xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-11 02:25:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Misc][LoRA] Clean up the function interface of Punica (#10917)

Browse Source 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
This commit is contained in:
Jee Jee Li 2024-12-05 21:22:28 +08:00 committed by GitHub
parent 39c89e71a8
commit 571da8fc43
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
5 changed files with 504 additions and 638 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

42
tests/lora/test_layers.py
View File

@ -565,7 +565,9 @@ def test_lm_head_logits_processor(dist_init, num_loras, device, vocab_size,
@pytest.mark.parametrize("num_loras", [1, 2, 4, 8]) @pytest.mark.parametrize("num_loras", [1, 2, 4, 8])
@pytest.mark.parametrize("device", CUDA_DEVICES) @pytest.mark.parametrize("device", CUDA_DEVICES)
@pytest.mark.parametrize("stage", STAGES) @pytest.mark.parametrize("stage", STAGES)
def test_linear_replicated(dist_init, num_loras, device, stage) -> None: @pytest.mark.parametrize("bias_enabled", [True, False])
def test_linear_replicated(dist_init, num_loras, device, stage,
bias_enabled) -> None:
torch.cuda.set_device(device) torch.cuda.set_device(device)
torch.set_default_device(device) torch.set_default_device(device)
@ -573,7 +575,8 @@ def test_linear_replicated(dist_init, num_loras, device, stage) -> None:
max_loras = 8 max_loras = 8
lora_config = LoRAConfig(max_loras=max_loras, lora_config = LoRAConfig(max_loras=max_loras,
max_lora_rank=8, max_lora_rank=8,
lora_dtype=torch.float16) lora_dtype=torch.float16,
bias_enabled=bias_enabled)
def create_random_linear_replicated_layer(): def create_random_linear_replicated_layer():
@ -585,7 +588,12 @@ def test_linear_replicated(dist_init, num_loras, device, stage) -> None:
lora_linear = ReplicatedLinearWithLoRA(linear) lora_linear = ReplicatedLinearWithLoRA(linear)
lora_linear.create_lora_weights(max_loras, lora_config) lora_linear.create_lora_weights(max_loras, lora_config)
assert (lora_linear.n_slices == len(lora_linear.lora_a_stacked) == len(
lora_linear.lora_b_stacked) == 1)
if bias_enabled:
assert len(lora_linear.lora_bias_stacked) == lora_linear.n_slices
else:
assert lora_linear.lora_bias_stacked is None
return linear, lora_linear return linear, lora_linear
for i in range(10): for i in range(10):
@ -669,8 +677,9 @@ def test_linear_replicated(dist_init, num_loras, device, stage) -> None:
@pytest.mark.parametrize("fully_shard", [True, False]) @pytest.mark.parametrize("fully_shard", [True, False])
@pytest.mark.parametrize("device", CUDA_DEVICES) @pytest.mark.parametrize("device", CUDA_DEVICES)
@pytest.mark.parametrize("stage", STAGES) @pytest.mark.parametrize("stage", STAGES)
@pytest.mark.parametrize("bias_enabled", [True, False])
def test_linear_parallel(dist_init, num_loras, orientation, fully_shard, def test_linear_parallel(dist_init, num_loras, orientation, fully_shard,
device, stage) -> None: device, stage, bias_enabled) -> None:
torch.cuda.set_device(device) torch.cuda.set_device(device)
torch.set_default_device(device) torch.set_default_device(device)
@ -679,7 +688,8 @@ def test_linear_parallel(dist_init, num_loras, orientation, fully_shard,
lora_config = LoRAConfig(max_loras=max_loras, lora_config = LoRAConfig(max_loras=max_loras,
max_lora_rank=8, max_lora_rank=8,
fully_sharded_loras=fully_shard, fully_sharded_loras=fully_shard,
lora_dtype=torch.float16) lora_dtype=torch.float16,
bias_enabled=bias_enabled)
def create_random_linear_parallel_layer(): def create_random_linear_parallel_layer():
if orientation == "row": if orientation == "row":
@ -700,7 +710,12 @@ def test_linear_parallel(dist_init, num_loras, orientation, fully_shard,
if not fully_shard else if not fully_shard else
ColumnParallelLinearWithShardedLoRA(linear)) ColumnParallelLinearWithShardedLoRA(linear))
lora_linear.create_lora_weights(max_loras, lora_config) lora_linear.create_lora_weights(max_loras, lora_config)
assert (lora_linear.n_slices == len(lora_linear.lora_a_stacked) == len(
lora_linear.lora_b_stacked) == 1)
if bias_enabled:
assert len(lora_linear.lora_bias_stacked) == lora_linear.n_slices
else:
assert lora_linear.lora_bias_stacked is None
return linear, lora_linear return linear, lora_linear
for i in range(10): for i in range(10):
@ -784,8 +799,9 @@ def test_linear_parallel(dist_init, num_loras, orientation, fully_shard,
@pytest.mark.parametrize("fully_shard", [True, False]) @pytest.mark.parametrize("fully_shard", [True, False])
@pytest.mark.parametrize("device", CUDA_DEVICES) @pytest.mark.parametrize("device", CUDA_DEVICES)
@pytest.mark.parametrize("stage", STAGES) @pytest.mark.parametrize("stage", STAGES)
@pytest.mark.parametrize("bias_enabled", [True, False])
def test_column_parallel_packed(dist_init, num_loras, repeats, fully_shard, def test_column_parallel_packed(dist_init, num_loras, repeats, fully_shard,
device, stage) -> None: device, stage, bias_enabled) -> None:
torch.cuda.set_device(device) torch.cuda.set_device(device)
torch.set_default_device(device) torch.set_default_device(device)
@ -794,7 +810,8 @@ def test_column_parallel_packed(dist_init, num_loras, repeats, fully_shard,
lora_config = LoRAConfig(max_loras=max_loras, lora_config = LoRAConfig(max_loras=max_loras,
max_lora_rank=8, max_lora_rank=8,
fully_sharded_loras=fully_shard, fully_sharded_loras=fully_shard,
lora_dtype=torch.float16) lora_dtype=torch.float16,
bias_enabled=bias_enabled)
def create_column_parallel_packed_layer(): def create_column_parallel_packed_layer():
if repeats == 2: if repeats == 2:
@ -832,10 +849,16 @@ def test_column_parallel_packed(dist_init, num_loras, repeats, fully_shard,
num_key_value_heads = 32 num_key_value_heads = 32
num_attention_heads = 32 num_attention_heads = 32
n_slices = repeats
lora_linear.create_lora_weights(max_loras, lora_linear.create_lora_weights(max_loras,
lora_config, lora_config,
model_config=FakeConfig()) model_config=FakeConfig())
assert (lora_linear.n_slices == len(lora_linear.lora_a_stacked) == len(
lora_linear.lora_b_stacked) == n_slices)
if bias_enabled:
assert len(lora_linear.lora_bias_stacked) == lora_linear.n_slices
else:
assert lora_linear.lora_bias_stacked is None
return linear, lora_linear return linear, lora_linear
for i in range(10): for i in range(10):
@ -911,7 +934,6 @@ def test_column_parallel_packed(dist_init, num_loras, repeats, fully_shard,
512, 512,
lora_config.lora_extra_vocab_size, lora_config.lora_extra_vocab_size,
) )
# lora_linear.set_mapping(*mapping_info)
lora_result = lora_linear(torch.cat(inputs))[0] lora_result = lora_linear(torch.cat(inputs))[0]
expected_result = linear(torch.cat(inputs))[0] expected_result = linear(torch.cat(inputs))[0]

175
vllm/lora/fully_sharded_layers.py
View File

@ -1,5 +1,5 @@
# pylint: disable=unused-argument # pylint: disable=unused-argument
from typing import TYPE_CHECKING, List, Optional, Union from typing import TYPE_CHECKING, List, Optional, Tuple, Union, cast
import torch import torch
import torch.nn as nn import torch.nn as nn
@ -32,6 +32,44 @@ def _fully_sharded_can_replace(can_replace):
return dec return dec
def _mcp_apply(x, bias, layer: ColumnParallelLinearWithLoRA):
"""
For `ColumnParallelLinearWithLoRA` or classes that inherit from
`ColumnParallelLinearWithLoRA`, they share the same `apply` logic.
