Skip to content

[TRTLLM-6654][feat] Add support for external multimodal embeddings #6263

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
chang-l merged 15 commits into from
Jul 30, 2025
Merged

[TRTLLM-6654][feat] Add support for external multimodal embeddings #6263

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
15 commits
Select commit Hold shift + click to select a range
3272099
Enable mm_emb as input for llama4
chang-l Jul 18, 2025
4467436
Minor update
chang-l Jul 18, 2025
f7e9188
Update to use shared tensor for main-worker comm.
chang-l Jul 21, 2025
d5d81f3
Format
chang-l Jul 21, 2025
b5ba5b6
tokenizer returns tensor
chang-l Jul 22, 2025
78a5278
Add unit tests
chang-l Jul 22, 2025
efdd13b
Format
chang-l Jul 22, 2025
23e7325
Avoid hard-code special image tokens
chang-l Jul 22, 2025
cd7393a
Address comment
chang-l Jul 24, 2025
69c1fc1
Update to cover corner cases
chang-l Jul 25, 2025
a20625c
Update to only accept either image_url or image_emb
chang-l Jul 27, 2025
d9c29f9
Format
chang-l Jul 27, 2025
0a83b39
Address comment
chang-l Jul 28, 2025
6f4a161
Merge branch 'main' into add_mmemb_input
chang-l Jul 29, 2025
c6cdf32
Merge remote-tracking branch 'upstream/main' into add_mmemb_input
chang-l Jul 30, 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
Prev Previous commit
Next Next commit
Address comment
  • Loading branch information
@chang-l
chang-l committed Jul 27, 2025
commit cd7393a1b4fdcfe96f37b08a0c69a05cb79a27df
2 changes: 1 addition & 1 deletion tensorrt_llm/_torch/models/modeling_llama.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -819,7 +819,7 @@ def __init__(self,
}).cuda()
load_sharded_checkpoint(self.encoder, model_path, strict=False)

def attch_multimodal_embeddings(
def attach_multimodal_embeddings(
self, inputs: TextPrompt, multimodal_embedding: Dict[str,
List[Dict[str,
Any]]],
Expand Down
2 changes: 1 addition & 1 deletion tensorrt_llm/_torch/models/modeling_llava_next.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -195,7 +195,7 @@ def _postprocess(self, input_ids, mm_features):
mm_features = mm_features.view(-1, mm_features.shape[-1])
return fused_input_ids, mm_features

def attch_multimodal_embeddings(
def attach_multimodal_embeddings(
self, inputs: TextPrompt,
multimodal_embedding: Dict[str, List[torch.Tensor]],
sampling_params: SamplingParams
Expand Down
2 changes: 1 addition & 1 deletion tensorrt_llm/llmapi/llm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -379,7 +379,7 @@ def generate_async(
inputs, sampling_params)
elif 'multi_modal_embeddings' in inputs:
mm_embedding_info = inputs['multi_modal_embeddings']
prompt_token_ids, extra_processed_inputs = self.input_processor.attch_multimodal_embeddings(
prompt_token_ids, extra_processed_inputs = self.input_processor.attach_multimodal_embeddings(
inputs, mm_embedding_info, sampling_params)
else:
with nvtx_range_debug("input_processor"):
Expand Down