Skip to content

[pull] main from NVIDIA:main#354

Merged
pull[bot] merged 5 commits intoyingguo-trt:mainfrom
NVIDIA:main
Apr 4, 2026
Merged

[pull] main from NVIDIA:main#354
pull[bot] merged 5 commits intoyingguo-trt:mainfrom
NVIDIA:main

Conversation

@pull
Copy link
Copy Markdown

@pull pull Bot commented Apr 4, 2026

See Commits and Changes for more details.


Created by pull[bot] (v2.0.0-alpha.4)

Can you help keep this open source service alive? 💖 Please sponsor : )

chang-l and others added 5 commits April 3, 2026 16:13
…2464)

Signed-off-by: Chang Liu <9713593+chang-l@users.noreply.github.com>
Signed-off-by: TensorRT LLM <90828364+tensorrt-cicd@users.noreply.github.com>
…ron (#12150)

Signed-off-by: Bo Deng <deemod@nvidia.com>
* Why?

Previously, the multimodal placeholder insertion was dictated by
hardcoded exception lists, which add cognitive burden when onboarding
new models, and did not account for the fact that different versions
of the same model architecture could have different types of chat
templates that require different handling.

In addition, all placeholders were either added before or after the
text, instead of possibly interleaved.

* What?

This commit addresses the above gaps by mimicking what is done in vLLM.

To that end, it:

1. introduces content format detection based on Jinja AST inspection of
   the chat template.
2. preserves the interleaved positions of text and media items during
   message parsing.
3. dispatches to the appropriate logic based on the (possibly
   auto-detected) content format before applying the chat template:
   either the template handles multimodal content natively (OpenAI-style
   dicts), or expects plain strings with placeholders pre-inserted.
4. inserts the multimodal placeholders.

Models can also explicitly declare their `content_format` during
registration if they are only meant to support one.

Signed-off-by: William Zhang <133824995+2ez4bz@users.noreply.github.com>
Signed-off-by: TensorRT LLM <90828364+tensorrt-cicd@users.noreply.github.com>
@pull pull Bot locked and limited conversation to collaborators Apr 4, 2026
@pull pull Bot added the ⤵️ pull label Apr 4, 2026
@pull pull Bot merged commit b6c5a71 into yingguo-trt:main Apr 4, 2026
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants