=[0])
optimizer.step()
logger.info(get_mem_info(prefix=f'[{n+1}/{NUM_STEPS}] Optimizer step '), ranks=[0])
step_time = time() - start
logger.info(
f'[{n+1}/{NUM_STEPS}] Loss:{loss.item():.3f}, Step time: {step_time:.3f}s, TFLOPS: {get_tflops_func(step_time):.3f}',
ranks=[0])
torch.cuda.synchronize()
if __name__ == '__main__':
main()
examples/tutorial/handson6/LICENSE deleted 100644 → 0
View file @ 351f0f64
Copyright (c) 2022 Robin Rombach and Patrick Esser and contributors
CreativeML Open RAIL-M
dated August 22, 2022
Section I: PREAMBLE
Multimodal generative models are being widely adopted and used, and have the potential to transform the way artists, among other individuals, conceive and benefit from AI or ML technologies as a tool for content creation.
Notwithstanding the current and potential benefits that these artifacts can bring to society at large, there are also concerns about potential misuses of them, either due to their technical limitations or ethical considerations.
In short, this license strives for both the open and responsible downstream use of the accompanying model. When it comes to the open character, we took inspiration from open source permissive licenses regarding the grant of IP rights. Referring to the downstream responsible use, we added use-based restrictions not permitting the use of the Model in very specific scenarios, in order for the licensor to be able to enforce the license in case potential misuses of the Model may occur. At the same time, we strive to promote open and responsible research on generative models for art and content generation.
Even though downstream derivative versions of the model could be released under different licensing terms, the latter will always have to include - at minimum - the same use-based restrictions as the ones in the original license (this license). We believe in the intersection between open and responsible AI development; thus, this License aims to strike a balance between both in order to enable responsible open-science in the field of AI.
This License governs the use of the model (and its derivatives) and is informed by the model card associated with the model.
NOW THEREFORE, You and Licensor agree as follows:
1. Definitions
- "License" means the terms and conditions for use, reproduction, and Distribution as defined in this document.
- "Data" means a collection of information and/or content extracted from the dataset used with the Model, including to train, pretrain, or otherwise evaluate the Model. The Data is not licensed under this License.
- "Output" means the results of operating a Model as embodied in informational content resulting therefrom.
- "Model" means any accompanying machine-learning based assemblies (including checkpoints), consisting of learnt weights, parameters (including optimizer states), corresponding to the model architecture as embodied in the Complementary Material, that have been trained or tuned, in whole or in part on the Data, using the Complementary Material.
- "Derivatives of the Model" means all modifications to the Model, works based on the Model, or any other model which is created or initialized by transfer of patterns of the weights, parameters, activations or output of the Model, to the other model, in order to cause the other model to perform similarly to the Model, including - but not limited to - distillation methods entailing the use of intermediate data representations or methods based on the generation of synthetic data by the Model for training the other model.
- "Complementary Material" means the accompanying source code and scripts used to define, run, load, benchmark or evaluate the Model, and used to prepare data for training or evaluation, if any. This includes any accompanying documentation, tutorials, examples, etc, if any.
- "Distribution" means any transmission, reproduction, publication or other sharing of the Model or Derivatives of the Model to a third party, including providing the Model as a hosted service made available by electronic or other remote means - e.g. API-based or web access.
- "Licensor" means the copyright owner or entity authorized by the copyright owner that is granting the License, including the persons or entities that may have rights in the Model and/or distributing the Model.
- "You" (or "Your") means an individual or Legal Entity exercising permissions granted by this License and/or making use of the Model for whichever purpose and in any field of use, including usage of the Model in an end-use application - e.g. chatbot, translator, image generator.
- "Third Parties" means individuals or legal entities that are not under common control with Licensor or You.
- "Contribution" means any work of authorship, including the original version of the Model and any modifications or additions to that Model or Derivatives of the Model thereof, that is intentionally submitted to Licensor for inclusion in the Model by the copyright owner or by an individual or Legal Entity authorized to submit on behalf of the copyright owner. For the purposes of this definition, "submitted" means any form of electronic, verbal, or written communication sent to the Licensor or its representatives, including but not limited to communication on electronic mailing lists, source code control systems, and issue tracking systems that are managed by, or on behalf of, the Licensor for the purpose of discussing and improving the Model, but excluding communication that is conspicuously marked or otherwise designated in writing by the copyright owner as "Not a Contribution."
- "Contributor" means Licensor and any individual or Legal Entity on behalf of whom a Contribution has been received by Licensor and subsequently incorporated within the Model.
Section II: INTELLECTUAL PROPERTY RIGHTS
Both copyright and patent grants apply to the Model, Derivatives of the Model and Complementary Material. The Model and Derivatives of the Model are subject to additional terms as described in Section III.
2. Grant of Copyright License. Subject to the terms and conditions of this License, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable copyright license to reproduce, prepare, publicly display, publicly perform, sublicense, and distribute the Complementary Material, the Model, and Derivatives of the Model.
3. Grant of Patent License. Subject to the terms and conditions of this License and where and as applicable, each Contributor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable (except as stated in this paragraph) patent license to make, have made, use, offer to sell, sell, import, and otherwise transfer the Model and the Complementary Material, where such license applies only to those patent claims licensable by such Contributor that are necessarily infringed by their Contribution(s) alone or by combination of their Contribution(s) with the Model to which such Contribution(s) was submitted. If You institute patent litigation against any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Model and/or Complementary Material or a Contribution incorporated within the Model and/or Complementary Material constitutes direct or contributory patent infringement, then any patent licenses granted to You under this License for the Model and/or Work shall terminate as of the date such litigation is asserted or filed.
Section III: CONDITIONS OF USAGE, DISTRIBUTION AND REDISTRIBUTION
4. Distribution and Redistribution. You may host for Third Party remote access purposes (e.g. software-as-a-service), reproduce and distribute copies of the Model or Derivatives of the Model thereof in any medium, with or without modifications, provided that You meet the following conditions:
Use-based restrictions as referenced in paragraph 5 MUST be included as an enforceable provision by You in any type of legal agreement (e.g. a license) governing the use and/or distribution of the Model or Derivatives of the Model, and You shall give notice to subsequent users You Distribute to, that the Model or Derivatives of the Model are subject to paragraph 5. This provision does not apply to the use of Complementary Material.
You must give any Third Party recipients of the Model or Derivatives of the Model a copy of this License;
You must cause any modified files to carry prominent notices stating that You changed the files;
You must retain all copyright, patent, trademark, and attribution notices excluding those notices that do not pertain to any part of the Model, Derivatives of the Model.
You may add Your own copyright statement to Your modifications and may provide additional or different license terms and conditions - respecting paragraph 4.a. - for use, reproduction, or Distribution of Your modifications, or for any such Derivatives of the Model as a whole, provided Your use, reproduction, and Distribution of the Model otherwise complies with the conditions stated in this License.
5. Use-based restrictions. The restrictions set forth in Attachment A are considered Use-based restrictions. Therefore You cannot use the Model and the Derivatives of the Model for the specified restricted uses. You may use the Model subject to this License, including only for lawful purposes and in accordance with the License. Use may include creating any content with, finetuning, updating, running, training, evaluating and/or reparametrizing the Model. You shall require all of Your users who use the Model or a Derivative of the Model to comply with the terms of this paragraph (paragraph 5).
6. The Output You Generate. Except as set forth herein, Licensor claims no rights in the Output You generate using the Model. You are accountable for the Output you generate and its subsequent uses. No use of the output can contravene any provision as stated in the License.
Section IV: OTHER PROVISIONS
7. Updates and Runtime Restrictions. To the maximum extent permitted by law, Licensor reserves the right to restrict (remotely or otherwise) usage of the Model in violation of this License, update the Model through electronic means, or modify the Output of the Model based on updates. You shall undertake reasonable efforts to use the latest version of the Model.
8. Trademarks and related. Nothing in this License permits You to make use of Licensors’ trademarks, trade names, logos or to otherwise suggest endorsement or misrepresent the relationship between the parties; and any rights not expressly granted herein are reserved by the Licensors.