"""
assert (layer.n_slices == len(layer.lora_a_stacked) == len(
layer.lora_b_stacked) == len(layer.output_slices))
if layer.lora_bias_stacked is not None:
assert layer.n_slices == len(layer.lora_bias_stacked)
output = layer.base_layer.quant_method.apply(layer.base_layer, x, bias)
x = x.view(-1, x.shape[-1])
output, out_orig_shape = output.view(-1, output.shape[-1]), output.shape
# Since communication is needed, the buffer is directly initialized as a
# tensor rather than a tuple of tensor.
buffers = torch.zeros(
(layer.n_slices, x.shape[0], layer.lora_a_stacked[0].shape[2]),
dtype=torch.float32,
device=x.device,
)
layer.punica_wrapper.add_shrink(buffers, x, layer.lora_a_stacked, 1.0)
buffers = tensor_model_parallel_all_gather(buffers)
layer.punica_wrapper.add_expand(output,
buffers,
layer.lora_b_stacked,
layer.lora_bias_stacked,
layer.output_slices,
offset_start=0,
add_input=True)
output = output.view(*out_orig_shape)
# now have column partitioned and packed output
return output
# these layers are based on the tensor parallelism strategy given in # these layers are based on the tensor parallelism strategy given in
# Y. Sheng et al., S-LoRA: Serving Thousands of Concurrent LoRA Adapters. 2023, # Y. Sheng et al., S-LoRA: Serving Thousands of Concurrent LoRA Adapters. 2023,
# https://arxiv.org/abs/2311.03285. # https://arxiv.org/abs/2311.03285.
@ -51,34 +89,15 @@ class ColumnParallelLinearWithShardedLoRA(ColumnParallelLinearWithLoRA):
# gather operation. # gather operation.
def slice_lora_a(self, lora_a: torch.Tensor) -> torch.Tensor: def slice_lora_a(self, lora_a: torch.Tensor) -> torch.Tensor:
tp_rank = get_tensor_model_parallel_rank() tp_rank = get_tensor_model_parallel_rank()
shard_size = self.lora_a_stacked.shape[2] shard_size = self.lora_a_stacked[0].shape[2]
start_idx = tp_rank * shard_size start_idx = tp_rank * shard_size
lora_a = lora_a[:, start_idx:start_idx + shard_size] lora_a = lora_a[:, start_idx:start_idx + shard_size]
return lora_a return lora_a
def apply(self, x: torch.Tensor, def apply(self,
bias: Optional[torch.Tensor]) -> torch.Tensor: x: torch.Tensor,
output = self.base_layer.quant_method.apply(self.base_layer, x, bias) bias: Optional[torch.Tensor] = None) -> torch.Tensor:
return _mcp_apply(x, bias, self)
x = x.view(-1, x.shape[-1])
output, out_orig_shape = output.view(-1,
output.shape[-1]), output.shape
buffer = torch.zeros(
(x.shape[0], self.lora_a_stacked.shape[2]),
dtype=torch.float32,
device=x.device,
)
self.punica_wrapper.add_shrink(buffer, x, self.lora_a_stacked, 1.0)
buffer = tensor_model_parallel_all_gather(buffer)
self.punica_wrapper.add_expand(output,
buffer,
self.lora_b_stacked,
self.bias_stacked,
add_input=True)
# now have column partitioned output
output = output.view(*out_orig_shape)
return output
@classmethod @classmethod
@_fully_sharded_can_replace @_fully_sharded_can_replace
@ -99,46 +118,6 @@ class ColumnParallelLinearWithShardedLoRA(ColumnParallelLinearWithLoRA):
) )
def _mcp_apply(x, bias, layer: QKVParallelLinearWithLora):
"""
MergedColumnParallelLinearWithShardedLoRA and
MergedQKVParallelLinearWithShardedLora share the same
LoRa weight application method.
The main difference is the step by shard_size for lora_b which can
vary for MergedQKVParallelLinearWithShardedLora but is constant for
MergedColumnParallelLinearWithShardedLoRA.
"""
# expecting 2 for column parallel and 3 for qkv
n = len(layer.lora_a_stacked)
output = layer.base_layer.quant_method.apply(layer.base_layer, x, bias)
x = x.view(-1, x.shape[-1])
output, out_orig_shape = output.view(-1, output.shape[-1]), output.shape
buffers = torch.zeros(
(n, x.shape[0], layer.lora_a_stacked[0].shape[2]),
dtype=torch.float32,
device=x.device,
)
for idx in range(n):
layer.punica_wrapper.add_shrink(buffers[idx], x,
layer.lora_a_stacked[idx], 1.0)
buffers = tensor_model_parallel_all_gather(buffers)
layer.punica_wrapper.add_expand_packed_nslice(
output,
buffers,
layer.lora_b_stacked,
layer.bias_stacked,
1.0,
layer.output_slices,
)
output = output.view(*out_orig_shape)
# now have column partitioned and packed output
return output
class MergedColumnParallelLinearWithShardedLoRA( class MergedColumnParallelLinearWithShardedLoRA(
MergedColumnParallelLinearWithLoRA): MergedColumnParallelLinearWithLoRA):
""" """
@ -162,8 +141,9 @@ class MergedColumnParallelLinearWithShardedLoRA(
] ]
return lora_a return lora_a
def apply(self, x: torch.Tensor, def apply(self,
bias: Optional[torch.Tensor]) -> torch.Tensor: x: torch.Tensor,
bias: Optional[torch.Tensor] = None) -> torch.Tensor:
return _mcp_apply(x, bias, self) return _mcp_apply(x, bias, self)
@classmethod @classmethod
@ -195,31 +175,15 @@ class QKVParallelLinearWithShardedLora(QKVParallelLinearWithLora):
def slice_lora_a(self, lora_a: torch.Tensor) -> torch.Tensor: def slice_lora_a(self, lora_a: torch.Tensor) -> torch.Tensor:
tp_rank = get_tensor_model_parallel_rank() tp_rank = get_tensor_model_parallel_rank()
shard_size = self.lora_a_stacked.shape[2] shard_size = self.lora_a_stacked[0].shape[2]
start_idx = tp_rank * shard_size start_idx = tp_rank * shard_size
lora_a = lora_a[:, start_idx:start_idx + shard_size] lora_a = lora_a[:, start_idx:start_idx + shard_size]
return lora_a return lora_a
def apply(self, x: torch.Tensor, def apply(self,
bias: Optional[torch.Tensor]) -> torch.Tensor: x: torch.Tensor,
output = self.base_layer.quant_method.apply(self.base_layer, x, bias) bias: Optional[torch.Tensor] = None) -> torch.Tensor:
return _mcp_apply(x, bias, self)
x = x.view(-1, x.shape[-1])
output, out_orig_shape = output.view(-1,
output.shape[-1]), output.shape
buffer = torch.zeros((x.shape[0], self.lora_a_stacked.shape[2]),
dtype=torch.float32,
device=x.device)
self.punica_wrapper.add_shrink(buffer, x, self.lora_a_stacked, 1.0)
buffer = tensor_model_parallel_all_gather(buffer)
self.punica_wrapper.add_expand(output,
buffer,
self.lora_b_stacked,
self.bias_stacked,
add_input=True)
# now have column partitioned output
output = output.view(*out_orig_shape)
return output
@classmethod @classmethod
@_fully_sharded_can_replace @_fully_sharded_can_replace
@ -260,8 +224,9 @@ class MergedQKVParallelLinearWithShardedLora(MergedQKVParallelLinearWithLora):
] ]
return lora_a return lora_a
def apply(self, x: torch.Tensor, def apply(self,
bias: Optional[torch.Tensor]) -> torch.Tensor: x: torch.Tensor,
bias: Optional[torch.Tensor] = None) -> torch.Tensor:
return _mcp_apply(x, bias, self) return _mcp_apply(x, bias, self)
@classmethod @classmethod
@ -294,7 +259,7 @@ class RowParallelLinearWithShardedLoRA(RowParallelLinearWithLoRA):
""" """
def slice_lora_b(self, lora_b: torch.Tensor) -> torch.Tensor: def slice_lora_b(self, lora_b: torch.Tensor) -> torch.Tensor:
shard_size = self.lora_b_stacked.shape[2] shard_size = self.lora_b_stacked[0].shape[2]
start_idx = self.tp_rank * shard_size start_idx = self.tp_rank * shard_size
end_idx = (self.tp_rank + 1) * shard_size end_idx = (self.tp_rank + 1) * shard_size
lora_b = lora_b[:, start_idx:end_idx] lora_b = lora_b[:, start_idx:end_idx]
@ -303,20 +268,24 @@ class RowParallelLinearWithShardedLoRA(RowParallelLinearWithLoRA):
def slice_bias(self, bias: torch.Tensor) -> torch.Tensor: def slice_bias(self, bias: torch.Tensor) -> torch.Tensor:
if bias is None: if bias is None:
return bias return bias
shard_size = self.bias_stacked.shape[2] self.lora_bias_stacked = cast(Tuple[torch.Tensor, ...],
self.lora_bias_stacked)
shard_size = self.lora_bias_stacked[0].shape[2]
start_idx = self.tp_rank * shard_size start_idx = self.tp_rank * shard_size
end_idx = (self.tp_rank + 1) * shard_size end_idx = (self.tp_rank + 1) * shard_size
bias = bias[start_idx:end_idx] bias = bias[start_idx:end_idx]
return bias return bias
def apply(self, x: torch.Tensor) -> torch.Tensor: def apply(self,
x: torch.Tensor,
bias: Optional[torch.Tensor] = None) -> torch.Tensor:
output = self.base_layer.quant_method.apply(self.base_layer, x) output = self.base_layer.quant_method.apply(self.base_layer, x)
x = x.view(-1, x.shape[-1]) x = x.view(-1, x.shape[-1])
output, out_orig_shape = output.view(-1, output, out_orig_shape = output.view(-1,
output.shape[-1]), output.shape output.shape[-1]), output.shape
buffer = torch.zeros( buffer = torch.zeros(
(x.shape[0], self.lora_a_stacked.shape[2]), (self.n_slices, x.shape[0], self.lora_a_stacked[0].shape[2]),
dtype=torch.float32, dtype=torch.float32,
device=x.device, device=x.device,
) )
@ -330,12 +299,18 @@ class RowParallelLinearWithShardedLoRA(RowParallelLinearWithLoRA):
# remains is a standard all_reduce. User should be aware though that # remains is a standard all_reduce. User should be aware though that
# the output is not the same as a normal row_parallel, it should be # the output is not the same as a normal row_parallel, it should be
# reduced before being used # reduced before being used
shard_size = self.lora_b_stacked.shape[2] # NOTE offset are based on the rank.
start_idx = self.tp_rank * shard_size shard_size = self.lora_b_stacked[0].shape[2]
self.punica_wrapper.add_expand_slice(output, buffer, offset_start = self.tp_rank * shard_size
self.lora_b_stacked, self.punica_wrapper.add_expand(
self.bias_stacked, start_idx, output,
shard_size) buffer,
self.lora_b_stacked,
self.lora_bias_stacked,
self.output_slices,
offset_start=offset_start,
add_input=True,
)
output = output.view(*out_orig_shape) output = output.view(*out_orig_shape)
return output return output

534
vllm/lora/layers.py
View File

@ -1,7 +1,7 @@
# pylint: disable=unused-argument # pylint: disable=unused-argument
import math import math
from dataclasses import dataclass from dataclasses import dataclass
from typing import TYPE_CHECKING, Dict, List, Optional, Tuple, Union from typing import TYPE_CHECKING, Dict, List, Optional, Tuple, Union, cast
import torch import torch
import torch.nn as nn import torch.nn as nn
@ -18,11 +18,14 @@ from vllm.distributed import (get_tensor_model_parallel_rank,
tensor_model_parallel_gather) tensor_model_parallel_gather)
from vllm.distributed.utils import divide from vllm.distributed.utils import divide
from vllm.lora.punica import PunicaWrapper from vllm.lora.punica import PunicaWrapper
# yapf: disable
from vllm.model_executor.layers.linear import (ColumnParallelLinear, from vllm.model_executor.layers.linear import (ColumnParallelLinear,
LinearBase,
MergedColumnParallelLinear, MergedColumnParallelLinear,
QKVParallelLinear, QKVParallelLinear,
ReplicatedLinear, ReplicatedLinear,
RowParallelLinear) RowParallelLinear)
# yapf: enable
from vllm.model_executor.layers.logits_processor import LogitsProcessor from vllm.model_executor.layers.logits_processor import LogitsProcessor
from vllm.model_executor.layers.rotary_embedding import ( from vllm.model_executor.layers.rotary_embedding import (
LinearScalingRotaryEmbedding, RotaryEmbedding) LinearScalingRotaryEmbedding, RotaryEmbedding)
@ -249,13 +252,10 @@ class VocabParallelEmbeddingWithLoRA(BaseLayerWithLoRA):
full_lora_a_embeddings.shape[1], full_lora_a_embeddings.shape[1],
-1, -1,
) )
self.punica_wrapper.add_lora_embedding(full_output,
# Embedding layer only need expand op full_lora_a_embeddings,
self.punica_wrapper.add_expand(full_output, self.lora_b_stacked,
full_lora_a_embeddings, add_input=True)
self.lora_b_stacked,
bias_all=None,
add_input=True)
return full_output.view_as(full_output_org) return full_output.view_as(full_output_org)
@classmethod @classmethod
@ -269,14 +269,19 @@ class VocabParallelEmbeddingWithLoRA(BaseLayerWithLoRA):
return type(source_layer) is VocabParallelEmbedding return type(source_layer) is VocabParallelEmbedding
class ReplicatedLinearWithLoRA(BaseLayerWithLoRA): class BaseLinearLayerWithLoRA(BaseLayerWithLoRA):
def __init__(self, base_layer: ReplicatedLinear) -> None: def __init__(self, base_layer: LinearBase):
super().__init__() super().__init__()
self.base_layer = base_layer self.base_layer = base_layer
self.input_size = self.base_layer.input_size self.input_size = self.base_layer.input_size
self.output_size = self.base_layer.output_size
self.device = _get_lora_device(self.base_layer) self.device = _get_lora_device(self.base_layer)
self.lora_bias_stacked: Optional[Tuple[torch.Tensor, ...]] = None
self.output_slices: Tuple[int, ...]
self.tp_size: int
self.output_size: int
self.n_slices: int
def create_lora_weights( def create_lora_weights(
self, self,
@ -285,39 +290,64 @@ class ReplicatedLinearWithLoRA(BaseLayerWithLoRA):
model_config: Optional[PretrainedConfig] = None, model_config: Optional[PretrainedConfig] = None,
) -> None: ) -> None:
self.lora_config = lora_config self.lora_config = lora_config
lora_a_output_size = lora_config.max_lora_rank #
self.lora_a_stacked = torch.zeros( if isinstance(self.base_layer, ReplicatedLinear):
max_loras, lora_a_out_size = lora_config.max_lora_rank
1, lora_b_out_size = self.output_size
lora_a_output_size,
self.input_size, elif isinstance(self.base_layer, ColumnParallelLinear):
dtype=lora_config.lora_dtype, lora_a_out_size = (lora_config.max_lora_rank if
device=self.device, not lora_config.fully_sharded_loras else divide(
) lora_config.max_lora_rank, self.tp_size))
self.lora_b_stacked = torch.zeros( lora_b_out_size = self.output_size
max_loras,
1, elif isinstance(self.base_layer, RowParallelLinear):
self.output_size, lora_a_out_size = lora_config.max_lora_rank
lora_config.max_lora_rank, lora_b_out_size = (self.output_size if
dtype=lora_config.lora_dtype, not lora_config.fully_sharded_loras else divide(
device=self.device, self.output_size, self.tp_size))
) else:
if lora_config.bias_enabled: raise NotImplementedError
self.bias_stacked = torch.zeros(
self.lora_a_stacked = tuple(
torch.zeros(
max_loras, max_loras,
1, 1,
self.output_size, lora_a_out_size,
self.input_size,
dtype=lora_config.lora_dtype, dtype=lora_config.lora_dtype,
device=self.device, device=self.device,
) ) for _ in range(self.n_slices))
else: self.lora_b_stacked = tuple(
self.bias_stacked = None torch.zeros(
max_loras,
1,
lora_b_out_size,
lora_config.max_lora_rank,
dtype=lora_config.lora_dtype,
device=self.device,
) for _ in range(self.n_slices))
if lora_config.bias_enabled:
lora_bias_out_size = lora_b_out_size
self.lora_bias_stacked = tuple(
torch.zeros(
max_loras,
1,
lora_bias_out_size,
dtype=lora_config.lora_dtype,
device=self.device,
) for _ in range(self.n_slices))
self.output_slices = (self.lora_b_stacked[0].shape[2], )
def reset_lora(self, index: int): def reset_lora(self, index: int):
self.lora_a_stacked[index] = 0 for s_index in range(self.n_slices):
self.lora_b_stacked[index] = 0 self.lora_a_stacked[s_index][index] = 0
if self.lora_config.bias_enabled: self.lora_b_stacked[s_index][index] = 0
self.bias_stacked[index] = 0 if self.lora_config.bias_enabled:
# Make mypy happy
self.lora_bias_stacked = cast(Tuple[torch.Tensor, ...],
self.lora_bias_stacked)
self.lora_bias_stacked[s_index][index] = 0
def set_lora( def set_lora(
self, self,
@ -325,29 +355,56 @@ class ReplicatedLinearWithLoRA(BaseLayerWithLoRA):
lora_a: torch.Tensor, lora_a: torch.Tensor,
lora_b: torch.Tensor, lora_b: torch.Tensor,
embeddings_tensor: Optional[torch.Tensor], embeddings_tensor: Optional[torch.Tensor],
bias: Optional[torch.Tensor] = None, lora_bias: Optional[torch.Tensor] = None,
): ):
# Except for QKVParallelLinearWithLora and
# MergedColumnParallelLinearWithLoRA, all other linear LoRA layers
# store weights in a tuple of size 1. These two layers will
# override this function.
assert (len(self.lora_a_stacked) == len(self.lora_b_stacked) ==
self.n_slices == 1)
self.reset_lora(index) self.reset_lora(index)
if self.tp_size > 1:
lora_a = self.slice_lora_a(lora_a)
lora_b = self.slice_lora_b(lora_b)
if lora_bias is not None:
lora_bias = self.slice_bias(lora_bias)
self.lora_a_stacked[index, self.lora_a_stacked[0][index,
0, :lora_a.shape[1], :lora_a.shape[0]].copy_( 0, :lora_a.shape[1], :lora_a.shape[0]].copy_(
lora_a.T, non_blocking=True) lora_a.T, non_blocking=True)
self.lora_b_stacked[index, self.lora_b_stacked[0][index,
0, :lora_b.shape[1], :lora_b.shape[0]].copy_( 0, :lora_b.shape[1], :lora_b.shape[0]].copy_(
lora_b.T, non_blocking=True) lora_b.T, non_blocking=True)
if bias is not None: if lora_bias is not None:
self.bias_stacked[index,
0, :bias.shape[0]].copy_(bias.T,
non_blocking=True)
def apply(self, x: torch.Tensor, self.lora_bias_stacked = cast(Tuple[torch.Tensor, ...],
bias: Optional[torch.Tensor]) -> torch.Tensor: self.lora_bias_stacked)
assert len(self.lora_bias_stacked)
self.lora_bias_stacked[0][index, 0, :lora_bias.shape[0]].copy_(
lora_bias.T, non_blocking=True)
def apply(self,
x: torch.Tensor,
bias: Optional[torch.Tensor] = None) -> torch.Tensor:
output = self.base_layer.quant_method.apply(self.base_layer, x, bias) output = self.base_layer.quant_method.apply(self.base_layer, x, bias)
self.punica_wrapper.add_lora(output, x, self.lora_a_stacked, self.punica_wrapper.add_lora_linear(output, x, self.lora_a_stacked,
self.lora_b_stacked, self.bias_stacked, self.lora_b_stacked,
1.0) self.lora_bias_stacked, 1.0,
self.output_slices)
return output return output
class ReplicatedLinearWithLoRA(BaseLinearLayerWithLoRA):
def __init__(self, base_layer: ReplicatedLinear) -> None:
super().__init__(base_layer, )
# To ensure interface compatibility, set to 1 always.
self.tp_size = 1
self.output_size = self.base_layer.output_size
self.n_slices = 1
def forward(self, input_): def forward(self, input_):
"""Forward of ReplicatedLinearWithLoRA """Forward of ReplicatedLinearWithLoRA
@ -380,73 +437,26 @@ class ReplicatedLinearWithLoRA(BaseLayerWithLoRA):
return type(source_layer) is ReplicatedLinear return type(source_layer) is ReplicatedLinear
class ColumnParallelLinearWithLoRA(BaseLayerWithLoRA): class ColumnParallelLinearWithLoRA(BaseLinearLayerWithLoRA):
""" """
LoRA on top of ColumnParallelLinear layer. LoRA on top of ColumnParallelLinear layer.
