[Model] Add Ernie4.5 VL Model Support by CSWYF3634076 · Pull Request #22514 · vllm-project/vllm

CSWYF3634076 · 2025-08-08T09:39:06Z

Purpose

Support Baidu Ernie4.5 VL model for vllm

Note: torch.compile is not supported. Due to the model limitations of multimodal experts and text experts, using torch.compile may fail to start

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

github-actions · 2025-08-08T09:39:15Z

👋 Hi! Thank you for contributing to the vLLM project.

💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.

Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.

To run CI, PR reviewers can either: Add ready label to the PR or enable auto-merge.

🚀

gemini-code-assist

Code Review

This pull request adds support for the Baidu Ernie4.5 VL model. The changes include new model implementation files, rotary embedding logic, and a custom processor. While the overall structure seems correct, there are several critical issues that need to be addressed. These include hardcoded values that make the code brittle, unsafe tensor operations that could lead to runtime errors, incorrect function calls in the processor, and some incomplete logic indicated by TODOs. Addressing these points will significantly improve the robustness and maintainability of the new model support.

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

ywang96

Thanks for the contritbution! Is this PR ready for review? If not, could you please convert this to a draft PR?

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

CSWYF3634076 · 2025-08-12T08:46:00Z

Thanks for the contritbution! Is this PR ready for review? If not, could you please convert this to a draft PR?

@ywang96 Hi, the PR is ready for review. Could you help review it?

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

DarkLight1337 · 2025-08-15T03:00:36Z

Sorry for the delay, I'll take a look later today

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

Signed-off-by: wangyafeng <wangyafeng@baidu.com> Signed-off-by: tc-mb <caitianchi@modelbest.cn>

hmellor · 2025-08-27T07:56:55Z

Why has decord been re-added as a dependency? In the past we went to the effort of removing it because it's been unmaintained for 3 years.

As far as I can tell nothing that was added in this PR needs it.

DarkLight1337 · 2025-08-27T07:59:47Z

It is only in the test dependencies, and is needed to load the model from HF Hub

hmellor · 2025-08-27T08:08:45Z

Ah I see. That'll be why nothing in vLLM appears to be using it. Thanks for explaining.

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

Signed-off-by: wangyafeng <wangyafeng@baidu.com> Signed-off-by: Xiao Yu <xiao.yu@amd.com>

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

justicel · 2025-11-25T15:53:24Z

It appears the decord requirement breaks this in production when trying to load the model. The decord lib isn't included in the vllm docker images and so this happens:

INFO 11-24 16:04:03 [scheduler.py:207] Chunked prefill is enabled with max_num_batched_tokens=2048.
(APIServer pid=1) INFO 11-24 16:04:03 [api_server.py:2056] vLLM API server version 0.11.2.dev201+g55c21c883
(APIServer pid=1) INFO 11-24 16:04:03 [utils.py:253] non-default args: {'model': '/models/ernie54-vl-28b-395bb358-6394-4721-be96-2fd42fd77299', 'trust_remote_code': True, 'served_model_name': ['ernie54-vl-28b'], 'gpu_memory_utilization': 0.95}
(APIServer pid=1) The argument `trust_remote_code` is to be used with Auto classes. It has no effect here and is ignored.
(APIServer pid=1) INFO 11-24 16:04:03 [config.py:508] Replacing legacy 'type' key with 'rope_type'
(APIServer pid=1) INFO 11-24 16:04:10 [model.py:630] Resolved architecture: Ernie4_5_VLMoeForConditionalGeneration
(APIServer pid=1) INFO 11-24 16:04:10 [model.py:1745] Using max model len 131072
(APIServer pid=1) INFO 11-24 16:04:10 [scheduler.py:207] Chunked prefill is enabled with max_num_batched_tokens=8192.
(APIServer pid=1) Encountered exception while importing decord: No module named 'decord'
(APIServer pid=1) Traceback (most recent call last):
(APIServer pid=1)   File "<frozen runpy>", line 198, in _run_module_as_main
(APIServer pid=1)   File "<frozen runpy>", line 88, in _run_code
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/vllm/entrypoints/openai/api_server.py", line 2178, in <module>
(APIServer pid=1)     uvloop.run(run_server(args))
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/uvloop/__init__.py", line 96, in run
(APIServer pid=1)     return __asyncio.run(
(APIServer pid=1)            ^^^^^^^^^^^^^^
(APIServer pid=1)   File "/usr/lib/python3.12/asyncio/runners.py", line 195, in run
(APIServer pid=1)     return runner.run(main)
(APIServer pid=1)            ^^^^^^^^^^^^^^^^
(APIServer pid=1)   File "/usr/lib/python3.12/asyncio/runners.py", line 118, in run
(APIServer pid=1)     return self._loop.run_until_complete(task)
(APIServer pid=1)            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=1)   File "uvloop/loop.pyx", line 1518, in uvloop.loop.Loop.run_until_complete
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/uvloop/__init__.py", line 48, in wrapper
(APIServer pid=1)     return await main
(APIServer pid=1)            ^^^^^^^^^^
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/vllm/entrypoints/openai/api_server.py", line 2106, in run_server
(APIServer pid=1)     await run_server_worker(listen_address, sock, args, **uvicorn_kwargs)
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/vllm/entrypoints/openai/api_server.py", line 2125, in run_server_worker
(APIServer pid=1)     async with build_async_engine_client(
(APIServer pid=1)                ^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=1)   File "/usr/lib/python3.12/contextlib.py", line 210, in __aenter__
(APIServer pid=1)     return await anext(self.gen)
(APIServer pid=1)            ^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/vllm/entrypoints/openai/api_server.py", line 196, in build_async_engine_client
(APIServer pid=1)     async with build_async_engine_client_from_engine_args(
(APIServer pid=1)                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=1)   File "/usr/lib/python3.12/contextlib.py", line 210, in __aenter__
(APIServer pid=1)     return await anext(self.gen)
(APIServer pid=1)            ^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/vllm/entrypoints/openai/api_server.py", line 237, in build_async_engine_client_from_engine_args
(APIServer pid=1)     async_llm = AsyncLLM.from_vllm_config(
(APIServer pid=1)                 ^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/vllm/utils/func_utils.py", line 116, in inner
(APIServer pid=1)     return fn(*args, **kwargs)
(APIServer pid=1)            ^^^^^^^^^^^^^^^^^^^
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/vllm/v1/engine/async_llm.py", line 219, in from_vllm_config
(APIServer pid=1)     return cls(
(APIServer pid=1)            ^^^^
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/vllm/v1/engine/async_llm.py", line 114, in __init__
(APIServer pid=1)     tokenizer = init_tokenizer_from_configs(self.model_config)
(APIServer pid=1)                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/vllm/transformers_utils/tokenizer.py", line 295, in init_tokenizer_from_configs
(APIServer pid=1)     return get_tokenizer(
(APIServer pid=1)            ^^^^^^^^^^^^^^
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/vllm/transformers_utils/tokenizer.py", line 221, in get_tokenizer
(APIServer pid=1)     tokenizer = AutoTokenizer.from_pretrained(
(APIServer pid=1)                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/transformers/models/auto/tokenization_auto.py", line 1122, in from_pretrained
(APIServer pid=1)     tokenizer_class = get_class_from_dynamic_module(class_ref, pretrained_model_name_or_path, **kwargs)
(APIServer pid=1)                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/transformers/dynamic_module_utils.py", line 604, in get_class_from_dynamic_module
(APIServer pid=1)     final_module = get_cached_module_file(
(APIServer pid=1)                    ^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/transformers/dynamic_module_utils.py", line 427, in get_cached_module_file
(APIServer pid=1)     modules_needed = check_imports(resolved_module_file)
(APIServer pid=1)                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
(APIServer pid=1)   File "/usr/local/lib/python3.12/dist-packages/transformers/dynamic_module_utils.py", line 260, in check_imports
(APIServer pid=1)     raise ImportError(
(APIServer pid=1) ImportError: This modeling file requires the following packages that were not found in your environment: decord. Run `pip install decord`