9. Disclaimer of Warranty. Unless required by applicable law or agreed to in writing, Licensor provides the Model and the Complementary Material (and each Contributor provides its Contributions) on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied, including, without limitation, any warranties or conditions of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A PARTICULAR PURPOSE. You are solely responsible for determining the appropriateness of using or redistributing the Model, Derivatives of the Model, and the Complementary Material and assume any risks associated with Your exercise of permissions under this License.
10. Limitation of Liability. In no event and under no legal theory, whether in tort (including negligence), contract, or otherwise, unless required by applicable law (such as deliberate and grossly negligent acts) or agreed to in writing, shall any Contributor be liable to You for damages, including any direct, indirect, special, incidental, or consequential damages of any character arising as a result of this License or out of the use or inability to use the Model and the Complementary Material (including but not limited to damages for loss of goodwill, work stoppage, computer failure or malfunction, or any and all other commercial damages or losses), even if such Contributor has been advised of the possibility of such damages.
11. Accepting Warranty or Additional Liability. While redistributing the Model, Derivatives of the Model and the Complementary Material thereof, You may choose to offer, and charge a fee for, acceptance of support, warranty, indemnity, or other liability obligations and/or rights consistent with this License. However, in accepting such obligations, You may act only on Your own behalf and on Your sole responsibility, not on behalf of any other Contributor, and only if You agree to indemnify, defend, and hold each Contributor harmless for any liability incurred by, or claims asserted against, such Contributor by reason of your accepting any such warranty or additional liability.
12. If any provision of this License is held to be invalid, illegal or unenforceable, the remaining provisions shall be unaffected thereby and remain valid as if such provision had not been set forth herein.
END OF TERMS AND CONDITIONS
Attachment A
Use Restrictions
You agree not to use the Model or Derivatives of the Model:
- In any way that violates any applicable national, federal, state, local or international law or regulation;
- For the purpose of exploiting, harming or attempting to exploit or harm minors in any way;
- To generate or disseminate verifiably false information and/or content with the purpose of harming others;
- To generate or disseminate personal identifiable information that can be used to harm an individual;
- To defame, disparage or otherwise harass others;
- For fully automated decision making that adversely impacts an individual’s legal rights or otherwise creates or modifies a binding, enforceable obligation;
- For any use intended to or which has the effect of discriminating against or harming individuals or groups based on online or offline social behavior or known or predicted personal or personality characteristics;
- To exploit any of the vulnerabilities of a specific group of persons based on their age, social, physical or mental characteristics, in order to materially distort the behavior of a person pertaining to that group in a manner that causes or is likely to cause that person or another person physical or psychological harm;
- For any use intended to or which has the effect of discriminating against individuals or groups based on legally protected characteristics or categories;
- To provide medical advice and medical results interpretation;
- To generate or disseminate information for the purpose to be used for administration of justice, law enforcement, immigration or asylum processes, such as predicting an individual will commit fraud/crime commitment (e.g. by text profiling, drawing causal relationships between assertions made in documents, indiscriminate and arbitrarily-targeted use).
examples/tutorial/handson6/README.md deleted 100644 → 0
View file @ 351f0f64
# Handson 6: Acceleration of Stable Diffusion
*[Colosssal-AI](https://github.com/hpcaitech/ColossalAI) provides a faster and lower cost solution for pretraining and
fine-tuning for AIGC (AI-Generated Content) applications such as the model [stable-diffusion](https://github.com/CompVis/stable-diffusion) from [Stability AI](https://stability.ai/).*
We take advantage of [Colosssal-AI](https://github.com/hpcaitech/ColossalAI) to exploit multiple optimization strategies
, e.g. data parallelism, tensor parallelism, mixed precision & ZeRO, to scale the training to multiple GPUs.
## Stable Diffusion
[Stable Diffusion](https://huggingface.co/CompVis/stable-diffusion) is a latent text-to-image diffusion
model.
Thanks to a generous compute donation from [Stability AI](https://stability.ai/) and support from [LAION](https://laion.ai/), we were able to train a Latent Diffusion Model on 512x512 images from a subset of the [LAION-5B](https://laion.ai/blog/laion-5b/) database.
Similar to Google's [Imagen](https://arxiv.org/abs/2205.11487),
this model uses a frozen CLIP ViT-L/14 text encoder to condition the model on text prompts.
