Set default dtype to half
This commit is contained in:
parent
de0fabbc5c
commit
1ce1333573
@ -1,3 +1,6 @@
|
|||||||
|
from typing import Union
|
||||||
|
|
||||||
|
import torch
|
||||||
import torch.nn as nn
|
import torch.nn as nn
|
||||||
|
|
||||||
from cacheflow.models.opt import OPTForCausalLM
|
from cacheflow.models.opt import OPTForCausalLM
|
||||||
@ -6,9 +9,23 @@ MODEL_CLASSES = {
|
|||||||
'opt': OPTForCausalLM,
|
'opt': OPTForCausalLM,
|
||||||
}
|
}
|
||||||
|
|
||||||
|
STR_DTYPE_TO_TORCH_DTYPE = {
|
||||||
|
'half': torch.half,
|
||||||
|
'float': torch.float,
|
||||||
|
'float16': torch.float16,
|
||||||
|
'float32': torch.float32,
|
||||||
|
}
|
||||||
|
|
||||||
def get_model(model_name: str) -> nn.Module:
|
|
||||||
|
def get_model(
|
||||||
|
model_name: str,
|
||||||
|
dtype: Union[torch.dtype, str],
|
||||||
|
) -> nn.Module:
|
||||||
|
if isinstance(dtype, str):
|
||||||
|
torch_dtype = STR_DTYPE_TO_TORCH_DTYPE[dtype.lower()]
|
||||||
|
else:
|
||||||
|
torch_dtype = dtype
|
||||||
for model_class, model in MODEL_CLASSES.items():
|
for model_class, model in MODEL_CLASSES.items():
|
||||||
if model_class in model_name:
|
if model_class in model_name:
|
||||||
return model.from_pretrained(model_name)
|
return model.from_pretrained(model_name, torch_dtype=torch_dtype)
|
||||||
raise ValueError(f'Invalid model name: {model_name}')
|
raise ValueError(f'Invalid model name: {model_name}')
|
||||||
|
|||||||
@ -14,6 +14,7 @@ class Controller:
|
|||||||
block_size: int,
|
block_size: int,
|
||||||
num_gpu_blocks: int,
|
num_gpu_blocks: int,
|
||||||
num_cpu_blocks: int,
|
num_cpu_blocks: int,
|
||||||
|
dtype: str = 'half',
|
||||||
) -> None:
|
) -> None:
|
||||||
self.node_id = node_id
|
self.node_id = node_id
|
||||||
self.num_workers = num_workers
|
self.num_workers = num_workers
|
||||||
@ -35,6 +36,7 @@ class Controller:
|
|||||||
block_size=block_size,
|
block_size=block_size,
|
||||||
num_gpu_blocks=num_gpu_blocks,
|
num_gpu_blocks=num_gpu_blocks,
|
||||||
num_cpu_blocks=num_cpu_blocks,
|
num_cpu_blocks=num_cpu_blocks,
|
||||||
|
dtype=dtype,
|
||||||
)
|
)
|
||||||
self.workers.append(worker)
|
self.workers.append(worker)
|
||||||
|
|
||||||
|
|||||||
@ -17,6 +17,7 @@ class Worker:
|
|||||||
block_size: int,
|
block_size: int,
|
||||||
num_gpu_blocks: int,
|
num_gpu_blocks: int,
|
||||||
num_cpu_blocks: int,
|
num_cpu_blocks: int,
|
||||||
|
dtype: str,
|
||||||
) -> None:
|
) -> None:
|
||||||
self.worker_id = worker_id
|
self.worker_id = worker_id
|
||||||
self.gpu_id = gpu_id
|
self.gpu_id = gpu_id
|
||||||
@ -26,7 +27,7 @@ class Worker:
|
|||||||
|
|
||||||
# Initialize the model.
|
# Initialize the model.
|
||||||
# FIXME(woosuk): This is a hack.
|
# FIXME(woosuk): This is a hack.
|
||||||
self.model = get_model(model_name).to(device=gpu_id)
|
self.model = get_model(model_name, dtype=dtype).to(device=self.device)
|
||||||
self.num_layers = self.model.config.num_hidden_layers
|
self.num_layers = self.model.config.num_hidden_layers
|
||||||
self.num_heads = self.model.config.num_attention_heads
|
self.num_heads = self.model.config.num_attention_heads
|
||||||
self.head_size = self.model.config.hidden_size // self.num_heads
|
self.head_size = self.model.config.hidden_size // self.num_heads
|
||||||
|
|||||||
Loading…
Reference in New Issue
Block a user