Skip to content

[BugFix] mm_post_fix #4005

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Sep 11, 2025
Merged

[BugFix] mm_post_fix #4005

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
bcca44e
mm_post_fix
xiaoxiaohehe001 Sep 9, 2025
ceb78bc
mm_post_fix_1
xiaoxiaohehe001 Sep 9, 2025
ed121ce
Merge branch 'develop' into mm_post_fix
xiaoxiaohehe001 Sep 10, 2025
7377991
Merge branch 'develop' into mm_post_fix
xiaoxiaohehe001 Sep 11, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 0 additions & 12 deletions fastdeploy/model_executor/models/ernie4_5_vl/ernie4_5_vl_moe.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,6 @@

if current_platform.is_cuda():
from fastdeploy.model_executor.ops.gpu import (
extract_text_token_output,
text_image_gather_scatter,
text_image_index_out,
)
Expand Down Expand Up @@ -544,17 +543,6 @@ def forward(
)

hidden_states = hidden_states + residual

max_seq_len, max_seq_len_index = paddle.topk(forward_meta.seq_lens_this_time, k=1)
hidden_states = extract_text_token_output(
max_seq_len,
max_seq_len_index.cast("int32"),
vl_moe_meta.image_token_num.cast("int32"),
forward_meta.seq_lens_this_time,
forward_meta.cu_seqlens_q,
hidden_states.cast("float32"),
).cast(self._dtype)

out = self.norm(hidden_states)

return out
Expand Down
42 changes: 20 additions & 22 deletions fastdeploy/worker/gpu_model_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1298,24 +1298,23 @@ def _dummy_run(
self.share_inputs["image_features"],
self.forward_meta,
)
hidden_states = model_output
else:
model_output = self.model(
ids_remove_padding=self.share_inputs["ids_remove_padding"],
forward_meta=self.forward_meta,
)

hidden_states = rebuild_padding(
model_output,
self.share_inputs["cu_seqlens_q"],
self.share_inputs["seq_lens_this_time"],
self.share_inputs["seq_lens_decoder"],
self.share_inputs["seq_lens_encoder"],
(
self.share_inputs["output_padding_offset"] if self.speculative_decoding else None
), # speculative decoding requires
self.parallel_config.max_model_len,
)
hidden_states = rebuild_padding(
model_output,
self.share_inputs["cu_seqlens_q"],
self.share_inputs["seq_lens_this_time"],
self.share_inputs["seq_lens_decoder"],
self.share_inputs["seq_lens_encoder"],
(
self.share_inputs["output_padding_offset"] if self.speculative_decoding else None
), # speculative decoding requires
self.parallel_config.max_model_len,
)

# 4. Execute spec decode
logits = self.model.compute_logits(hidden_states)
Expand Down Expand Up @@ -1608,21 +1607,20 @@ class at the server level, which is too granular for ModelRunner.
self.share_inputs["image_features"],
self.forward_meta,
)
hidden_states = model_output
else:
model_output = self.model(
ids_remove_padding=self.share_inputs["ids_remove_padding"],
forward_meta=self.forward_meta,
)
hidden_states = rebuild_padding(
model_output,
self.share_inputs["cu_seqlens_q"],
self.share_inputs["seq_lens_this_time"],
self.share_inputs["seq_lens_decoder"],
self.share_inputs["seq_lens_encoder"],
(self.share_inputs["output_padding_offset"] if self.speculative_decoding else None),
self.parallel_config.max_model_len,
)
hidden_states = rebuild_padding(
model_output,
self.share_inputs["cu_seqlens_q"],
self.share_inputs["seq_lens_this_time"],
self.share_inputs["seq_lens_decoder"],
self.share_inputs["seq_lens_encoder"],
(self.share_inputs["output_padding_offset"] if self.speculative_decoding else None),
self.parallel_config.max_model_len,
)

# 4. Compute logits, Sample
logits = self.model.compute_logits(hidden_states)
Expand Down
Loading