<p id="diffusion_train" align="center">
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/diffusion_train.png" width=800/>
</p>
[Stable Diffusion with Colossal-AI](https://github.com/hpcaitech/ColossalAI/tree/main/examples/images/diffusion) provides **6.5x faster training and pretraining cost saving, the hardware cost of fine-tuning can be almost 7X cheaper** (from RTX3090/4090 24GB to RTX3050/2070 8GB).
<p id="diffusion_demo" align="center">
<img src="https://raw.githubusercontent.com/hpcaitech/public_assets/main/colossalai/img/diffusion_demo.png" width=800/>
</p>
## Requirements
A suitable [conda](https://conda.io/) environment named `ldm` can be created
and activated with:
```
conda env create -f environment.yaml
conda activate ldm
```
You can also update an existing [latent diffusion](https://github.com/CompVis/latent-diffusion) environment by running
```
conda install pytorch torchvision -c pytorch
pip install transformers==4.19.2 diffusers invisible-watermark
pip install -e .
```
### Install [Colossal-AI v0.1.10](https://colossalai.org/download/) From Our Official Website
```
pip install colossalai==0.1.10+torch1.11cu11.3 -f https://release.colossalai.org
```
### Install [Lightning](https://github.com/Lightning-AI/lightning)
We use the Sep. 2022 version with commit id as `b04a7aa`.
```
git clone https://github.com/Lightning-AI/lightning && cd lightning && git reset --hard b04a7aa
pip install -r requirements.txt && pip install .
```
> The specified version is due to the interface incompatibility caused by the latest update of [Lightning](https://github.com/Lightning-AI/lightning), which will be fixed in the near future.
## Dataset
The DataSet is from [LAION-5B](https://laion.ai/blog/laion-5b/), the subset of [LAION](https://laion.ai/),
you should the change the `data.file_path` in the `config/train_colossalai.yaml`
## Training
we provide the script `train.sh` to run the training task , and two Stategy in `configs`:`train_colossalai.yaml`, `train_ddp.yaml`
for example, you can run the training from colossalai by
```
python main.py --logdir /tmp -t --postfix test -b config/train_colossalai.yaml
```
- you can change the `--logdir` the save the log information and the last checkpoint
### Training config
you can change the trainging config in the yaml file
- accelerator: acceleratortype, default 'gpu'
- devices: device number used for training, default 4
- max_epochs: max training epochs
- precision: usefp16 for training or not, default 16, you must use fp16 if you want to apply colossalai
## Comments
- Our codebase for the diffusion models builds heavily on [OpenAI's ADM codebase](https://github.com/openai/guided-diffusion)
, [lucidrains](https://github.com/lucidrains/denoising-diffusion-pytorch),
[Stable Diffusion](https://github.com/CompVis/stable-diffusion), [Lightning](https://github.com/Lightning-AI/lightning) and [Hugging Face](https://huggingface.co/CompVis/stable-diffusion).
Thanks for open-sourcing!
- The implementation of the transformer encoder is from [x-transformers](https://github.com/lucidrains/x-transformers) by [lucidrains](https://github.com/lucidrains?tab=repositories).
- The implementation of [flash attention](https://github.com/HazyResearch/flash-attention) is from [HazyResearch](https://github.com/HazyResearch).
## BibTeX
```
@article{bian2021colossal,
title={Colossal-AI: A Unified Deep Learning System For Large-Scale Parallel Training},
author={Bian, Zhengda and Liu, Hongxin and Wang, Boxiang and Huang, Haichen and Li, Yongbin and Wang, Chuanrui and Cui, Fan and You, Yang},
journal={arXiv preprint arXiv:2110.14883},
year={2021}
}
@misc{rombach2021highresolution,
title={High-Resolution Image Synthesis with Latent Diffusion Models},
author={Robin Rombach and Andreas Blattmann and Dominik Lorenz and Patrick Esser and Björn Ommer},
year={2021},
eprint={2112.10752},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
@article{dao2022flashattention,
title={FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness},
author={Dao, Tri and Fu, Daniel Y. and Ermon, Stefano and Rudra, Atri and R{\'e}, Christopher},
journal={arXiv preprint arXiv:2205.14135},
year={2022}
}
```
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment