Skip to content

[None][fix] Fix MTP 1-model sampler #10369

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
mikeiovine merged 1 commit into from
Jan 23, 2026
Merged

[None][fix] Fix MTP 1-model sampler #10369

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
9 changes: 4 additions & 5 deletions tensorrt_llm/_torch/pyexecutor/model_engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -48,8 +48,7 @@
get_spec_metadata,
update_spec_config_from_model_config)
from ..speculative.drafting_loops import BaseDraftingLoopWrapper
from ..speculative.eagle3 import (Eagle3OneModelSpecMetadata,
Eagle3ResourceManager, Eagle3SpecMetadata)
from ..speculative.eagle3 import Eagle3ResourceManager, Eagle3SpecMetadata
from ..speculative.mtp import SampleStateTensorsMTP
from ..speculative.utils import SpecDecodingTensor
from ..utils import (get_model_extra_attrs,
Expand Down Expand Up @@ -2684,9 +2683,9 @@ def previous_seq_slots_device():
num_accepted_draft_tokens)]
if isinstance(spec_metadata, Eagle3SpecMetadata):
spec_metadata.request_accepted_path = request_accepted_path
if isinstance(spec_metadata, Eagle3OneModelSpecMetadata):
spec_metadata.populate_sampling_params_for_one_model(
scheduled_requests.all_requests())
# No-op for non 1-model
spec_metadata.populate_sampling_params_for_one_model(
scheduled_requests.all_requests())
spec_metadata.prepare()
inputs['spec_metadata'] = spec_metadata

Expand Down
16 changes: 6 additions & 10 deletions tensorrt_llm/_torch/pyexecutor/py_executor_creator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -281,16 +281,12 @@ def create_py_executor(
)
llm_args.disable_overlap_scheduler = True

if spec_config is not None and spec_config.spec_dec_mode.use_one_engine():
if not spec_config.allow_advanced_sampling:
logger.warning(
f"Falling back to greedy decoding for {spec_config.decoding_type}. If you "
"want to use non-greedy sampling, please set allow_advanced_sampling=True."
)
elif spec_config.spec_dec_mode.is_mtp_one_model():
logger.warning(
"Advanced sampling is not supported for MTP yet - this will be added soon."
)
if spec_config is not None and spec_config.spec_dec_mode.use_one_engine(
) and not spec_config.allow_advanced_sampling:
logger.warning(
f"Falling back to greedy decoding for {spec_config.decoding_type}. If you "
"want to use non-greedy sampling, please set allow_advanced_sampling=True."
)

if mm_encoder_only:
llm_args.mm_encoder_only = True
Expand Down
1 change: 1 addition & 0 deletions tensorrt_llm/_torch/speculative/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,6 +31,7 @@ def get_spec_metadata(spec_config,
mtp_num_modules=spec_config.num_nextn_predict_layers,
max_num_requests=max_num_requests,
mtp_hidden_states_manager=spec_resource_manager,
allow_advanced_sampling=spec_config.allow_advanced_sampling,
)
if spec_config.spec_dec_mode.is_mtp_eagle():
return Eagle3SpecMetadata(
Expand Down