DarkLight1337 · 2025-11-25T15:56:27Z

Yes, that's intentional. You have to install these additional dependencies manually because of licensing constraints. See #28722

justicel · 2025-11-25T16:45:59Z

@DarkLight1337 Thanks. By any chance do you know who to talk to (or should I just submit a pull request), about adding details about this requirement to the Ernie 4.5 page(s)? https://docs.vllm.ai/projects/recipes/en/latest/Ernie/Ernie4.5-VL.html

ywang96 · 2025-11-25T19:31:48Z

@DarkLight1337 Thanks. By any chance do you know who to talk to (or should I just submit a pull request), about adding details about this requirement to the Ernie 4.5 page(s)? https://docs.vllm.ai/projects/recipes/en/latest/Ernie/Ernie4.5-VL.html

Feel free to submit a PR! vLLM-recipe is a community project

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

CSWYF3634076 added 2 commits August 8, 2025 16:28

[Model] Add Ernie4.5 VL

dee4372

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

[Model] Add Ernie4.5 VL v2 annotation organization

af1d864

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

CSWYF3634076 requested review from DarkLight1337, hmellor and ywang96 as code owners August 8, 2025 09:39

mergify Bot added documentation Improvements or additions to documentation new-model Requests to new models labels Aug 8, 2025

gemini-code-assist Bot reviewed Aug 8, 2025

View reviewed changes

CSWYF3634076 added 2 commits August 8, 2025 17:53

[Model] Add Ernie4.5 VL v2 annotation organization

99773ff

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

[Model] Add Ernie4.5 VL v3 fix variable name

8442d5d

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

ywang96 reviewed Aug 9, 2025

View reviewed changes

CSWYF3634076 marked this pull request as draft August 9, 2025 07:21

CSWYF3634076 mentioned this pull request Aug 11, 2025

[Feature]: Ernie-4.5 vision support #20732

Closed

1 task

CSWYF3634076 added 5 commits August 11, 2025 21:55

[Model] Add Ernie4.5 VL v4 fix code-assist issue

42004bc

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

fix format by pre-commit

9a509cd

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

[Model] Add Ernie4.5 VL v5 fix format by pre-commit

8d3d62b

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

[Model] Add Ernie4.5 VL v5 fix format by pre-commit

c227368

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

[Model] Add Ernie4.5 VL v5 fix format by pre-commit

080f818

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

CSWYF3634076 marked this pull request as ready for review August 12, 2025 08:41

CSWYF3634076 requested a review from ywang96 August 12, 2025 08:46

CSWYF3634076 changed the title ~~[WIP][Model] Add Ernie4.5 VL Model Support~~ [Model] Add Ernie4.5 VL Model Support Aug 12, 2025

[Model] Add Ernie4.5 VL v5 add trust_remote_code tag

01f2231

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

Isotr0py reviewed Aug 15, 2025

View reviewed changes

DarkLight1337 reviewed Aug 15, 2025

View reviewed changes

Comment thread vllm/model_executor/layers/rotary_embedding/ernie45_vl_rope.py Outdated

Comment thread vllm/model_executor/models/ernie45_vl.py Outdated

Isotr0py reviewed Aug 15, 2025

View reviewed changes

Comment thread vllm/model_executor/models/ernie45_vl.py Outdated

Isotr0py reviewed Aug 15, 2025

View reviewed changes

Comment thread vllm/model_executor/models/registry.py Outdated

Comment thread vllm/model_executor/models/ernie45_vl.py Outdated

Comment thread vllm/model_executor/models/ernie45_vl.py Outdated

[Model] Add Ernie4.5 VL v6 rename and fix comments

d4ee345

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

tc-mb pushed a commit to tc-mb/vllm that referenced this pull request Aug 27, 2025

