Skip to content

[V1] implement tree sampler for draft token acceptance #22752

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
wants to merge 1 commit into
base: main
Choose a base branch
from
+808 −143
Open

[V1] implement tree sampler for draft token acceptance #22752

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
16 changes: 9 additions & 7 deletions tests/v1/e2e/test_spec_decode.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -162,12 +162,6 @@ def test_eagle_correctness(
mm_enabled: bool,
attn_backend: str,
):
if attn_backend == "TREE_ATTN":
# TODO: Fix this flaky test
pytest.skip(
"TREE_ATTN is flaky in the test disable for now until it can be "
"reolved (see https://github.com/vllm-project/vllm/issues/22922)")

# Generate test prompts inside the function instead of using fixture
test_prompts = get_test_prompts(mm_enabled)
'''
Expand Down Expand Up @@ -222,7 +216,15 @@ def test_eagle_correctness(

# Heuristic: expect at least 66% of the prompts to match exactly
# Upon failure, inspect the outputs to check for inaccuracy.
assert matches > int(0.66 * len(ref_outputs))
accuracy_threshold = 0.66

if attn_backend == "TREE_ATTN":
# The tree attention backend uses Triton kernels, which exhibit
# floating-point nondeterminism. Reducing the threshold to 50%
# to prevent flaky tests.
accuracy_threshold = 0.50

assert matches > int(accuracy_threshold * len(ref_outputs))
del spec_llm
torch.cuda.empty_cache()
cleanup_dist_env_and_memory()
Loading