Skip to content

Support absorb dict for awq #1920

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

Merged
merged 7 commits into from
Jul 16, 2024
Merged

Support absorb dict for awq #1920

Changes from 1 commit
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
f6f2cda
support absorb dict for awq
Kaihui-intel Jul 15, 2024
1bf8db0
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Jul 15, 2024
cc61136
update entry
Kaihui-intel Jul 15, 2024
6f23425
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Jul 15, 2024
089a49c
update ut
Kaihui-intel Jul 15, 2024
cfef742
Merge branch 'kaihui/absorb_dict' of https://github.com/intel/neural-…
Kaihui-intel Jul 15, 2024
88ce975
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Jul 15, 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
Prev Previous commit
Next Next commit
update ut 
Signed-off-by: Kaihui-intel <[email protected]>
  • Loading branch information
@Kaihui-intel
Kaihui-intel committed Jul 15, 2024
commit 089a49c8d1d6e77eb0ec434b5e7f041a2bee90d4
55 changes: 12 additions & 43 deletions test/3x/torch/quantization/weight_only/test_awq.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -159,49 +159,19 @@ def test_quant_lm_head(self):
), "The tied lm_head weight is not deep copied, please check!"

def test_awq_absorb_to_layer(self):
absorb_to_layer = {
(
"transformer.h.0.attn.q_proj",
"transformer.h.0.attn.k_proj",
"transformer.h.0.attn.v_proj",
"transformer.h.0.mlp.fc_in",
): "transformer.h.0.ln_1",
("transformer.h.0.attn.out_proj",): "transformer.h.0.attn.out_proj",
("transformer.h.0.mlp.fc_out",): "transformer.h.0.mlp.fc_out",
(
"transformer.h.1.attn.q_proj",
"transformer.h.1.attn.k_proj",
"transformer.h.1.attn.v_proj",
"transformer.h.1.mlp.fc_in",
): "transformer.h.1.ln_1",
("transformer.h.1.attn.out_proj",): "transformer.h.1.attn.out_proj",
("transformer.h.1.mlp.fc_out",): "transformer.h.1.mlp.fc_out",
(
"transformer.h.2.attn.q_proj",
"transformer.h.2.attn.k_proj",
"transformer.h.2.attn.v_proj",
"transformer.h.2.mlp.fc_in",
): "transformer.h.2.ln_1",
("transformer.h.2.attn.out_proj",): "transformer.h.2.attn.out_proj",
("transformer.h.2.mlp.fc_out",): "transformer.h.2.mlp.fc_out",
(
"transformer.h.3.attn.q_proj",
"transformer.h.3.attn.k_proj",
"transformer.h.3.attn.v_proj",
"transformer.h.3.mlp.fc_in",
): "transformer.h.3.ln_1",
("transformer.h.3.attn.out_proj",): "transformer.h.3.attn.out_proj",
("transformer.h.3.mlp.fc_out",): "transformer.h.3.mlp.fc_out",
(
"transformer.h.4.attn.q_proj",
"transformer.h.4.attn.k_proj",
"transformer.h.4.attn.v_proj",
"transformer.h.4.mlp.fc_in",
): "transformer.h.4.ln_1",
("transformer.h.4.attn.out_proj",): "transformer.h.4.attn.out_proj",
("transformer.h.4.mlp.fc_out",): "transformer.h.4.mlp.fc_out",
absorb_layer_dict = {
"ln_1": (
"attn.q_proj",
"attn.k_proj",
"attn.v_proj",
"mlp.fc_in",
),
"attn.out_proj": "attn.out_proj",
"mlp.fc_out": ("mlp.fc_out"),
}
quant_config = AWQConfig(absorb_to_layer=absorb_to_layer)


quant_config = AWQConfig(absorb_layer_dict=absorb_layer_dict)
logger.info(f"Test AWQ with config {quant_config}")
# prepare + convert API
model = prepare(
Expand All @@ -212,7 +182,6 @@ def test_awq_absorb_to_layer(self):
calib_func(model)
model = convert(model)
out1 = model(self.example_inputs)

quant_config = AWQConfig()
logger.info(f"Test AWQ with config {quant_config}")

Expand Down