karylab_agents/vllm
1
0
Fork 0
forked from Karylab-cklius/vllm
Code Pull Requests 2 Actions 1 Packages Activity

fix: prefix DeepSeek V4 MTP projections (#44821)

Browse Source 
Signed-off-by: Yufeng He <40085740+he-yufeng@users.noreply.github.com>
This commit is contained in:
Yufeng He
2026-06-10 08:47:04 -07:00
committed by GitHub
parent de900fa7e5
commit 4673ca1d78
2 changed files with 4 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
vllm/models/deepseek_v4/amd/mtp.py
+2
View File
@@ -86,6 +86,7 @@ class DeepSeekV4MultiTokenPredictorLayer(nn.Module):
bias=False,
return_bias=False,
quant_config=quant_config,
prefix=f"{prefix}.e_proj",
)
self.h_proj = ReplicatedLinear(
config.hidden_size,
@@ -93,6 +94,7 @@ class DeepSeekV4MultiTokenPredictorLayer(nn.Module):
bias=False,
return_bias=False,
quant_config=quant_config,
prefix=f"{prefix}.h_proj",
)
self.hc_eps = config.hc_eps
vllm/models/deepseek_v4/nvidia/mtp.py
+2
View File
@@ -92,6 +92,7 @@ class DeepSeekV4MultiTokenPredictorLayer(nn.Module):
bias=False,
return_bias=False,
quant_config=quant_config,
prefix=f"{prefix}.e_proj",
)
self.h_proj = ReplicatedLinear(
config.hidden_size,
@@ -99,6 +100,7 @@ class DeepSeekV4MultiTokenPredictorLayer(nn.Module):
bias=False,
return_bias=False,
quant_config=quant_config,
prefix=f"{prefix}.h_proj",
)
self.hc_eps = config.hc_eps