iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[Bugfix][MiniCPM-o] Fix cuda/cpu device mismatch in Resampler2_5 pos_embed (#43844)

Browse Source 
Signed-off-by: Parth Ashwin Jain <parthash@amd.com>
Co-authored-by: Parth Ashwin Jain <parthash@amd.com>
Co-authored-by: Cyrus Leung <tlleungac@connect.ust.hk>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Parth Ashwin Jain
2026-06-08 23:28:26 -07:00
committed by GitHub
co-authored by Parth Ashwin Jain Cyrus Leung mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent f843ac1a1c
commit e6fc848d4f
1 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
vllm/model_executor/models/minicpmv.py
+3 -1
View File
@@ -217,7 +217,9 @@ class Resampler2_5(BaseResampler):
for i in range(bs):
tgt_h, tgt_w = tgt_sizes[i].tolist()
pos_embed.append(
self.pos_embed[:tgt_h, :tgt_w, :].reshape((tgt_h * tgt_w, -1)).to(dtype)
self.pos_embed[:tgt_h, :tgt_w, :]
.reshape((tgt_h * tgt_w, -1))
.to(device=device, dtype=dtype)
) # patches * D
key_padding_mask[i, patch_len[i] :] = True
pos_embed = torch.nn.utils.rnn.pad_sequence(