add-glm4v model #1157
add-glm4v model #1157
Conversation
Summary of ChangesHello @SangChengC, I'm Gemini Code Assist1! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request integrates the GLM4V multimodal model into the LightLLM framework. It provides the necessary vision processing capabilities, including a custom vision transformer and a specialized tokenizer, to enable the model to understand and respond to visual inputs. The changes also refine the rotary embedding configuration to support the specific requirements of the GLM4V architecture. Highlights
🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console. Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request adds support for the GLM-4V model. The changes are well-structured, introducing a new model implementation with its visual component, transformer layers, and weight loading logic. The implementation thoughtfully reuses and extends existing components from Llama and Qwen models, such as the RoPE implementation and image processor. Modifications to the base Llama model configuration handling and the mrope Triton kernel have been made to accommodate GLM-4V's specific architecture. I've provided a few minor suggestions to improve code quality, mainly fixing typos in error messages and removing redundant code.
| elif self.data_type in ["fp32", "float32"]: | ||
| self.data_type = torch.float32 | ||
| else: | ||
| raise ValueError(f"Unsupport datatype {self.data_type}!") |
There was a problem hiding this comment.
There is a typo in the error message. 'Unsupport' should be 'Unsupported'.
| raise ValueError(f"Unsupport datatype {self.data_type}!") | |
| raise ValueError(f"Unsupported datatype {self.data_type}!") |
| img_tensors.append(pixel_values) | ||
| img_grids.append(image_grid_thw) | ||
| else: | ||
| raise Exception("Unsupport input types: {} for {}".format(type(img), img)) |
There was a problem hiding this comment.
There is a typo in the exception message. 'Unsupport' should be 'Unsupported'.
| raise Exception("Unsupport input types: {} for {}".format(type(img), img)) | |
| raise Exception("Unsupported input types: {} for {}".format(type(img), img)) |
| @@ -0,0 +1,104 @@ | |||
| import torch | |||
| import torch.functional as F | |||
There was a problem hiding this comment.
The import torch.functional as F is unused in this file and can be removed. Note that torch.functional is also a deprecated alias for torch.nn.functional.
| class Glm4VPreAndPostLayerWeight(Qwen2PreAndPostLayerWeight): | ||
| def __init__(self, data_type, network_config, mode): | ||
| super().__init__(data_type, network_config, mode) | ||
| return |
There was a problem hiding this comment.
This return statement is redundant at the end of an __init__ method, which implicitly returns None. It can be removed for cleaner code.
| def load_hf_weights(self, weights): | ||
| rename_weight_keys(weights) | ||
| super().load_hf_weights(weights) | ||
| return |
There was a problem hiding this comment.
This return statement is redundant at the end of a method that doesn't explicitly return a value. It can be removed for cleaner code.
|
|
||
| def __init__(self, kvargs): | ||
| super().__init__(kvargs) | ||
| return |
There was a problem hiding this comment.
This return statement is redundant at the end of an __init__ method, which implicitly returns None. It can be removed for cleaner code.
| repair_config(self.config, same_names=["num_hidden_layers", "n_layer"]) | ||
| if self.finetune_config: | ||
| self.config["vocab_size"] = self.finetune_config.vocab_size | ||
| return |
There was a problem hiding this comment.
This return statement is redundant at the end of a method that doesn't explicitly return a value. It can be removed for cleaner code.
No description provided.