xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-17 23:46:21 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Kimi-Linear] Correct prefixes and add compatibility to AWQ quants (#27834)

Browse Source 
Signed-off-by: toncao <cpatonn@gmail.com>
Co-authored-by: toncao <cpatonn@gmail.com>
This commit is contained in:
toncao 2025-10-31 16:36:37 +07:00 committed by GitHub
parent 36960501d3
commit e5ef4dfc11
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/model_executor/models/kimi_linear.py
View File

@ -155,6 +155,7 @@ class KimiMoE(nn.Module):
hidden_act=config.hidden_act, hidden_act=config.hidden_act,
quant_config=quant_config, quant_config=quant_config,
reduce_results=False, reduce_results=False,
prefix=f"{prefix}.shared_experts",
) )
def forward(self, hidden_states: torch.Tensor) -> torch.Tensor: def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
@ -340,7 +341,7 @@ class KimiDecoderLayer(nn.Module):
self.block_sparse_moe = KimiMoE( self.block_sparse_moe = KimiMoE(
config=config, config=config,
quant_config=quant_config, quant_config=quant_config,
prefix=f"{prefix}.mlp", prefix=f"{prefix}.block_sparse_moe",
) )
self.mlp = self.block_sparse_moe self.mlp = self.block_sparse_moe
else: else: