hf_text-generation-inference/server/text_generation_server/models/t5.py

246 lines
9.3 KiB
Python
Raw Normal View History

2023-02-07 10:25:17 -07:00
import torch
import torch.distributed
from typing import List, Optional, Tuple
from accelerate import init_empty_weights
from safetensors import safe_open
from transformers import (
AutoTokenizer,
AutoModelForSeq2SeqLM,
AutoConfig,
)
from transformers.models.t5.parallel_layers import (
TensorParallelColumnLinear,
TensorParallelEmbedding,
TensorParallelRowLinear,
)
2023-03-07 10:52:22 -07:00
from text_generation_server.models import Seq2SeqLM
from text_generation_server.utils import (
2023-02-07 10:25:17 -07:00
initialize_torch_distributed,
weight_files,
)
HAS_BITS_AND_BYTES = True
try:
import bitsandbytes as bnb
from bitsandbytes.nn import Int8Params
except Exception as e:
HAS_BITS_AND_BYTES = False
class T5Sharded(Seq2SeqLM):
def __init__(
self, model_id: str, revision: Optional[str] = None, quantize: bool = False
):
self.process_group, self.rank, self.world_size = initialize_torch_distributed()
self.master = self.rank == 0
if torch.cuda.is_available():
device = torch.device(f"cuda:{self.rank}")
dtype = torch.bfloat16 if torch.cuda.is_bf16_supported() else torch.float32
2023-02-07 10:25:17 -07:00
else:
device = torch.device("cpu")
dtype = torch.float32
tokenizer = AutoTokenizer.from_pretrained(
model_id, revision=revision, padding_side="left", truncation_side="left"
2023-02-07 10:25:17 -07:00
)
config = AutoConfig.from_pretrained(
model_id, revision=revision, tp_parallel=True
)
tokenizer.bos_token_id = config.decoder_start_token_id
torch.distributed.barrier(group=self.process_group)
filenames = weight_files(model_id, revision=revision, extension=".safetensors")
with init_empty_weights():
model = AutoModelForSeq2SeqLM.from_config(config)
torch.distributed.barrier(group=self.process_group)
self.load_weights(
model,
filenames,
quantize=quantize,
device=device,
dtype=dtype,
2023-02-07 10:25:17 -07:00
rank=self.rank,
world_size=self.world_size,
)
self.model = model.eval()
2023-02-07 10:25:17 -07:00
torch.distributed.barrier(group=self.process_group)
super(Seq2SeqLM, self).__init__(
tokenizer=tokenizer,
device=device,
)
@staticmethod
def load_weights(
model,
filenames: List[str],
quantize: bool,
device: torch.device,
dtype: torch.dtype,
2023-02-07 10:25:17 -07:00
rank: int,
world_size: int,
):
parameters = dict(model.named_parameters())
for file in filenames:
with safe_open(
file, framework="pt", device=str(device) if not quantize else "cpu"
) as f:
for name in f.keys():
module_name, param_name = name.rsplit(".", 1)
module = model.get_submodule(module_name)
current_parameter_tensor = parameters.get(name, None)
slice_ = f.get_slice(name)
if isinstance(module, TensorParallelColumnLinear):
size = slice_.get_shape()[0]
block_size = size // world_size
start = rank * block_size
stop = (rank + 1) * block_size
tensor = slice_[start:stop]
elif isinstance(module, TensorParallelRowLinear):
if param_name == "weight":
size = slice_.get_shape()[1]
block_size = size // world_size
start = rank * block_size
stop = (rank + 1) * block_size
tensor = slice_[:, start:stop]
else:
tensor = slice_[:]
# XXX: Hack for Rowlinear to add the bias only once.
if rank != 0:
tensor = torch.zeros_like(tensor)
elif isinstance(module, TensorParallelEmbedding):
size = slice_.get_shape()[0]
block_size = size // world_size
start = rank * block_size
stop = (rank + 1) * block_size
tensor = slice_[start:stop]
elif name == "lm_head.weight":
size = slice_.get_shape()[0]
block_size = size // world_size
start = rank * block_size
stop = (rank + 1) * block_size
tensor = slice_[start:stop]
elif "relative_attention_bias.weight" in name:
size = slice_.get_shape()[1]
block_size = size // world_size
start = rank * block_size
stop = (rank + 1) * block_size
tensor = slice_[:, start:stop]
else:
try:
tensor = slice_[:]
except:
tensor = f.get_tensor(name)
if (
current_parameter_tensor is not None
and current_parameter_tensor.shape != tensor.shape
):
raise ValueError(
f"Name {name} -- Current {current_parameter_tensor.shape} and got {tensor.shape}"
)
tensor = tensor.contiguous().to(dtype)
2023-02-07 10:25:17 -07:00
if quantize:
if not HAS_BITS_AND_BYTES:
raise ImportError(
"bitsandbytes is not available on your machine either because it is not installed "
"or you don't have a GPU.\n"
"You can install it with `pip install bitsandbytes`."
)
if (
type(module)
in [TensorParallelRowLinear, TensorParallelColumnLinear]
and param_name == "weight"
):
tensor = Int8Params(
tensor,
has_fp16_weights=False,
requires_grad=False,
).to(device)
state = bnb.MatmulLtState()
state.threshold = 6.0
state.has_fp16_weights = False
state.memory_efficient_backward = False
state.use_pool = True
state.CB = tensor.CB
state.SCB = tensor.SCB
tensor.CB = None
tensor.SCB = None
def replace_linear(state):
def linear(input, weight, bias):
out = bnb.matmul(
input,
weight,
state=state,
threshold=state.threshold,
bias=bias,
)
if state.CB is not None:
# we converted 8-bit row major to turing/ampere format
# in the first inference pass
# we no longer need the row-major weight
del state.CB
weight.data = state.CxB
return out
return linear
module.linear = replace_linear(state)
else:
tensor = tensor.to(device)
if current_parameter_tensor is not None:
module._parameters[param_name] = tensor
else:
module._buffers[param_name] = tensor
def forward(
self,
input_ids,
attention_mask,
decoder_input_ids,
decoder_attention_mask: Optional,
encoder_last_hidden_state: Optional,
past_key_values: Optional = None,
) -> Tuple[
torch.Tensor,
torch.Tensor,
List[Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]],
]:
# Model Forward
outputs = self.model.forward(
input_ids=input_ids,
attention_mask=attention_mask,
decoder_input_ids=decoder_input_ids,
decoder_attention_mask=decoder_attention_mask,
encoder_outputs=encoder_last_hidden_state,
past_key_values=past_key_values,
use_cache=True,
)
# Logits are sharded, so we need to gather them
logits = [torch.empty_like(outputs.logits) for _ in range(self.world_size)]
torch.distributed.all_gather(logits, outputs.logits, group=self.process_group)
logits = torch.cat(logits, dim=2)
return (
logits,
outputs.encoder_last_hidden_state,
outputs.past_key_values,
)