chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

Enhance runtime memory check in CI (#15192)

Browse Source
This commit is contained in:
Liangsheng Yin
2025-12-16 21:40:38 +08:00
committed by GitHub
parent b399e3ac4f
commit ecb401ed42
8 changed files with 23 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/speculative/eagle_info.py
View File

@@ -114,8 +114,6 @@ class EagleVerifyInput(SpecInput, EagleVerifyInputV2Mixin):
len(batch.input_ids),
)
end_offset = batch.seq_lens + self.draft_token_num
for req in batch.reqs:
req.kv_allocated_len += 1
else:
prefix_lens = batch.seq_lens
prefix_lens_cpu = batch.seq_lens_cpu

2
python/sglang/srt/speculative/eagle_worker.py
View File

@@ -382,8 +382,6 @@ class EAGLEWorker(TpModelWorker):
# [ topk 0 ] [ topk 1 ]
# [iter=0, iter=1, iter=2] [iter=0, iter=1, iter=2]
if self.page_size == 1:
for req in batch.reqs:
req.kv_allocated_len += self.speculative_num_steps * self.topk
# TODO: We only need self.speculative_num_steps - 1 * topk cache loc
out_cache_loc, token_to_kv_pool_state_backup = alloc_token_slots(
batch.tree_cache,