[Bugfix]Fix MiniCPM's LoRA bug #9286

jeejeelee · 2024-10-11T14:43:21Z

FILL IN THE PR DESCRIPTION HERE

BEFORE SUBMITTING, PLEASE READ THE CHECKLIST BELOW AND FILL IN THE DESCRIPTION ABOVE

PR Checklist (Click to Expand)

Thank you for your contribution to vLLM! Before submitting the pull request, please ensure the PR meets the following criteria. This helps vLLM maintain the code quality and improve the efficiency of the review process.

PR Title and Classification

Only specific types of PRs will be reviewed. The PR title is prefixed appropriately to indicate the type of change. Please use one of the following:

[Bugfix] for bug fixes.
[CI/Build] for build or continuous integration improvements.
[Doc] for documentation fixes and improvements.
[Model] for adding a new model or improving an existing model. Model name should appear in the title.
[Frontend] For changes on the vLLM frontend (e.g., OpenAI API server, LLM class, etc.)
[Kernel] for changes affecting CUDA kernels or other compute kernels.
[Core] for changes in the core vLLM logic (e.g., LLMEngine, AsyncLLMEngine, Scheduler, etc.)
[Hardware][Vendor] for hardware-specific changes. Vendor name should appear in the prefix (e.g., [Hardware][AMD]).
[Misc] for PRs that do not fit the above categories. Please use this sparingly.

Note: If the PR spans more than one category, please include all relevant prefixes.

Code Quality

The PR need to meet the following code quality standards:

We adhere to Google Python style guide and Google C++ style guide.
Pass all linter checks. Please use format.sh to format your code.
The code need to be well-documented to ensure future contributors can easily understand the code.
Include sufficient tests to ensure the project to stay correct and robust. This includes both unit tests and integration tests.
Please add documentation to docs/source/ if the PR modifies the user-facing behaviors of vLLM. It helps vLLM user understand and utilize the new features or changes.

Adding or changing kernels

Each custom kernel needs a schema and one or more implementations to be registered with PyTorch.

Make sure custom ops are registered following PyTorch guidelines: Custom C++ and CUDA Operators and The Custom Operators Manual
Custom operations that return Tensors require meta-functions. Meta-functions should be implemented and registered in python so that dynamic dims can be handled automatically. See above documents for a description of meta-functions.
Use torch.libary.opcheck() to test the function registration and meta-function for any registered ops. See tests/kernels for examples.
When changing the C++ signature of an existing op, the schema must be updated to reflect the changes.
If a new custom type is needed, see the following document: Custom Class Support in PT2.

Notes for Large Changes

Please keep the changes as concise as possible. For major architectural changes (>500 LOC excluding kernel/data/config/test), we would expect a GitHub issue (RFC) discussing the technical design and justification. Otherwise, we will tag it with rfc-required and might not go through the PR.

What to Expect for the Reviews

The goal of the vLLM team is to be a transparent reviewing machine. We would like to make the review process transparent and efficient and make sure no contributor feel confused or frustrated. However, the vLLM team is small, so we need to prioritize some PRs over others. Here is what you can expect from the review process:

After the PR is submitted, the PR will be assigned to a reviewer. Every reviewer will pick up the PRs based on their expertise and availability.
After the PR is assigned, the reviewer will provide status update every 2-3 days. If the PR is not reviewed within 7 days, please feel free to ping the reviewer or the vLLM team.
After the review, the reviewer will put an action-required label on the PR if there are changes required. The contributor should address the comments and ping the reviewer to re-review the PR.
Please respond to all comments within a reasonable time frame. If a comment isn't clear or you disagree with a suggestion, feel free to ask for clarification or discuss the suggestion.

Thank You

Finally, thank you for taking the time to read these guidelines and for your interest in contributing to vLLM. Your contributions make vLLM a great tool for everyone!

github-actions · 2024-10-11T14:43:35Z

👋 Hi! Thank you for contributing to the vLLM project.
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 do one of these:

Add ready label to the PR
Enable auto-merge.

🚀

DarkLight1337 · 2024-10-11T14:46:27Z

@ML-GCN can you try out this PR and see if it fixes the issue for you?