[Model] Add Ernie4.5 VL Model Support (vllm-project#22514)

d49f0a7

Signed-off-by: wangyafeng <wangyafeng@baidu.com> Signed-off-by: tc-mb <caitianchi@modelbest.cn>

m1namuci mentioned this pull request Aug 27, 2025

Feature Request: Support for ERNIE-4.5-VL ggml-org/llama.cpp#15512

Closed

4 tasks

epwalsh pushed a commit to epwalsh/vllm that referenced this pull request Aug 28, 2025

[Model] Add Ernie4.5 VL Model Support (vllm-project#22514)

eaed096

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

xiao-llm pushed a commit to xiao-llm/vllm that referenced this pull request Aug 28, 2025

[Model] Add Ernie4.5 VL Model Support (vllm-project#22514)

0e55d10

Signed-off-by: wangyafeng <wangyafeng@baidu.com> Signed-off-by: Xiao Yu <xiao.yu@amd.com>

zhewenl pushed a commit to zhewenl/vllm that referenced this pull request Aug 28, 2025

[Model] Add Ernie4.5 VL Model Support (vllm-project#22514)

66b2f22

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

zhewenl pushed a commit to zhewenl/vllm that referenced this pull request Sep 3, 2025

[Model] Add Ernie4.5 VL Model Support (vllm-project#22514)

cdf9ba3

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

jinyouzhi pushed a commit to jinyouzhi/vllm that referenced this pull request Sep 11, 2025

[Model] Add Ernie4.5 VL Model Support (vllm-project#22514)

d8a5cb8

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

ABC12345anouys pushed a commit to ABC12345anouys/vllm that referenced this pull request Sep 25, 2025

[Model] Add Ernie4.5 VL Model Support (vllm-project#22514)

072c58a

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

jinyouzhi pushed a commit to jinyouzhi/vllm that referenced this pull request Oct 24, 2025

[Model] Add Ernie4.5 VL Model Support (vllm-project#22514)

024cc89

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

mystous pushed a commit to mystous/vllm_hybrid that referenced this pull request May 10, 2026

[Model] Add Ernie4.5 VL Model Support (vllm-project#22514)

1af5181

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

my-other-github-account pushed a commit to my-other-github-account/vllm that referenced this pull request May 15, 2026

[Model] Add Ernie4.5 VL Model Support (vllm-project#22514)

13ab1ab

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

my-other-github-account pushed a commit to my-other-github-account/vllm that referenced this pull request May 15, 2026

[Model] Add Ernie4.5 VL Model Support (vllm-project#22514)

4a0b0be

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

0826joyce pushed a commit to 0826joyce/vllm-serving-optimization that referenced this pull request May 19, 2026

[Model] Add Ernie4.5 VL Model Support (vllm-project#22514)

d29c659

Signed-off-by: wangyafeng <wangyafeng@baidu.com>

Uh oh!

Uh oh!

Conversation

CSWYF3634076 commented Aug 8, 2025 • edited by github-actions Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Purpose

Uh oh!

github-actions Bot commented Aug 8, 2025

Uh oh!

gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ywang96 left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

CSWYF3634076 commented Aug 12, 2025

Uh oh!

DarkLight1337 commented Aug 15, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

hmellor commented Aug 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

DarkLight1337 commented Aug 27, 2025

Uh oh!

hmellor commented Aug 27, 2025

Uh oh!

justicel commented Nov 25, 2025

Uh oh!

DarkLight1337 commented Nov 25, 2025

Uh oh!

justicel commented Nov 25, 2025

Uh oh!

ywang96 commented Nov 25, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

CSWYF3634076 commented Aug 8, 2025 •

edited by github-actions Bot

Loading

ywang96 left a comment •

edited

Loading

hmellor commented Aug 27, 2025 •

edited

Loading