Skip to content

fix: Fix trtllm sampler beam width bug #4507

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 2 commits into from
May 21, 2025
Merged

fix: Fix trtllm sampler beam width bug #4507

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
1d1ce7f
Fix TRTLLMSampler.
dcampora May 20, 2025
e1aeeef
Added type hint.
dcampora May 20, 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
13 changes: 7 additions & 6 deletions tensorrt_llm/_torch/pyexecutor/sampler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,5 @@
from abc import ABC, abstractmethod
from collections.abc import Iterable
from dataclasses import dataclass

import torch
Expand Down Expand Up @@ -561,7 +562,7 @@ def setup_sampler_step(self, requests):
self.model_config, self.world_config, self.decoding_config,
requests, self.store["buffer_manager"], self.logits_datatype,
self.store["decoder_input_buffers"],
self.algs.decoder.decoder_state, self.beam_width,
self.algs.decoder.decoder_state, self.beam_width(requests),
self.store["cuda_stream"])

if len(decoder_requests):
Expand All @@ -578,15 +579,15 @@ def setup_sampler_step(self, requests):
decoder_requests)

@staticmethod
def beam_width(scheduled_requests: ScheduledRequests) -> int:
for req in scheduled_requests.all_requests:
def beam_width(scheduled_requests: Iterable[LlmRequest]) -> int:
for req in scheduled_requests:
return req.sampling_config.beam_width
raise ValueError("No beam width found")
return 0

def sample_async(self, scheduled_requests: ScheduledRequests,
model_outputs) -> SampleStateTRTLLM:
batch_size = scheduled_requests.batch_size
beam_width = self.beam_width(scheduled_requests)
beam_width = self.beam_width(scheduled_requests.all_requests)

logits = model_outputs["logits"].reshape((batch_size, beam_width, -1))

Expand Down Expand Up @@ -659,7 +660,7 @@ def update_requests(self, state: SampleStateTRTLLM):

scheduled_requests = state.scheduled_requests
assert scheduled_requests.batch_size > 0
beam_width = self.beam_width(scheduled_requests)
beam_width = self.beam_width(scheduled_requests.all_requests)
sampler_event = state.sampler_event

if sampler_event:
Expand Down