LoRA B is sliced for tensor parallelism. LoRA B is sliced for tensor parallelism.
There are two types for the `base_layer`:
1. ColumnParallelLinear, e.g.`dense_h_to_4h` in `FalconForCausalLM`.
2. MergedColumnParallelLinear, e.g.`gate_up_proj` in `Phi3ForCausalLM`.
""" """
def __init__(self, base_layer: ColumnParallelLinear) -> None: def __init__(self, base_layer: ColumnParallelLinear) -> None:
super().__init__() super().__init__(base_layer)
# The base_layer type is ColumnParallelLinear or # The base_layer type is ColumnParallelLinear or
# MergedColumnParallelLinear, their weight sharding logic is # MergedColumnParallelLinear, their weight sharding logic is
# inconsistent when TP is greater than 1. # inconsistent when TP is greater than 1.
self.is_merged_col_linear = type( self.is_merged_col_linear = type(
base_layer) is MergedColumnParallelLinear base_layer) is MergedColumnParallelLinear
self.base_layer = base_layer
self.tp_size = get_tensor_model_parallel_world_size() self.tp_size = get_tensor_model_parallel_world_size()
self.input_size = self.base_layer.input_size
self.output_size = self.base_layer.output_size_per_partition self.output_size = self.base_layer.output_size_per_partition
self.device = _get_lora_device(self.base_layer) # There is only one LoRA layer
self.n_slices = 1
def create_lora_weights(
self,
max_loras: int,
lora_config: LoRAConfig,
model_config: Optional[PretrainedConfig] = None,
) -> None:
self.lora_config = lora_config
self.tp_size = get_tensor_model_parallel_world_size()
lora_a_output_size_per_partition = (
lora_config.max_lora_rank if not lora_config.fully_sharded_loras
else divide(lora_config.max_lora_rank, self.tp_size))
self.lora_a_stacked = torch.zeros(
max_loras,
1,
lora_a_output_size_per_partition,
self.input_size,
dtype=lora_config.lora_dtype,
device=self.device,
)
self.lora_b_stacked = torch.zeros(
max_loras,
1,
self.output_size,
lora_config.max_lora_rank,
dtype=lora_config.lora_dtype,
device=self.device,
)
if lora_config.bias_enabled:
self.bias_stacked = torch.zeros(
max_loras,
1,
self.output_size,
dtype=lora_config.lora_dtype,
device=self.device,
)
else:
self.bias_stacked = None
self.output_dim = self.lora_b_stacked.shape[2]
def reset_lora(self, index: int):
self.lora_a_stacked[index] = 0
self.lora_b_stacked[index] = 0
if self.lora_config.bias_enabled:
self.bias_stacked[index] = 0
def slice_lora_a(self, lora_a: torch.Tensor) -> torch.Tensor: def slice_lora_a(self, lora_a: torch.Tensor) -> torch.Tensor:
return lora_a return lora_a
@ -485,40 +495,6 @@ class ColumnParallelLinearWithLoRA(BaseLayerWithLoRA):
bias = bias[start_idx:end_idx] bias = bias[start_idx:end_idx]
return bias return bias
def set_lora(
self,
index: int,
lora_a: torch.Tensor,
lora_b: torch.Tensor,
embeddings_tensor: Optional[torch.Tensor],
bias: Optional[torch.Tensor] = None,
):
self.reset_lora(index)
if self.tp_size > 1:
lora_a = self.slice_lora_a(lora_a)
lora_b = self.slice_lora_b(lora_b)
bias = self.slice_bias(bias)
self.lora_a_stacked[index,
0, :lora_a.shape[1], :lora_a.shape[0]].copy_(
lora_a.T, non_blocking=True)
self.lora_b_stacked[index,
0, :lora_b.shape[1], :lora_b.shape[0]].copy_(
lora_b.T, non_blocking=True)
if bias is not None:
self.bias_stacked[index,
0, :bias.shape[0]].copy_(bias.T,
non_blocking=True)
def apply(self, x: torch.Tensor,
bias: Optional[torch.Tensor]) -> torch.Tensor:
output = self.base_layer.quant_method.apply(self.base_layer, x, bias)
self.punica_wrapper.add_lora(output, x, self.lora_a_stacked,
self.lora_b_stacked, self.bias_stacked,
1.0)
return output
def forward(self, input_): def forward(self, input_):
"""Forward of ColumnParallelLinear """Forward of ColumnParallelLinear
@ -568,6 +544,8 @@ class MergedColumnParallelLinearWithLoRA(ColumnParallelLinearWithLoRA):
def __init__(self, base_layer: MergedColumnParallelLinear) -> None: def __init__(self, base_layer: MergedColumnParallelLinear) -> None:
super().__init__(base_layer) super().__init__(base_layer)
# There are two LoRA layers
self.n_slices = len(self.base_layer.output_sizes)
def create_lora_weights( def create_lora_weights(
self, self,
@ -575,9 +553,13 @@ class MergedColumnParallelLinearWithLoRA(ColumnParallelLinearWithLoRA):
lora_config: LoRAConfig, lora_config: LoRAConfig,
model_config: Optional[PretrainedConfig] = None, model_config: Optional[PretrainedConfig] = None,
) -> None: ) -> None:
"""
The main reason for overriding this function is to enhance code
maintainability.
"""
self.lora_config = lora_config self.lora_config = lora_config
n_slices = 2
if not (len(self.base_layer.output_sizes) == n_slices if not (len(self.base_layer.output_sizes) == self.n_slices == 2
and self.base_layer.output_sizes[0] and self.base_layer.output_sizes[0]
== self.base_layer.output_sizes[1]): == self.base_layer.output_sizes[1]):
raise ValueError( raise ValueError(
@ -598,7 +580,7 @@ class MergedColumnParallelLinearWithLoRA(ColumnParallelLinearWithLoRA):
self.input_size, self.input_size,
dtype=lora_config.lora_dtype, dtype=lora_config.lora_dtype,
device=self.device, device=self.device,
) for _ in range(n_slices)) ) for _ in range(self.n_slices))
self.lora_b_stacked = tuple( self.lora_b_stacked = tuple(
torch.zeros( torch.zeros(
max_loras, max_loras,
@ -607,30 +589,19 @@ class MergedColumnParallelLinearWithLoRA(ColumnParallelLinearWithLoRA):
lora_config.max_lora_rank, lora_config.max_lora_rank,
dtype=lora_config.lora_dtype, dtype=lora_config.lora_dtype,
device=self.device, device=self.device,
) for _ in range(n_slices)) ) for _ in range(self.n_slices))
if lora_config.bias_enabled: if lora_config.bias_enabled:
self.bias_stacked = tuple( self.lora_bias_stacked = tuple(
torch.zeros( torch.zeros(
max_loras, max_loras,
1, 1,
self.output_size // 2, self.output_size // 2,
dtype=lora_config.lora_dtype, dtype=lora_config.lora_dtype,
device=self.device, device=self.device,
) for _ in range(n_slices)) ) for _ in range(self.n_slices))
else:
self.bias_stacked = None
self.output_dim = self.lora_b_stacked[0].shape[2] self.output_dim = self.lora_b_stacked[0].shape[2]
self.output_slices = (self.output_dim, self.output_dim) self.output_slices = (self.output_dim, self.output_dim)
def reset_lora(self, index: int):
self.lora_a_stacked[0][index] = 0
self.lora_a_stacked[1][index] = 0
self.lora_b_stacked[0][index] = 0
self.lora_b_stacked[1][index] = 0
if self.lora_config.bias_enabled:
self.bias_stacked[0][index] = 0
self.bias_stacked[1][index] = 0