ML-GCN · 2024-10-12T01:57:48Z

@ML-GCN can you try out this PR and see if it fixes the issue for you?

ok i try now

ML-GCN · 2024-10-12T02:59:05Z

@ML-GCN can you try out this PR and see if it fixes the issue for you?

not yet

from the error shows that only like 'q_proj', 'k_proj', 'v_proj', 'o_proj', 'gate_proj', 'up_proj', 'down_proj' ... this linear layer
but the cpm3-4b Trainable linear layer is
"kv_a_proj_with_mqa",
"up_proj",
"q_b_proj",
"kv_b_proj",
"o_proj",
"gate_proj",
"down_proj",
"q_a_proj"

jeejeelee · 2024-10-12T03:55:50Z

@ML-GCN could you provibe your lora config?

ML-GCN · 2024-10-12T04:43:04Z

@ML-GCN could you provibe your lora config?

{
"alpha_pattern": {},
"auto_mapping": null,
"base_model_name_or_path": "/dataNfs/pre-trained/MiniCPM3-4B",
"bias": "none",
"fan_in_fan_out": false,
"inference_mode": true,
"init_lora_weights": true,
"layer_replication": null,
"layers_pattern": null,
"layers_to_transform": null,
"loftq_config": {},
"lora_alpha": 16,
"lora_dropout": 0.1,
"megatron_config": null,
"megatron_core": "megatron.core",
"modules_to_save": null,
"peft_type": "LORA",
"r": 8,
"rank_pattern": {},
"revision": null,
"target_modules": [
"kv_a_proj_with_mqa",
"up_proj",
"q_b_proj",
"kv_b_proj",
"o_proj",
"gate_proj",
"down_proj",
"q_a_proj"
],
"task_type": "CAUSAL_LM",
"use_dora": false,
"use_rslora": false
}

jeejeelee · 2024-10-12T05:20:54Z

@ML-GCN Please try it again

ML-GCN · 2024-10-12T07:41:09Z

@ML-GCN can you try out this PR and see if it fixes the issue for you?

i try the first commit i found the output is error by origin params MiniCPM3-4B (https://huggingface.co/openbmb/MiniCPM3-4B)

example i input 你好 but the output is

Can you take a look？thanks you

ML-GCN · 2024-10-12T07:42:43Z

@ML-GCN Please try it again

There was no error reported but the output is error As I mentioned above

jeejeelee · 2024-10-12T08:19:19Z

It seems this PR is not responsible for the unsatisfactory result.

the main branch:

Prompt: 'Hello, my name is', Generated text: " [Your Name], and I'm a student in the field of computer science"
Prompt: 'The president of the United States is', Generated text: ' the head of state and the head of government of the United States. The role'
Prompt: 'The capital of France is', Generated text: ' Paris.'
Prompt: 'The future of AI is', Generated text: ' often associated with two categories: strong AI and weak AI.\n\n- Strong'
Prompt: '你好', Generated text: '！ 区隔理论是古斯塔夫.勒庞在《乌合'

current PR branch:

Prompt: 'Hello, my name is', Generated text: " [Your Name], and I'm a student in the field of computer science"
Prompt: 'The president of the United States is', Generated text: ' the head of state and the head of government of the United States. The role'
Prompt: 'The capital of France is', Generated text: ' Paris.'
Prompt: 'The future of AI is', Generated text: ' often associated with two categories: strong AI and weak AI.\n\n- Strong'
Prompt: '你好', Generated text: '！ 区隔理论是古斯塔夫.勒庞在《乌合'

ML-GCN · 2024-10-12T08:25:10Z

It seems this PR is not responsible for the unsatisfactory result.

the main branch:

Prompt: 'Hello, my name is', Generated text: " [Your Name], and I'm a student in the field of computer science"
Prompt: 'The president of the United States is', Generated text: ' the head of state and the head of government of the United States. The role'
Prompt: 'The capital of France is', Generated text: ' Paris.'
Prompt: 'The future of AI is', Generated text: ' often associated with two categories: strong AI and weak AI.\n\n- Strong'
Prompt: '你好', Generated text: '！ 区隔理论是古斯塔夫.勒庞在《乌合'

current PR branch:

Prompt: 'Hello, my name is', Generated text: " [Your Name], and I'm a student in the field of computer science"
Prompt: 'The president of the United States is', Generated text: ' the head of state and the head of government of the United States. The role'
Prompt: 'The capital of France is', Generated text: ' Paris.'
Prompt: 'The future of AI is', Generated text: ' often associated with two categories: strong AI and weak AI.\n\n- Strong'
Prompt: '你好', Generated text: '！ 区隔理论是古斯塔夫.勒庞在《乌合'

ok i try again thanks you

jeejeelee · 2024-10-12T08:33:35Z

I also tested the generated result using https://huggingface.co/openbmb/MiniCPM3-4B#inference-with-vllm. The result is :

你好，我是MiniCPM系列模型，由面壁智能和OpenBMB开源社区开发。详细信息请访问 https://github.com/OpenBMB/

ML-GCN · 2024-10-12T08:38:39Z

I also tested the generated result using https://huggingface.co/openbmb/MiniCPM3-4B#inference-with-vllm. The result is :
你好，我是MiniCPM系列模型，由面壁智能和OpenBMB开源社区开发。详细信息请访问 https://github.com/OpenBMB/

yes this is right but you first test the input is 你好 is error result

jeejeelee · 2024-10-12T08:48:24Z

@ML-GCN The first test is based on https://github.com/vllm-project/vllm/blob/main/examples/offline_inference.py, possibly due to the lack of the chat template.

jeejeelee · 2024-10-12T09:16:25Z

@ML-GCN sry, I don't quite understand what you mean.

ML-GCN · 2024-10-12T09:28:39Z

@ML-GCN sry, I don't quite understand what you mean.

o it is my error I read it wrong
sorry

jeejeelee · 2024-10-12T09:31:03Z

@ML-GCN sry, I don't quite understand what you mean.

o it is my error I read it wrong sorry

No worries. Does this PR resolve your previous issue?

ML-GCN · 2024-10-12T10:19:23Z

@ML-GCN sry, I don't quite understand what you mean.

o it is my error I read it wrong sorry

No worries. Does this PR resolve your previous issue?

yes thanks you very much
When will it be updated to Docker

ML-GCN

this pr sove my request thank you

DarkLight1337 · 2024-10-12T10:59:20Z

The introduction of GLM-4V is causing LoRA tests to fail on main. Can you fix it in this PR as well? I think you need to update LoRAModelManager.

         # Used to indicate whether the model is a multimodal model
-        self.supports_mm: bool = supports_multimodal(self.model)
+        self.supports_mm = (supports_multimodal(self.model)
+                            # In case the model only supports LoRA for
+                            # text modules (e.g. ChatGLM)
+                            and hasattr(self.model, "get_mm_mapping"))

jeejeelee · 2024-10-12T15:19:18Z

@DarkLight1337 I have made the modifications. I don't understand why GLM4V is integrated in this way, it a bit confusing. I think it would be better to use different model script to distinguish between VL and LLM.

DarkLight1337 · 2024-10-12T15:31:43Z

@DarkLight1337 I have made the modifications. I don't understand why GLM4V is integrated in this way, it a bit confusing. I think it would be better to use different model script to distinguish between VL and LLM.

This is because they use the same HF architecture name, so we can't distinguish between them when deciding which vLLM model to load.

DarkLight1337

LoRA tests pass should this should be good to merge.

Signed-off-by: Alvant <alvasian@yandex.ru>

Signed-off-by: Amit Garg <mitgarg17495@gmail.com>

Signed-off-by: Sumit Dubey <sumit.dubey2@ibm.com>

Signed-off-by: Maxime Fournioux <55544262+mfournioux@users.noreply.github.com>

Modfiy MiniCPM lm_head

86e01f9

DarkLight1337 added the ready ONLY add when PR is ready to merge/full CI is needed label Oct 11, 2024

jeejeelee added 2 commits October 12, 2024 13:19

Fix mincpm3 lora mapping bug

4ccacc3

Merge branch 'vllm-project:main' into modify-mincpm-embedding

b16c91f

Fix typo

03ff616

ML-GCN reviewed Oct 12, 2024

View reviewed changes

jeejeelee added 2 commits October 12, 2024 23:12

Fix ChatGLM LoRA bug

09ed306

Merge branch 'vllm-project:main' into modify-mincpm-embedding

cf0ba74

DarkLight1337 approved these changes Oct 12, 2024

View reviewed changes

DarkLight1337 enabled auto-merge (squash) October 12, 2024 15:56

simon-mo merged commit 250e26a into vllm-project:main Oct 12, 2024
58 of 61 checks passed

jeejeelee deleted the modify-mincpm-embedding branch October 13, 2024 01:00

Alvant pushed a commit to compressa-ai/vllm that referenced this pull request Oct 26, 2024

[Bugfix]Fix MiniCPM's LoRA bug (vllm-project#9286)

ab33c25

Signed-off-by: Alvant <alvasian@yandex.ru>

garg-amit pushed a commit to garg-amit/vllm that referenced this pull request Oct 28, 2024

[Bugfix]Fix MiniCPM's LoRA bug (vllm-project#9286)

57f262e

Signed-off-by: Amit Garg <mitgarg17495@gmail.com>

sumitd2 pushed a commit to sumitd2/vllm that referenced this pull request Nov 14, 2024

[Bugfix]Fix MiniCPM's LoRA bug (vllm-project#9286)

c9dd12e

Signed-off-by: Sumit Dubey <sumit.dubey2@ibm.com>

KuntaiDu pushed a commit to KuntaiDu/vllm that referenced this pull request Nov 20, 2024

[Bugfix]Fix MiniCPM's LoRA bug (vllm-project#9286)

4088bd6

mfournioux pushed a commit to mfournioux/vllm that referenced this pull request Nov 20, 2024

[Bugfix]Fix MiniCPM's LoRA bug (vllm-project#9286)

f980fe0

Signed-off-by: Maxime Fournioux <55544262+mfournioux@users.noreply.github.com>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Bugfix]Fix MiniCPM's LoRA bug #9286

[Bugfix]Fix MiniCPM's LoRA bug #9286

jeejeelee commented Oct 11, 2024 •

edited

Loading

github-actions bot commented Oct 11, 2024

DarkLight1337 commented Oct 11, 2024

ML-GCN commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

jeejeelee commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

jeejeelee commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

jeejeelee commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

jeejeelee commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

jeejeelee commented Oct 12, 2024

jeejeelee commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

jeejeelee commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

ML-GCN left a comment

DarkLight1337 commented Oct 12, 2024 •

edited

Loading

jeejeelee commented Oct 12, 2024

DarkLight1337 commented Oct 12, 2024 •

edited

Loading

DarkLight1337 left a comment

[Bugfix]Fix MiniCPM's LoRA bug #9286

[Bugfix]Fix MiniCPM's LoRA bug #9286

Conversation

jeejeelee commented Oct 11, 2024 • edited Loading

PR Title and Classification

Code Quality

Adding or changing kernels

Notes for Large Changes

What to Expect for the Reviews

Thank You

github-actions bot commented Oct 11, 2024

DarkLight1337 commented Oct 11, 2024

ML-GCN commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

jeejeelee commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

jeejeelee commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

jeejeelee commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

jeejeelee commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

jeejeelee commented Oct 12, 2024

jeejeelee commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

jeejeelee commented Oct 12, 2024

ML-GCN commented Oct 12, 2024

ML-GCN left a comment

Choose a reason for hiding this comment

DarkLight1337 commented Oct 12, 2024 • edited Loading

jeejeelee commented Oct 12, 2024

DarkLight1337 commented Oct 12, 2024 • edited Loading

DarkLight1337 left a comment

Choose a reason for hiding this comment

jeejeelee commented Oct 11, 2024 •

edited

Loading

DarkLight1337 commented Oct 12, 2024 •

edited

Loading

DarkLight1337 commented Oct 12, 2024 •

edited

Loading