Skip to content

[None][fix] fix for rejection sampling warmup crash (upstream Pr #15233) #12

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
chungen28 merged 1 commit into from
Jun 25, 2026
Merged

[None][fix] fix for rejection sampling warmup crash (upstream Pr #15233) #12

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
23 changes: 15 additions & 8 deletions tensorrt_llm/_torch/modules/embedding.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -160,24 +160,31 @@ def pre_comm_embedding_ops(
gather_output: bool,
padding_size: int,
):
# Generate the mask for the input if needed.
if tp_mode == TensorParallelMode.COLUMN:
input_, input_mask = get_masked_input_and_mask(
input_,
vocab_start_index,
vocab_end_index,
)
else:
# flashinfer's rejection kernel (chain_speculative_sampling) pads non-accepted
# tokens with -1. When the full vocab is local (non-TP or ROW TP), mask
# out-of-range ids (e.g. -1) over [0, weight.shape[0]) to avoid an OOB
# embedding lookup.
input_, input_mask = get_masked_input_and_mask(
input_,
0,
weight.shape[0],
)

# Get the embeddings.
output = F.embedding(input_, weight)

# Mask or pad the output if needed.
if tp_mode == TensorParallelMode.COLUMN:
output.masked_fill_(input_mask, 0)
elif tp_mode == TensorParallelMode.ROW:
if gather_output:
if tp_rank == tp_size - 1 and padding_size > 0:
output = F.pad(output, (0, padding_size))
output.masked_fill_(input_mask, 0)

if tp_mode == TensorParallelMode.ROW and gather_output:
if tp_rank == tp_size - 1 and padding_size > 0:
output = F.pad(output, (0, padding_size))

return output

Expand Down
Loading