def slice_lora_a( def slice_lora_a(
self, lora_a: List[Union[torch.Tensor, None]] self, lora_a: List[Union[torch.Tensor, None]]
) -> List[Union[torch.Tensor, None]]: ) -> List[Union[torch.Tensor, None]]:
@ -668,15 +639,15 @@ class MergedColumnParallelLinearWithLoRA(ColumnParallelLinearWithLoRA):
lora_a: torch.Tensor, lora_a: torch.Tensor,
lora_b: torch.Tensor, lora_b: torch.Tensor,
embeddings_tensor: Optional[torch.Tensor], embeddings_tensor: Optional[torch.Tensor],
bias: Optional[torch.Tensor] = None, lora_bias: Optional[torch.Tensor] = None,
): ):
self.reset_lora(index) self.reset_lora(index)
if self.tp_size > 1: if self.tp_size > 1:
lora_a = self.slice_lora_a(lora_a) lora_a = self.slice_lora_a(lora_a)
lora_b = self.slice_lora_b(lora_b) lora_b = self.slice_lora_b(lora_b)
if bias is not None: if lora_bias is not None:
bias = self.slice_bias(bias) lora_bias = self.slice_bias(lora_bias)
if lora_a[0] is not None: if lora_a[0] is not None:
self.lora_a_stacked[0][ self.lora_a_stacked[0][
@ -685,10 +656,11 @@ class MergedColumnParallelLinearWithLoRA(ColumnParallelLinearWithLoRA):
self.lora_b_stacked[0][ self.lora_b_stacked[0][
index, 0, :lora_b[0].shape[1], :lora_b[0].shape[0]].copy_( index, 0, :lora_b[0].shape[1], :lora_b[0].shape[0]].copy_(
lora_b[0].T, non_blocking=True) lora_b[0].T, non_blocking=True)
if bias is not None and bias[0] is not None: if lora_bias is not None and lora_bias[0] is not None:
self.bias_stacked[0][index, self.lora_bias_stacked = cast(Tuple[torch.Tensor, ...],
0, :bias[0].shape[0]].copy_(bias[0].T, self.lora_bias_stacked)
non_blocking=True) self.lora_bias_stacked[0][index, 0, :lora_bias[0].shape[0]].copy_(
lora_bias[0].T, non_blocking=True)
if lora_a[1] is not None: if lora_a[1] is not None:
self.lora_a_stacked[1][ self.lora_a_stacked[1][
index, 0, :lora_a[1].shape[1], :lora_a[1].shape[0]].copy_( index, 0, :lora_a[1].shape[1], :lora_a[1].shape[0]].copy_(
@ -696,18 +668,11 @@ class MergedColumnParallelLinearWithLoRA(ColumnParallelLinearWithLoRA):
self.lora_b_stacked[1][ self.lora_b_stacked[1][
index, 0, :lora_b[1].shape[1], :lora_b[1].shape[0]].copy_( index, 0, :lora_b[1].shape[1], :lora_b[1].shape[0]].copy_(
lora_b[1].T, non_blocking=True) lora_b[1].T, non_blocking=True)
if bias is not None and bias[1] is not None: if lora_bias is not None and lora_bias[1] is not None:
self.bias_stacked[1][index, self.lora_bias_stacked = cast(Tuple[torch.Tensor, ...],
0, :bias[1].shape[0]].copy_(bias[1].T, self.lora_bias_stacked)
non_blocking=True) self.lora_bias_stacked[1][index, 0, :lora_bias[1].shape[0]].copy_(
lora_bias[1].T, non_blocking=True)
def apply(self, x: torch.Tensor,
bias: Optional[torch.Tensor]) -> torch.Tensor:
output = self.base_layer.quant_method.apply(self.base_layer, x, bias)
self.punica_wrapper.add_lora_packed_nslice(
output, x, self.lora_a_stacked, self.lora_b_stacked,
self.bias_stacked, 1.0, (self.output_dim, self.output_dim))
return output
@classmethod @classmethod
@_not_fully_sharded_can_replace @_not_fully_sharded_can_replace
@ -737,7 +702,6 @@ class QKVParallelLinearWithLora(ColumnParallelLinearWithLoRA):
def __init__(self, base_layer: QKVParallelLinear) -> None: def __init__(self, base_layer: QKVParallelLinear) -> None:
super().__init__(base_layer) super().__init__(base_layer)
self.tp_size = get_tensor_model_parallel_world_size()
self.q_proj_total_size = (self.base_layer.total_num_heads * self.q_proj_total_size = (self.base_layer.total_num_heads *
self.base_layer.head_size) self.base_layer.head_size)
self.q_proj_shard_size = (self.base_layer.num_heads * self.q_proj_shard_size = (self.base_layer.num_heads *
@ -746,6 +710,8 @@ class QKVParallelLinearWithLora(ColumnParallelLinearWithLoRA):
self.base_layer.head_size) self.base_layer.head_size)
self.kv_proj_total_size = (self.base_layer.total_num_kv_heads * self.kv_proj_total_size = (self.base_layer.total_num_kv_heads *
self.base_layer.head_size) self.base_layer.head_size)
# There is only one LoRA layer
self.n_slices = 1
def slice_lora_b(self, lora_b: torch.Tensor) -> torch.Tensor: def slice_lora_b(self, lora_b: torch.Tensor) -> torch.Tensor:
tp_rank = get_tensor_model_parallel_rank() tp_rank = get_tensor_model_parallel_rank()
@ -780,32 +746,6 @@ class QKVParallelLinearWithLora(ColumnParallelLinearWithLoRA):
bias = torch.cat([bias_q, bias_k, bias_v], dim=1) bias = torch.cat([bias_q, bias_k, bias_v], dim=1)
return bias return bias
def set_lora(
self,
index: int,
lora_a: torch.Tensor,
lora_b: torch.Tensor,
embeddings_tensor: Optional[torch.Tensor],
bias: Optional[torch.Tensor] = None,
):
self.reset_lora(index)
if self.tp_size > 1:
lora_a = self.slice_lora_a(lora_a)
lora_b = self.slice_lora_b(lora_b)
if bias is not None:
bias = self.slice_bias(bias)
self.lora_a_stacked[index,
0, :lora_a.shape[1], :lora_a.shape[0]].copy_(
lora_a.T, non_blocking=True)
self.lora_b_stacked[index,
0, :lora_b.shape[1], :lora_b.shape[0]].copy_(
lora_b.T, non_blocking=True)
if bias is not None:
self.bias_stacked[index,
0, :bias.shape[0]].copy_(bias.T,
non_blocking=True)
@classmethod @classmethod
@_not_fully_sharded_can_replace @_not_fully_sharded_can_replace
def can_replace_layer(cls, source_layer: nn.Module, def can_replace_layer(cls, source_layer: nn.Module,
@ -828,6 +768,10 @@ class MergedQKVParallelLinearWithLora(ColumnParallelLinearWithLoRA):
def __init__(self, base_layer: QKVParallelLinear) -> None: def __init__(self, base_layer: QKVParallelLinear) -> None:
super().__init__(base_layer) super().__init__(base_layer)
# There are three LoRA layer.
self.n_slices = len(self.base_layer.output_sizes)
self.tp_size = get_tensor_model_parallel_world_size()
self.tp_rank = get_tensor_model_parallel_rank()
def create_lora_weights( def create_lora_weights(
self, self,
@ -835,9 +779,16 @@ class MergedQKVParallelLinearWithLora(ColumnParallelLinearWithLoRA):
lora_config: LoRAConfig, lora_config: LoRAConfig,
model_config: Optional[PretrainedConfig] = None, model_config: Optional[PretrainedConfig] = None,
) -> None: ) -> None:
"""
The main reason for overloading this function is to handle inconsistent
weight dimensions in qkv lora.
"""
self.lora_config = lora_config self.lora_config = lora_config
self.tp_size = get_tensor_model_parallel_world_size()
self.tp_rank = get_tensor_model_parallel_rank() if not (len(self.base_layer.output_sizes) == self.n_slices == 3):
raise ValueError(
"LoRAColumnParallelLinear3Slice requires 3 slices.")
