Skip to content

fix: use vision_size for Gemma4 specializations to support chunked embedding in vLLM v1 (#1135) #1146

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
Rishin Raj (quic-rishinr) merged 1 commit into from
Jul 6, 2026
Merged

fix: use vision_size for Gemma4 specializations to support chunked embedding in vLLM v1 (#1135) #1146

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
14 changes: 10 additions & 4 deletions QEfficient/transformers/models/gemma4/modeling_gemma4.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1179,7 +1179,13 @@ def get_specializations(
prefill_seq_len = prefill_seq_len if prefill_seq_len else 32
ctx_len = ctx_len if ctx_len else constants.INTERN_CTX_LEN
max_patches = self._get_vision_max_patches()
mm_tokens_per_image = self._get_mm_tokens_per_image()
user_vision_size = compiler_options.pop("vision_size", None)
if user_vision_size:
if user_vision_size >= ctx_len:
raise ValueError("vision_size must be less than ctx_len")
vision_size = user_vision_size
else:
vision_size = self._get_mm_tokens_per_image()

vision = [{"batch_size": batch_size, "max_patches": max_patches}]

Expand All @@ -1190,7 +1196,7 @@ def build_lang_prefill_spec(comp_ctx_lengths: Optional[int] = None):
"ctx_len": ctx_len,
"sliding_window": self.model.language_model.config.sliding_window,
"vision_batch_size": batch_size,
"vision_tokens": mm_tokens_per_image,
"vision_size": vision_size,
}
if comp_ctx_lengths is not None:
spec["comp_ctx_lengths"] = comp_ctx_lengths
Expand All @@ -1209,7 +1215,7 @@ def build_lang_decode_spec(comp_ctx_lengths: Optional[int] = None):
"ctx_len": ctx_len,
"sliding_window": self.model.language_model.config.sliding_window,
"vision_batch_size": batch_size,
"vision_tokens": mm_tokens_per_image,
"vision_size": vision_size,
}
if comp_ctx_lengths is not None:
spec["comp_ctx_lengths"] = comp_ctx_lengths
Expand Down Expand Up @@ -1237,7 +1243,7 @@ def get_onnx_dynamic_axes(
}
lang_dynamic_axes = {
"input_ids": {0: "batch_size", 1: "seq_len"},
"vision_embeds": {0: "vision_batch_size", 1: "vision_tokens"},
"vision_embeds": {0: "vision_batch_size", 1: "vision_size"},
"position_ids": {0: "batch_size", 1: "seq_len"},
"mm_token_type_ids": {0: "batch_size", 1: "seq_len"},
}
Expand Down
Loading