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

[loader] support presharded fused mlp loading (#19519)

Browse Source
This commit is contained in:
Yilong Zhao
2026-02-27 20:37:24 -08:00
committed by GitHub
parent 71620122c9
commit 79b1d2bac6
1 changed files with 7 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/linear.py
+7 -2
View File
@@ -575,8 +575,13 @@ class MergedColumnParallelLinear(ColumnParallelLinear):
current_shard_offset = 0 current_shard_offset = 0
shard_offsets: List[Tuple[int, int, int]] = [] shard_offsets: List[Tuple[int, int, int]] = []
for i, output_size in enumerate(self.output_sizes): for i, output_size in enumerate(self.output_sizes):
shard_offsets.append((i, current_shard_offset, output_size)) effective_size = (
current_shard_offset += output_size output_size // self.tp_size
if self.use_presharded_weights
else output_size
)
shard_offsets.append((i, current_shard_offset, effective_size))
current_shard_offset += effective_size
packed_dim = getattr(param, "packed_dim", None) packed_dim = getattr(param, "packed_dim", None)
use_bitsandbytes_4bit = getattr(param, "use_bitsandbytes_4bit", False) use_bitsandbytes_4bit = getattr(param, "use_bitsandbytes_4bit", False)