self.q_proj_shard_size = (self.base_layer.num_heads * self.q_proj_shard_size = (self.base_layer.num_heads *
self.base_layer.head_size) self.base_layer.head_size)
self.kv_proj_shard_size = (self.base_layer.num_kv_heads * self.kv_proj_shard_size = (self.base_layer.num_kv_heads *
@ -902,7 +853,7 @@ class MergedQKVParallelLinearWithLora(ColumnParallelLinearWithLoRA):
), ),
) )
if lora_config.bias_enabled: if lora_config.bias_enabled:
self.bias_stacked = ( self.lora_bias_stacked = (
torch.zeros( torch.zeros(
max_loras, max_loras,
1, 1,
@ -925,9 +876,6 @@ class MergedQKVParallelLinearWithLora(ColumnParallelLinearWithLoRA):
device=self.device, device=self.device,
), ),
) )
else:
self.bias_stacked = None
self.output_slices = ( self.output_slices = (
self.q_proj_shard_size, self.q_proj_shard_size,
self.kv_proj_shard_size, self.kv_proj_shard_size,
@ -939,18 +887,6 @@ class MergedQKVParallelLinearWithLora(ColumnParallelLinearWithLoRA):
self.indices: torch.Tensor self.indices: torch.Tensor
self.indices_len: List[int] self.indices_len: List[int]
def reset_lora(self, index: int):
self.lora_a_stacked[0][index] = 0
self.lora_b_stacked[0][index] = 0
self.lora_a_stacked[1][index] = 0
self.lora_b_stacked[1][index] = 0
self.lora_a_stacked[2][index] = 0
self.lora_b_stacked[2][index] = 0
if self.lora_config.bias_enabled:
self.bias_stacked[0][index] = 0
self.bias_stacked[1][index] = 0
self.bias_stacked[2][index] = 0
def slice_lora_a( def slice_lora_a(
self, lora_a: List[Union[torch.Tensor, None]] self, lora_a: List[Union[torch.Tensor, None]]
) -> List[Union[torch.Tensor, None]]: ) -> List[Union[torch.Tensor, None]]:
@ -1000,15 +936,15 @@ class MergedQKVParallelLinearWithLora(ColumnParallelLinearWithLoRA):
lora_a: torch.Tensor, lora_a: torch.Tensor,
lora_b: torch.Tensor, lora_b: torch.Tensor,
embeddings_tensor: Optional[torch.Tensor], embeddings_tensor: Optional[torch.Tensor],
bias: Optional[torch.Tensor] = None, lora_bias: Optional[torch.Tensor] = None,
): ):
self.reset_lora(index) self.reset_lora(index)
if self.tp_size > 1: if self.tp_size > 1:
lora_a = self.slice_lora_a(lora_a) lora_a = self.slice_lora_a(lora_a)
lora_b = self.slice_lora_b(lora_b) lora_b = self.slice_lora_b(lora_b)
if bias is not None: if lora_bias is not None:
bias = self.slice_bias(bias) lora_bias = self.slice_bias(lora_bias)
if lora_b[0] is not None: if lora_b[0] is not None:
lora_b_q = lora_b[0] lora_b_q = lora_b[0]
@ -1039,26 +975,24 @@ class MergedQKVParallelLinearWithLora(ColumnParallelLinearWithLoRA):
index, 0, :lora_a[2].shape[1], :lora_a[2].shape[0]].copy_( index, 0, :lora_a[2].shape[1], :lora_a[2].shape[0]].copy_(
lora_a[2].T, non_blocking=True) lora_a[2].T, non_blocking=True)
if bias is not None: if lora_bias is not None:
if bias[0] is not None: self.lora_bias_stacked = cast(Tuple[torch.Tensor, ...],
self.bias_stacked[0][index, 0, :bias[0].shape[0]].copy_( self.lora_bias_stacked)
bias[0].T, non_blocking=True) if lora_bias[0] is not None:
if bias[1] is not None: self.lora_bias_stacked[0][index,
self.bias_stacked[1][index, 0, :bias[1].shape[0]].copy_( 0, :lora_bias[0].shape[0]].copy_(
bias[1].T, non_blocking=True) lora_bias[0].T,
if bias[2] is not None: non_blocking=True)
self.bias_stacked[2][index, 0, :bias[2].shape[0]].copy_( if lora_bias[1] is not None:
bias[2].T, non_blocking=True) self.lora_bias_stacked[1][index,
0, :lora_bias[1].shape[0]].copy_(
def apply(self, x: torch.Tensor, lora_bias[1].T,
bias: Optional[torch.Tensor]) -> torch.Tensor: non_blocking=True)
output = self.base_layer.quant_method.apply(self.base_layer, x, bias) if lora_bias[2] is not None:
self.punica_wrapper.add_lora_packed_nslice(output, x, self.lora_bias_stacked[2][index,
self.lora_a_stacked, 0, :lora_bias[2].shape[0]].copy_(
self.lora_b_stacked, lora_bias[2].T,
self.bias_stacked, 1.0, non_blocking=True)
self.output_slices)
return output
@classmethod @classmethod
@_not_fully_sharded_can_replace @_not_fully_sharded_can_replace
@ -1073,76 +1007,25 @@ class MergedQKVParallelLinearWithLora(ColumnParallelLinearWithLoRA):
and len(packed_modules_list) == 3) and len(packed_modules_list) == 3)
class RowParallelLinearWithLoRA(BaseLayerWithLoRA): class RowParallelLinearWithLoRA(BaseLinearLayerWithLoRA):
def __init__(self, base_layer: RowParallelLinear) -> None: def __init__(self, base_layer: RowParallelLinear) -> None:
super().__init__() super().__init__(base_layer)
self.base_layer = base_layer
self.tp_size = get_tensor_model_parallel_world_size()
# reset input_size
self.input_size = self.base_layer.input_size_per_partition self.input_size = self.base_layer.input_size_per_partition
self.output_size = self.base_layer.output_size self.output_size = self.base_layer.output_size
self.device = _get_lora_device(self.base_layer)
def create_lora_weights(
self,
max_loras: int,
lora_config: LoRAConfig,
model_config: Optional[PretrainedConfig] = None,
) -> None:
self.lora_config = lora_config
self.tp_rank = get_tensor_model_parallel_rank() self.tp_rank = get_tensor_model_parallel_rank()
self.lora_a_stacked = torch.zeros( # There is only one LoRA layer.
( self.n_slices = 1
max_loras,
1,
lora_config.max_lora_rank,
self.input_size,
),
dtype=lora_config.lora_dtype,
device=self.device,
)
tp_size = get_tensor_model_parallel_world_size()
lora_b_output_size_per_partition = (
self.output_size if not lora_config.fully_sharded_loras else
divide(self.output_size, tp_size))
self.lora_b_stacked = torch.zeros(
(
max_loras,
1,
lora_b_output_size_per_partition,
lora_config.max_lora_rank,
),
dtype=lora_config.lora_dtype,
device=self.device,
)
if lora_config.bias_enabled:
self.bias_stacked = torch.zeros(
(
max_loras,
1,
self.output_size,
),
dtype=lora_config.lora_dtype,
device=self.device,
)
else:
self.bias_stacked = None
# Lazily initialized
self.indices: torch.Tensor
self.indices_len: List[int]
def reset_lora(self, index: int):
self.lora_a_stacked[index] = 0
self.lora_b_stacked[index] = 0
if self.lora_config.bias_enabled:
self.bias_stacked[index] = 0
def slice_lora_a(self, lora_a: torch.Tensor) -> torch.Tensor: def slice_lora_a(self, lora_a: torch.Tensor) -> torch.Tensor:
tensor_model_parallel_rank = get_tensor_model_parallel_rank()
shard_size = self.input_size shard_size = self.input_size
start_idx = tensor_model_parallel_rank * shard_size start_idx = self.tp_rank * shard_size
end_idx = (tensor_model_parallel_rank + 1) * shard_size end_idx = (self.tp_rank + 1) * shard_size
lora_a = lora_a[start_idx:end_idx, :] lora_a = lora_a[start_idx:end_idx, :]
return lora_a return lora_a
@ -1152,40 +1035,6 @@ class RowParallelLinearWithLoRA(BaseLayerWithLoRA):
def slice_bias(self, bias: torch.Tensor) -> torch.Tensor: def slice_bias(self, bias: torch.Tensor) -> torch.Tensor:
return bias return bias
def set_lora(
self,
index: int,
lora_a: torch.Tensor,
lora_b: torch.Tensor,
embeddings_tensor: Optional[torch.Tensor],
bias: Optional[torch.Tensor] = None,
):
self.reset_lora(index)
if self.base_layer.tp_size > 1:
lora_a = self.slice_lora_a(lora_a)
lora_b = self.slice_lora_b(lora_b)
if bias is not None:
bias = self.slice_bias(bias)
self.lora_a_stacked[index,
0, :lora_a.shape[1], :lora_a.shape[0]].copy_(
lora_a.T, non_blocking=True)
self.lora_b_stacked[index,
0, :lora_b.shape[1], :lora_b.shape[0]].copy_(
lora_b.T, non_blocking=True)
if bias is not None:
self.bias_stacked[index,
0, :bias.shape[0]].copy_(bias.T,
non_blocking=True)
def apply(self, x: torch.Tensor) -> torch.Tensor:
output = self.base_layer.quant_method.apply(self.base_layer, x)
self.punica_wrapper.add_lora(output, x, self.lora_a_stacked,
self.lora_b_stacked, self.bias_stacked,
1.0)
return output
def forward(self, input_): def forward(self, input_):
"""Forward of RowParallelLinear """Forward of RowParallelLinear
@ -1203,10 +1052,9 @@ class RowParallelLinearWithLoRA(BaseLayerWithLoRA):
input_parallel = input_ input_parallel = input_
else: else:
# TODO: simplify code below # TODO: simplify code below
tp_rank = get_tensor_model_parallel_rank()
splitted_input = split_tensor_along_last_dim( splitted_input = split_tensor_along_last_dim(
input_, num_partitions=self.base_layer.tp_size) input_, num_partitions=self.base_layer.tp_size)
input_parallel = splitted_input[tp_rank].contiguous() input_parallel = splitted_input[self.tp_rank].contiguous()
