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

[CI][Spec Decode] fix: broken test for EAGLE model #11972

Open
wants to merge 4 commits into
base: main
Choose a base branch
from
+10 −2
Open

[CI][Spec Decode] fix: broken test for EAGLE model #11972

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
5a73145
fix: eagle test failed
llsj14 Jan 12, 2025
efbfc8c
add eagle test to test pipeline
llsj14 Jan 12, 2025
f6dac92
update test-pipeline.yaml
llsj14 Jan 12, 2025
f6625db
update test-pipeline
llsj14 Jan 12, 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
4 changes: 3 additions & 1 deletion .buildkite/test-pipeline.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -230,13 +230,15 @@ steps:
- pytest -v -s test_logits_processor.py
- pytest -v -s model_executor/test_guided_processors.py

- label: Speculative decoding tests # 30min
- label: Speculative decoding tests # 40min
source_file_dependencies:
- vllm/spec_decode
- tests/spec_decode
- vllm/model_executor/models/eagle.py
commands:
- pytest -v -s spec_decode/e2e/test_multistep_correctness.py
- VLLM_ATTENTION_BACKEND=FLASH_ATTN pytest -v -s spec_decode --ignore=spec_decode/e2e/test_multistep_correctness.py
- pytest -v -s spec_decode/e2e/test_eagle_correctness.py

- label: LoRA Test %N # 15min each
mirror_hardwares: [amd]
Expand Down
8 changes: 7 additions & 1 deletion vllm/model_executor/models/eagle.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,11 @@

class DummyInputLayerNorm(nn.Module):

def __init__(self, weight=None, bias=None):
super().__init__()
self.weight = nn.Parameter(weight) if weight is not None else None
self.bias = nn.Parameter(bias) if bias is not None else None

def forward(self, x):
return x

Expand Down Expand Up @@ -69,7 +74,8 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):

# Modify layer normalization and residual connections as suggested
# in the EAGLE framework: https://github.com/SafeAILab/EAGLE
self.model.model.layers[0].input_layernorm = DummyInputLayerNorm()
self.model.model.layers[0].input_layernorm = DummyInputLayerNorm(
weight=self.model.model.layers[0].input_layernorm.weight)
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

With the standard EAGLE model checkpoint, this part is not necessary. However, the EAGLE model ("abhigoyal/vllm-eagle-llama-68m-random") used in spec_decode/e2e/test_eagle_correctness.py contains an input layernorm in its checkpoint. If I simply remove the weight in this DummyInputLayerNorm, it could cause errors when loading the model's weights.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: Can you please add a comment on the need to have these weights for some of the unit tests that we have?

self.model.model.norm = DummyOutputNorm()

self.orig_vocab_size = config.vocab_size
Expand Down
Loading