apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 16:48:03 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

support mtp in ep64 (#4280)
Some checks failed
CE Compile Job / ce_job_pre_check (push) Has been cancelled
Details
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
Details
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
Details
CE Compile Job / Show Code Archive Output (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
Details
CE Compile Job / CE_UPLOAD (push) Has been cancelled
Details

Browse Source
This commit is contained in:
freeliuzc
2025-09-26 15:38:03 +08:00
committed by GitHub
parent dcf633c4d9
commit b176cba474
2 changed files with 11 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
fastdeploy/spec_decode/base.py
View File

@@ -38,13 +38,21 @@ class Proposer(ABC):
Init Speculative proposer
"""
cfg.parallel_config.tp_group = None
cfg.parallel_config.ep_group = None
self.cfg = deepcopy(cfg)
cfg.parallel_config.tp_group = dist.get_group(
cfg.parallel_config.data_parallel_rank + envs.FD_TP_GROUP_GID_OFFSET
)
cfg.parallel_config.ep_group = dist.get_group(
cfg.parallel_config.data_parallel_size + envs.FD_TP_GROUP_GID_OFFSET
)
self.cfg.parallel_config.tp_group = dist.get_group(
cfg.parallel_config.data_parallel_rank + envs.FD_TP_GROUP_GID_OFFSET
)
self.cfg.parallel_config.ep_group = dist.get_group(
cfg.parallel_config.data_parallel_size + envs.FD_TP_GROUP_GID_OFFSET
)
self.parallel_config = self.cfg.parallel_config
self.model_config = self.cfg.model_config
self.speculative_config = self.cfg.speculative_config

3
fastdeploy/spec_decode/mtp.py
View File

@@ -695,6 +695,9 @@ class MTPProposer(Proposer):
if substep != self.num_model_steps - 1:
target_hidden_states = self._get_self_hidden_states(hidden_states)
else:
if hasattr(self.model, "empty_input_forward"):
self.model.empty_input_forward()
def _get_self_hidden_states(self, hidden_states):
target_hidden_states = eagle_get_self_hidden_states(