[EPLB] Mask padding in EPLB load recording (#38128)

Signed-off-by: ilmarkov <markovilya197@gmail.com>
Signed-off-by: Markov Ilya <markovilya19@gmail.com>
Co-authored-by: Markov Ilya <markovilya19@gmail.com>
This commit is contained in:
Ilya Markov
2026-06-28 19:43:58 -07:00
committed by GitHub
co-authored by Markov Ilya
parent 311ad689ad
commit a2abce646f
17 changed files with 289 additions and 14 deletions
+12
View File
@@ -8,6 +8,7 @@ from typing import Any
import torch
import torch.nn as nn
from vllm.config import ModelConfig
from vllm.distributed.eplb.eplb_state import EplbState
from vllm.logger import init_logger
from vllm.model_executor.models.interfaces import (
@@ -90,6 +91,7 @@ class EPLBController:
draft_model,
speculative_config.draft_model_config,
)
speculator.set_eplb_state(self.state)
self._has_registered_models = True
return True
@@ -135,6 +137,16 @@ class EPLBController:
log_stats=self.parallel_config.eplb_config.log_balancedness,
)
def prepare_forward(
self,
model_config: ModelConfig,
num_unpadded_tokens: int,
ubatch_slices: list | None = None,
) -> None:
if self.state is None or not self.parallel_config.enable_eplb:
return
self.state.prepare_forward(model_config, num_unpadded_tokens, ubatch_slices)
def setup_from_mapping(
self,
model: nn.Module,