Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Misc][Refactor] Generalize linear_method to be quant_method #4373

Merged
merged 3 commits into from
Apr 26, 2024
Merged

[Misc][Refactor] Generalize linear_method to be quant_method #4373

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
5908873
[Refactor] Generalize linear_method to be quant_method
comaniac Apr 23, 2024
b5e5300
revert
comaniac Apr 25, 2024
461dd82
Merge branch 'main' into quant_method
comaniac Apr 25, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion tests/quantization/test_fp8.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,5 +20,5 @@ def test_load_fp16_model(vllm_runner) -> None:

model = llm.model.llm_engine.model_executor.driver_worker.model_runner.model
fc1 = model.model.decoder.layers[0].fc1
assert isinstance(fc1.linear_method, Fp8LinearMethod)
assert isinstance(fc1.quant_method, Fp8LinearMethod)
assert fc1.weight.dtype == torch.float8_e4m3fn
4 changes: 2 additions & 2 deletions tests/tensorizer_loader/test_tensorizer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -50,10 +50,10 @@ def test_load_with_tensorizer(mock_agent, tensorizer_config):
mock_agent_instance.deserialize.return_value = MagicMock()

result = load_with_tensorizer(tensorizer_config,
linear_method=mock_linear_method)
quant_method=mock_linear_method)

mock_agent.assert_called_once_with(tensorizer_config,
linear_method=mock_linear_method)
quant_method=mock_linear_method)
mock_agent_instance.deserialize.assert_called_once()
assert result == mock_agent_instance.deserialize.return_value

Expand Down
30 changes: 13 additions & 17 deletions vllm/lora/layers.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -389,10 +389,9 @@ def set_mapping(
self.indices = base_indices
self.indices_len = indices_len

def apply_weights(self, x: torch.Tensor,
bias: Optional[torch.Tensor]) -> torch.Tensor:
output = self.base_layer.linear_method.apply_weights(
self.base_layer, x, bias)
def apply(self, x: torch.Tensor,
bias: Optional[torch.Tensor]) -> torch.Tensor:
output = self.base_layer.quant_method.apply(self.base_layer, x, bias)
_apply_lora(
x,
self.lora_a_stacked,
Expand All @@ -416,7 +415,7 @@ def forward(self, input_):
if not self.base_layer.skip_bias_add else None)

# Matrix multiply.
output_parallel = self.apply_weights(input_, bias)
output_parallel = self.apply(input_, bias)
if self.base_layer.gather_output:
# All-gather across the partitions.
output = tensor_model_parallel_all_gather(output_parallel)
Expand Down Expand Up @@ -523,10 +522,9 @@ def set_lora(
index, 0, :lora_b[1].shape[1], :lora_b[1].shape[0]].copy_(
lora_b[1].T, non_blocking=True)

def apply_weights(self, x: torch.Tensor,
bias: Optional[torch.Tensor]) -> torch.Tensor:
output = self.base_layer.linear_method.apply_weights(
self.base_layer, x, bias)
def apply(self, x: torch.Tensor,
bias: Optional[torch.Tensor]) -> torch.Tensor:
output = self.base_layer.quant_method.apply(self.base_layer, x, bias)
_apply_lora_packed_nslice(
x,
self.lora_a_stacked,
Expand Down Expand Up @@ -765,10 +763,9 @@ def set_lora(
index, 0, :lora_a[2].shape[1], :lora_a[2].shape[0]].copy_(
lora_a[2].T, non_blocking=True)

def apply_weights(self, x: torch.Tensor,
bias: Optional[torch.Tensor]) -> torch.Tensor:
output = self.base_layer.linear_method.apply_weights(
self.base_layer, x, bias)
def apply(self, x: torch.Tensor,
bias: Optional[torch.Tensor]) -> torch.Tensor:
output = self.base_layer.quant_method.apply(self.base_layer, x, bias)
_apply_lora_packed_nslice(
x,
self.lora_a_stacked,
Expand Down Expand Up @@ -862,9 +859,8 @@ def set_mapping(
self.indices = base_indices
self.indices_len = indices_len

def apply_weights(self, x: torch.Tensor) -> torch.Tensor:
output = self.base_layer.linear_method.apply_weights(
self.base_layer, x)
def apply(self, x: torch.Tensor) -> torch.Tensor:
output = self.base_layer.quant_method.apply(self.base_layer, x)
_apply_lora(
x,
self.lora_a_stacked,
Expand Down Expand Up @@ -897,7 +893,7 @@ def forward(self, input_):
input_parallel = splitted_input[tp_rank].contiguous()

# Matrix multiply.
output_parallel = self.apply_weights(input_parallel)
output_parallel = self.apply(input_parallel)
if self.base_layer.reduce_results and self.base_layer.tp_size > 1:
output_ = tensor_model_parallel_all_reduce(output_parallel)
else:
Expand Down
Loading
Loading