Skip to content

Implement propagate_error argument #2008

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

Open
kylesayrs wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Implement propagate_error argument #2008

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
864b4d0
implement propagate_error
kylesayrs Nov 10, 2025
ffc4415
optimization
kylesayrs Nov 10, 2025
66f36d2
fix typo
kylesayrs Nov 10, 2025
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
17 changes: 13 additions & 4 deletions src/llmcompressor/args/dataset_arguments.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -223,10 +223,19 @@ class DatasetArguments(CustomDatasetArguments):
quantization_aware_calibration: bool = field(
default=True,
metadata={
"help": "Whether to enable quantization-aware calibration in the pipeline. "
"When True, quantization is applied during forward pass in calibration. "
"When False, quantization is disabled during forward pass in calibration. "
"Default is set to True."
"help": "Only relevant for the sequential pipeline. "
"If True, quantization is applied during forward pass in calibration. "
"If False, quantization is disabled during forward pass in calibration. "
"Default is True."
},
)
propagate_error: bool = field(
default=True,
metadata={
"help": "Only relevant for the sequential pipeline. If True, use quantized "
"layer outputs as the inputs to the next sequential layer. If False, use "
"unquantized layer outputs as the inputs to the next sequential layer. "
"Default is True"
},
)

Expand Down
12 changes: 8 additions & 4 deletions src/llmcompressor/pipelines/sequential/pipeline.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -97,7 +97,11 @@ def __call__(
# do a preliminary pass to trigger modifier hooks
for batch_idx in tqdm(range(len(dataloader)), desc=calib_desc):
inputs = activations.fetch(batch_idx, subgraph.input_names)
subgraph.forward(model, **inputs)
outputs = subgraph.forward(model, **inputs)

if not dataset_args.propagate_error:
activations.update(batch_idx, outputs)
activations.delete(batch_idx, subgraph.consumed_names)

LifecycleCallbacks.sequential_epoch_end(subgraph)

Expand All @@ -106,10 +110,10 @@ def __call__(
with HooksMixin.disable_hooks():
for batch_idx in tqdm(range(len(dataloader)), desc=prop_desc):
inputs = activations.fetch(batch_idx, subgraph.input_names)
output = subgraph.forward(model, **inputs)
outputs = subgraph.forward(model, **inputs)

if subgraph_index < num_subgraphs - 1:
activations.update(batch_idx, output)
if dataset_args.propagate_error:
activations.update(batch_idx, outputs)
activations.delete(batch_idx, subgraph.consumed_names)

# redundant, finish any remaining compression
Expand Down
Loading