# Matrix multiply. # Matrix multiply.
output_parallel = self.apply(input_parallel) output_parallel = self.apply(input_parallel)

8
vllm/lora/models.py
View File

@ -555,17 +555,17 @@ class LoRAModelManager(AdapterModelManager):
input_dim, input_dim,
output_dim, output_dim,
rank, rank,
module.lora_a_stacked.dtype, module.lora_a_stacked[0].dtype,
"cpu", "cpu",
embeddings_tensor_dim=embeddings_tensor_dim, embeddings_tensor_dim=embeddings_tensor_dim,
bias_enabled=bias_enabled) bias_enabled=bias_enabled)
else: else:
lora = LoRALayerWeights.create_dummy_lora_weights( lora = LoRALayerWeights.create_dummy_lora_weights(
module_name, module_name,
module.lora_a_stacked.shape[-1], module.lora_a_stacked[0].shape[-1],
module.lora_b_stacked.shape[-2], module.lora_b_stacked[0].shape[-2],
rank, rank,
module.lora_a_stacked.dtype, module.lora_a_stacked[0].dtype,
"cpu", "cpu",
bias_enabled=bias_enabled, bias_enabled=bias_enabled,
) )

383
vllm/lora/punica.py
View File

@ -362,7 +362,7 @@ class PunicaWrapper:
long_lora_len = self.indices_len[4] long_lora_len = self.indices_len[4]
return self._long_lora_indices[:long_lora_len] return self._long_lora_indices[:long_lora_len]
def shrink_prefill( def _shrink_prefill(
self, self,
y: torch.Tensor, y: torch.Tensor,
x: torch.Tensor, x: torch.Tensor,
@ -380,7 +380,7 @@ class PunicaWrapper:
scale, scale,
) )
def shrink_decode( def _shrink_decode(
self, self,
y: torch.Tensor, y: torch.Tensor,
x: torch.Tensor, x: torch.Tensor,
@ -389,7 +389,7 @@ class PunicaWrapper:
): ):
bgmv_shrink(x, w_t_all, y, self.token_lora_indices, scale) bgmv_shrink(x, w_t_all, y, self.token_lora_indices, scale)
def expand_prefill( def _expand_prefill(
self, self,
y: torch.Tensor, y: torch.Tensor,
x: torch.Tensor, x: torch.Tensor,
@ -407,7 +407,7 @@ class PunicaWrapper:
add_input, add_input,
) )
def expand_decode( def _expand_decode(
self, self,
y: torch.Tensor, y: torch.Tensor,
x: torch.Tensor, x: torch.Tensor,
@ -416,7 +416,7 @@ class PunicaWrapper:
): ):
bgmv_expand(x, w_t_all, y, self.token_lora_indices, add_input) bgmv_expand(x, w_t_all, y, self.token_lora_indices, add_input)
def expand_slice_prefill( def _expand_slice_prefill(
self, self,
y: torch.Tensor, y: torch.Tensor,
x: torch.Tensor, x: torch.Tensor,
@ -438,7 +438,7 @@ class PunicaWrapper:
add_input, add_input,
) )
def expand_slice_decode( def _expand_slice_decode(
self, self,
y: torch.Tensor, y: torch.Tensor,
x: torch.Tensor, x: torch.Tensor,
@ -450,41 +450,35 @@ class PunicaWrapper:
bgmv_expand_slice(x, w_t_all, y, self.token_lora_indices, y_offset, bgmv_expand_slice(x, w_t_all, y, self.token_lora_indices, y_offset,
y_slice_size, add_input) y_slice_size, add_input)
def apply_bias( def _apply_expand(self,
self, y: torch.Tensor,
indices: torch.Tensor, x: torch.Tensor,
output: torch.Tensor, w_t_all: torch.Tensor,
bias_stacked: torch.Tensor, y_offset: Optional[int],
): y_slice_size: Optional[int],
"""Applies bias to output add_input: bool = True):
"""
Input shapes: Perform the ` y[:,y_offset:y_offset+y_slice_size]+=x@w_t_all`
bias_stacked: (num_loras, output_dim) computation, which is suitable for the
indices: (batch_size) GEMM of lora'b.
output: (batch_size, output_dim)
""" """
org_output = output
output = output.view(-1, output.shape[-1])
indices = indices.view(-1)
bias_stacked = bias_stacked.view(-1, bias_stacked.shape[-1]) expand_slice_fun: Callable = (self._expand_slice_prefill
bias_stacked = bias_stacked[indices] if self.is_prefill else
bias_stacked[indices == -1] = 0 self._expand_slice_decode)
output += bias_stacked expand_slice_fun(y, x, w_t_all, y_offset, y_slice_size, add_input)
return output.view_as(org_output) def _apply_bias(
def apply_bias_packed_nslice(
self, self,
indices: torch.Tensor, indices: torch.Tensor,
output: torch.Tensor, output: torch.Tensor,
output_slices: Tuple[int, ...], output_slices: Tuple[int, ...],
bias_stacked: Tuple[Optional[torch.Tensor], ...], lora_bias_stacked: Tuple[Optional[torch.Tensor], ...],
): ):
"""Applies bias to output """Applies bias to output
Input shapes: Input shapes:
bias_stacked: 3 element tuple of (num_loras, output_dim) lora_bias_stacked: 3 element tuple of (num_loras, output_dim)
indices: (batch_size) indices: (batch_size)
output: (batch_size, q_slice_size + 2*kv_slice_size) output: (batch_size, q_slice_size + 2*kv_slice_size)
output_slices: n-1 element tuple of (slice_size...), output_slices: n-1 element tuple of (slice_size...),
@ -496,7 +490,7 @@ class PunicaWrapper:
offset_left = 0 offset_left = 0
for slice_idx, slice in enumerate(output_slices): for slice_idx, slice in enumerate(output_slices):
bias = bias_stacked[slice_idx] bias = lora_bias_stacked[slice_idx]
if bias is not None: if bias is not None:
bias = bias.view(-1, bias.shape[-1]) bias = bias.view(-1, bias.shape[-1])
bias = bias[indices] bias = bias[indices]
@ -506,7 +500,7 @@ class PunicaWrapper:
return output.view_as(org_output) return output.view_as(org_output)
def add_shrink( def _apply_shrink(
self, self,
y: torch.Tensor, y: torch.Tensor,
x: torch.Tensor, x: torch.Tensor,
@ -517,188 +511,215 @@ class PunicaWrapper:
Perform the ` y+=x@w_t_all` computation, which is suitable for the Perform the ` y+=x@w_t_all` computation, which is suitable for the
GEMM of lora'a. GEMM of lora'a.
When `is_prefill is` true, it indicates that it is currently the When `is_prefill is` true, it indicates that it is currently the
prefill stage, and the `shrink_prefill` function should be called. prefill stage, and the `_shrink_prefill` function should be called.
Otherwise, it is the decode stage, and the shrink_decode function Otherwise, it is the decode stage, and the _shrink_decode function
should be called. should be called.
""" """
shrink_fun: Callable = (self.shrink_prefill y_org = y
if self.is_prefill else self.shrink_decode) y = y.view(-1, y.shape[-1])
shrink_fun: Callable = (self._shrink_prefill
if self.is_prefill else self._shrink_decode)
shrink_fun(y, x, w_t_all, scale) shrink_fun(y, x, w_t_all, scale)
y = y.view_as(y_org)
def add_shrink(
self,
y: Union[Tuple[torch.Tensor, ...], torch.Tensor],
x: torch.Tensor,
lora_a_stacked: Tuple[torch.Tensor, ...],
scale: float,
):
"""
Performs GEMM for multiple slices of lora_a.
When `is_prefill is` true, it indicates that it is currently the
prefill stage, and the `_shrink_prefill` function should be called.
Otherwise, it is the decode stage, and the _shrink_decode function
should be called.
Semantics:
for i in range(len(lora_a_stacked)):
y[i] += (x @ lora_a_stacked[i]) * scale
Args:
y (Union[Tuple[torch.Tensor, ...], torch.Tensor]): Output tensors
x (torch.Tensor): Input tensor
lora_a_stacked (Tuple[torch.Tensor, ...]): lora_a's weights
scale (float): Scaling factor for the operation
"""
x = x.view(-1, x.shape[-1])
# TODO fuse these kernels
for slice_idx in range(len(lora_a_stacked)):
self._apply_shrink(y[slice_idx], x, lora_a_stacked[slice_idx],
scale)
def add_expand( def add_expand(
self,
y: torch.Tensor,
x: Union[Tuple[torch.Tensor, ...], torch.Tensor],
lora_b_stacked: Tuple[torch.Tensor, ...],
lora_bias_stacked: Optional[Tuple[torch.Tensor, ...]],
output_slices: Tuple[int, ...],
offset_start: int = 0,
add_input=True,
) -> None:
"""
Performs GEMM and bias addition for multiple slices of lora_b.
Semantics:
for i in range(len(lora_b_stacked)):
slice = output_slices[i]
y[:, offset:offset+slice] += x[i] @ lora_b_stacked[i] +
lora_bias_stacked[i]
offset += slice
Args:
y (torch.Tensor): Output tensor.
x (Union[Tuple[torch.Tensor, ...], torch.Tensor]): Input tensors
lora_b_stacked (Tuple[torch.Tensor, ...]): lora_b's weight
lora_bias_stacked (Optional[Tuple[torch.Tensor, ...]]):
bias's weight
output_slices (Tuple[int, ...]): Every slice's size
add_input (bool): Defaults to True.
"""
y_org = y
y = y.view(-1, y.shape[-1])
offset_left = offset_start
if lora_bias_stacked is not None:
self._apply_bias(self.token_lora_indices, y, output_slices,
lora_bias_stacked)
for slice_idx in range(len(lora_b_stacked)):
self._apply_expand(
y,
x[slice_idx],
lora_b_stacked[slice_idx],
offset_left,
output_slices[slice_idx],
add_input=add_input,
)
offset_left += output_slices[slice_idx]
y = y.view_as(y_org)
def add_lora_embedding(
self, self,
y: torch.Tensor, y: torch.Tensor,
x: torch.Tensor, x: torch.Tensor,
w_t_all: torch.Tensor, lora_b_stacked: torch.Tensor,
bias_all: Optional[torch.Tensor],
add_input: bool = True, add_input: bool = True,
): ):
""" """
Perform the ` y+=x@w_t_all+bias` computation, which is suitable for the Applies lora specifically for VocabParallelEmbeddingWithLoRA.
GEMM of lora'b.
When `is_prefill` is true, it indicates that it is currently the
prefill stage, and the `expand_prefill` function should be called.
Otherwise, it is the decode stage, and the expand_decode function
should be called.
