minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

config: borrowed-record reads follow the config bags (#35908)

Browse Source 
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
Cheng Wan
2026-08-23 01:19:20 -07:00
committed by GitHub
co-authored by Claude Opus 5
parent 64aa859da2
commit 362c2ee849
65 changed files with 617 additions and 377 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/benchmark/one_batch.py
+6 -6
View File
@@ -79,10 +79,10 @@ from sglang.srt.layers.quantization.fp8_utils import initialize_fp8_gemm_config
from sglang.srt.managers.schedule_batch import Req, ScheduleBatch
from sglang.srt.managers.scheduler_components.dp_attn import prepare_mlp_sync_batch_raw
from sglang.srt.mem_cache.base_prefix_cache import EvictParams
from sglang.srt.model_executor.cuda_graph_config import Phase
from sglang.srt.model_executor.cuda_graph_config import Phase, cuda_graph_fully_disabled
from sglang.srt.model_executor.forward_batch_info import ForwardBatch
from sglang.srt.model_executor.model_runner import ModelRunner
from sglang.srt.runtime_context import get_parallel
from sglang.srt.runtime_context import get_parallel, get_schedule
from sglang.srt.sampling.sampling_params import SamplingParams
from sglang.srt.server_args import PortArgs, ServerArgs
from sglang.srt.speculative.spec_info import SpeculativeAlgorithm
@@ -487,7 +487,7 @@ class TreeCacheNamespace(SimpleNamespace):
def extend(reqs, model_runner):
# Create dummy tree_cache for benchmarks (no prefix caching, just allocation)
dummy_tree_cache = TreeCacheNamespace(
page_size=model_runner.server_args.page_size,
page_size=get_schedule().page_size,
device=model_runner.device,
token_to_kv_pool_allocator=model_runner.token_to_kv_pool_allocator,
)
@@ -542,14 +542,14 @@ def _maybe_prepare_mlp_sync_batch(batch: ScheduleBatch, model_runner):
prepare_mlp_sync_batch_raw(
batch,
model_runner=model_runner,
dp_size=model_runner.server_args.dp_size,
dp_size=get_parallel().dp_size,
attn_tp_size=get_parallel().attn_tp_size,
attn_cp_size=model_runner.ps.attn_cp_size,
tp_group=model_runner.tp_group,
get_idle_batch=None,
disable_cuda_graph=model_runner.server_args.disable_cuda_graph,
disable_cuda_graph=cuda_graph_fully_disabled(),
require_mlp_tp_gather=require_mlp_tp_gather(model_runner.server_args),
disable_overlap_schedule=model_runner.server_args.disable_overlap_schedule,
disable_overlap_schedule=get_schedule().disable_overlap_schedule,
offload_tags=set(),
)