Integrate lmdeploy pipeline api

[lvhan028]

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

Please describe the motivation of this PR and the goal you want to achieve through this PR.

Modification

Please briefly describe what modification is made in this PR.

BC-breaking (Optional)

Does the modification introduce changes that break the backward compatibility of the downstream repositories?
If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR.

Use cases (Optional)

If this PR introduces a new feature, it is better to list some use cases here and update the documentation.

Checklist

Before PR:

• Pre-commit or other linting tools are used to fix the potential lint issues.
• Bug fixes are fully covered by unit tests, the case that causes the bug should be added in the unit tests.
• The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
• The documentation has been modified accordingly, like docstring or example tutorials.

After PR:

• If the modification has potential influence on downstream or other related projects, this PR should be tested with those projects.
• CLA has been signed and all committers have signed the CLA in this PR.

[tonysy]

This bs exceeds the max QPS?

[lvhan028]

The default max_batch_size of lmdeploy engines is 256 on A100
batch_size as far as I know is the length of messages passed to the generate in TurboMindModelwithChatTemplate.
The greater batch_size is, the higher throughput lmdeploy inference engine can achieve.

[lvhan028]

It was found that the accuracy degraded compared to the engine integration.
I am debugging it and will let you know once it is ready

[zhulinJulia24]

lgtm

[zhulinJulia24]

use newest lmdeploy code with opencompass main and this pr to do evaluation (batch_size = 1):
main code: https://github.com/InternLM/lmdeploy/actions/runs/11251926631
pr code: https://github.com/InternLM/lmdeploy/actions/runs/11252037673
there scores are different.

use vimdiff to compass predictions. Is that expected? @lvhan028

[lvhan028]

@zhulinJulia24 in main branch, BOS token is missed when tokenizing the message.

opencompass/opencompass/models/turbomind_with_tf_above_v4_33.py

Line 178 in d2ab51a

input_ids = self.tokenizer.encode(prompt, add_bos=False)

In this PR, BOS token is added.
That's the reason

[tonysy]

LGTM