"""
if bias_all is not None:
y = self.apply_bias(self.token_lora_indices, y, bias_all)
expand_fun: Callable = (self.expand_prefill
if self.is_prefill else self.expand_decode)
expand_fun(y, x, w_t_all, add_input)
def add_expand_slice(self,
y: torch.Tensor,
x: torch.Tensor,
w_t_all: torch.Tensor,
bias_all: Optional[torch.Tensor],
y_offset: Optional[int],
y_slice_size: Optional[int],
add_input: bool = True):
"""
Similar to `add_expand`
"""
if bias_all is not None:
y = self.apply_bias(self.token_lora_indices, y, bias_all)
expand_slice_fun: Callable = (self.expand_slice_prefill
if self.is_prefill else
self.expand_slice_decode)
expand_slice_fun(y, x, w_t_all, y_offset, y_slice_size, add_input)
def add_expand_packed_nslice(self, y: torch.Tensor, x: torch.Tensor,
lora_b_stacked: Tuple[torch.Tensor, ...],
bias_stacked: Optional[Tuple[torch.Tensor,
...]],
scale: float,
output_slices: Tuple[int, ...]) -> None:
"""
Similar to `add_expand`
"""
y_org = y
y = y.view(-1, y.shape[-1])
offset_left = 0
if bias_stacked is not None:
self.apply_bias_packed_nslice(self.token_lora_indices, y,
output_slices, bias_stacked)
for slice_idx in range(len(lora_b_stacked)):
self.add_expand_slice(y,
x[slice_idx],
lora_b_stacked[slice_idx],
None,
offset_left,
output_slices[slice_idx],
add_input=True)
offset_left += output_slices[slice_idx]
y = y.view_as(y_org)
def add_lora(self,
y: torch.Tensor,
x: torch.Tensor,
wa_t_all: torch.Tensor,
wb_t_all: torch.Tensor,
bias_all: Optional[torch.Tensor],
scale: float,
y_offset: Optional[int] = None,
y_slice_size: Optional[int] = None,
*,
buffer: Optional[torch.Tensor] = None) -> None:
"""
Semantics: Semantics:
y[i] += ( y += x @ lora_b_stacked
x[i].unsqueeze(0)
@ wa_t_all[indices[i], layer_idx, :, :].transpose(-1, -2)
@ wb_t_all[indices[i], layer_idx, :, :].transpose(-1, -2)
* scale
).squeeze(0)+bias[i]
Args: Args:
y (torch.Tensor): Output tensor. Will be changed in-place. y (torch.Tensor): Output tensor.
x (torch.Tensor): Input tensor x (torch.Tensor): Input tensor.
wa_t_all (torch.Tensor): lora_a's weight lora_b_stacked (torch.Tensor): lora_b's weights.
wb_t_all (torch.Tensor): lora_b's weight add_input (bool): Default to True.
bias_all: (torch.Tensor): lora's bias
scale (float): Scaling factor.
y_offset (Optional[int], optional): Offset to apply to the starting
column of y.
y_slice_size (Optional[int], optional): Size of the y column slice.
buffer (Optional[torch.Tensor], optional): Defaults to None.
""" """
y_org = y
y = y.view(-1, y.shape[-1]) # Embedding layer only need expand op
x = x.view(-1, x.shape[-1]) expand_fun: Callable = (self._expand_prefill
r = wb_t_all.size(-1) if self.is_prefill else self._expand_decode)
expand_fun(y, x, lora_b_stacked, add_input)
def add_lora_linear(
self,
y: torch.Tensor,
x: torch.Tensor,
lora_a_stacked: Tuple[torch.Tensor, ...],
lora_b_stacked: Tuple[torch.Tensor, ...],
lora_bias_stacked: Optional[Tuple[torch.Tensor, ...]],
scale: float,
output_slices: Tuple[int, ...],
*,
buffer: Optional[Tuple[torch.Tensor, ...]] = None) -> None:
"""
Applicable to linear-related lora.
Semantics:
for i in range(len(lora_a_stacked)):
y[i] += (
x[i].unsqueeze(0)
@ lora_a_stacked[indices[i], layer_idx, :, :]
@ lora_b_stacked[indices[i], layer_idx, :, :]
* scale
).squeeze(0)+lora_bias_stacked[i]
Args:
y (torch.Tensor): Output tensor. Will be changed in-place.
x (torch.Tensor): Input tensor
lora_a_stacked (Tuple[torch.Tensor, ...]): lora_a's weight.
lora_b_stacked (Tuple[torch.Tensor, ...]): lora_b's weight.
lora_bias_stacked (Optional[Tuple[torch.Tensor, ...]]): lora's bias.
scale (float): Scaling factor.
output_slices (Tuple[int, ...]): Every slice's size.
buffer (Optional[Tuple[torch.Tensor, ...]]): Defaults to None.
"""
assert len(lora_a_stacked) == len(lora_b_stacked) == len(output_slices)
if lora_bias_stacked is not None:
assert len(lora_bias_stacked) == len(output_slices)
y = self._apply_bias(self.token_lora_indices, y, output_slices,
lora_bias_stacked)
if buffer is None: if buffer is None:
r = lora_b_stacked[0].size(-1)
# We set the buffer to be float32 by default ,refer to: # We set the buffer to be float32 by default ,refer to:
# https://github.com/triton-lang/triton/issues/1387 # https://github.com/triton-lang/triton/issues/1387
buffer = torch.zeros((x.size(0), r), buffer = tuple(
dtype=torch.float32, torch.zeros(
device=x.device) (x.size(0), r), dtype=torch.float32, device=x.device)
if bias_all is not None: for _ in range(len(output_slices)))
y = self.apply_bias(self.token_lora_indices, y, bias_all) self.add_shrink(buffer, x, lora_a_stacked, scale)
self.add_shrink(buffer, x, wa_t_all, scale) self.add_expand(y,
if y_offset is None and y_slice_size is None: buffer,
self.add_expand(y, buffer, wb_t_all, bias_all=None, add_input=True) lora_b_stacked,
else: None,
self.add_expand_slice(y, output_slices,
buffer, add_input=True)
wb_t_all,
None,
y_offset,
y_slice_size,
add_input=True)
y = y.view_as(y_org)
def add_lora_packed_nslice(self, y: torch.Tensor, x: torch.Tensor,
lora_a_stacked: Tuple[torch.Tensor, ...],
lora_b_stacked: Tuple[torch.Tensor, ...],
bias_all: Tuple[Optional[torch.Tensor],
...], scale: float,
output_slices: Tuple[int, ...]) -> None:
"""
Applies lora to each input. Similar to add_lora, This method is
used for layers that are composed of multiple sublayers
(slices) packed together.
"""
y_org = y
x = x.view(-1, x.shape[-1])
y = y.view(-1, y.shape[-1])
offset_left = 0
if bias_all is not None:
y = self.apply_bias_packed_nslice(self.token_lora_indices, y,
output_slices, bias_all)
# TODO fuse these kernels
for slice_idx in range(len(output_slices)):
self.add_lora(y, x, lora_a_stacked[slice_idx],
lora_b_stacked[slice_idx], None, scale, offset_left,
output_slices[slice_idx])
offset_left += output_slices[slice_idx]
y = y.view_as(y_org)
def add_lora_logits(self, def add_lora_logits(self,
y: torch.Tensor, y: torch.Tensor,
x: torch.Tensor, x: torch.Tensor,
wa_t_all: torch.Tensor, lora_a_stacked: torch.Tensor,
wb_t_all: torch.Tensor, lora_b_stacked: torch.Tensor,
scale, scale,
*, *,
buffer: Optional[torch.Tensor] = None) -> None: buffer: Optional[torch.Tensor] = None) -> None:
""" """
LogitsProcessorWithLoRA always using bgmv Applies lora specifically for LogitsProcessorWithLoRA.
"""
Semantics:
buffer = (x @ lora_a_stacked) * scale
y += buffer @ lora_b_stacked
Args:
y (torch.Tensor): Output tensor.
x (torch.Tensor): Input tensor.
lora_a_stacked (torch.Tensor): lora_a's weights.
lora_b_stacked (torch.Tensor):lora_b's weights.
scale (float): Scaling factor.
buffer (Optional[torch.Tensor]):Default to None.
"""
y_org = y y_org = y
y = y.view(-1, y.shape[-1]) y = y.view(-1, y.shape[-1])
x = x.view(-1, x.shape[-1]) x = x.view(-1, x.shape[-1])
r = wb_t_all.size(-1) r = lora_b_stacked.size(-1)
if buffer is None: if buffer is None:
# We set the buffer to be float32 by default ,refer to: # We set the buffer to be float32 by default ,refer to:
# https://github.com/triton-lang/triton/issues/1387 # https://github.com/triton-lang/triton/issues/1387
buffer = torch.zeros((x.size(0), r), buffer = torch.zeros((x.size(0), r),
dtype=torch.float32, dtype=torch.float32,
device=x.device) device=x.device)
# LogitsProcessorWithLoRA always using bgmv.
bgmv_shrink(x, wa_t_all, buffer, self.sampler_indices, scale) bgmv_shrink(x, lora_a_stacked, buffer, self.sampler_indices, scale)
bgmv_expand(buffer, wb_t_all, y, self.sampler_indices, add_inputs=True) bgmv_expand(buffer,
lora_b_stacked,
y,
self.sampler_indices,
add_inputs=True)
y = y.view_as(y_org) y = y.view_as(y_org)