Skip to content

[None][fix] MTP Speculative Sampling #11076

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

Closed
IzzyPutterman wants to merge 1 commit into from
Closed

[None][fix] MTP Speculative Sampling #11076

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
5 changes: 3 additions & 2 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 @@ -52,7 +52,7 @@
from ..speculative.drafting_loops import BaseDraftingLoopWrapper
from ..speculative.eagle3 import (Eagle3OneModelSpecMetadata,
Eagle3ResourceManager, Eagle3SpecMetadata)
from ..speculative.mtp import SampleStateTensorsMTP
from ..speculative.mtp import MTPSpecMetadata, SampleStateTensorsMTP
from ..speculative.utils import SpecDecodingTensor
from ..utils import (get_model_extra_attrs,
set_per_request_piecewise_cuda_graph_flag,
Expand Down Expand Up @@ -2756,7 +2756,8 @@ 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):
if isinstance(spec_metadata,
(Eagle3OneModelSpecMetadata, MTPSpecMetadata)):
spec_metadata.populate_sampling_params_for_one_model(
scheduled_requests.all_requests())
spec_metadata.prepare()
Expand Down
2 changes: 1 addition & 1 deletion tensorrt_llm/_torch/speculative/interface.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -376,7 +376,7 @@ def __init__(self):
super().__init__()
self.guided_decoder: Optional["CapturableGuidedDecoder"] = None
self.force_num_accepted_tokens = get_force_num_accepted_tokens()
self.use_flashinfer = IS_FLASHINFER_AVAILABLE and flashinfer.__version__ >= "0.6.0"
self.use_flashinfer = IS_FLASHINFER_AVAILABLE and flashinfer.__version__ >= "0.6.3"
self.seed = 0
self.offset = 0

Expand Down
3 changes: 2 additions & 1 deletion tensorrt_llm/_torch/speculative/one_model_sampler.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,7 +73,8 @@ def apply_temperature(
return logits.div_(temp.unsqueeze(dim=1))


@torch.compile(options={"max-autotune": True})
# Broken with current ToT (Jan 27)
# @torch.compile(options={"max-autotune": True})
def sampling_batch_spec_dec_one_model(
logits: torch.Tensor,
temperatures: torch.Tensor,
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
Loading