-
Notifications
You must be signed in to change notification settings - Fork 329
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[Feature] Support LLaVA #196
Merged
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This was
linked to
issues
Nov 2, 2023
pppppM
approved these changes
Dec 26, 2023
llkn-2
pushed a commit
to llkn-2/xtuner
that referenced
this pull request
Jul 31, 2024
* v1 * add load_image * update cfg image url * del fig * update * temp * update convert * update chat_mm * add exclude_frozen_parameters for deepspeed * update chat * update xtuner help msg * fix bugs * revert bf16 deepspeed * fix bugs * add visual_select_layer for chat * improve pth_to_hf * rename projecter_pth to pretrained_pth * temp * update requirements * add cfgs * update * fix pre-commit * optim chat * optim chat * Delete xtuner/model/unused.py * move dispatch to a deeper folder * add projector * update * del model/projector * fix bugs * add docs * update * update * update * update * enhance resume for map_fn * update import * add llava_internlm_chat_7b_clip_vit_large_p14 * update dispatch * update dispatch * add link * update max_length * update max_length * update hyp * align * move yi flash attn * fix pre-commit * update deepspeed requirements * add mmbench script * install openpyxl * add entry_point for mmbench * save args * update mmbench * update max_length * add llama2 qlora * update mmbench * fix mmbench bugs * use osp instead of os.path * refactor pth_to_hf * update chat and mmbench to support --llava * align to chat * update entry_point * add vicuna template * add vicuna_7b_v15 * fix pre-commit * add vicuna_7b_v1.5 qlora * skip_special_tokens for decode text * remove do_sample * add warmup * fix pre-commit * Update dataset_prepare.md * Update dataset_prepare.md * Add KEEP_STSTEM for template * remove * fix vicuna template * clean cfgs * add cfgs * fix pre-commit * add --language for mmbench * fix bugs * fix pretrain bug * support visual_encoder lora * fix bugs * add paramwise_cfg * remove print_peft_model_trainable_parameters * fix bugs * add paramwise_cfg for DeepSpeedOptimWrapper * fix engine deepspeed paramwise_cfg bug * fix encode_fn bug * fix * fix pad_image_to_square bugs * Add space for system to avoid mismatch of 'USER' token * revert to adding bos_token at each conv * revert for paramwise_cfg * better cfgs? * fix import bug * fix import bug * pretrain align * update prepare_inputs_labels_for_multimodal * 1792 * support length_grouped_samplers * 1792 * remove KEEP_SYSTEM * remove system in cfg * update 336 cfg * add torch_dtype for mmbench and chat * group 50 * quant for pretrain * update cfgs * refactor cfgs * add length for concat dataset * update requirements * fix typo * add template for internlm pretrain * no zh * remove 20b cfgs * fix pre-commit * revert invalid input * rename * Update README.md * Update README_zh-CN.md * fix pre-commit * remove llava_zh from docs * qlora 512 * rename llava map_fn * update cfgs * update model urls * add docs link * add llava docs * update docs * update urls * add citation * fix README * move * update * vicuna pretrain with prompt * rename * add results * fix pre-commit * update * update * update * update * update * update * update * update * update * update * update * update * Update README.md * Update README_zh-CN.md * Update README_zh.md * Update README_zh.md * Update README.md * Update README_zh.md * Update README.md * Update README.md * fix typo * fix * Update README.md * Update README_zh-CN.md * rename * auto cn_string * fix pre-commit * rename * remove language * add VLMEvalKit * rename VLLM to VLM * add the download links of MMBench * update * update readme * update * update * update merge * fix cfg bug * Update README.md * Update README_zh.md * update * fix * update requirements * Update runtime.txt * Update runtime.txt * Update runtime.txt * Update README.md * Update README.md * Update README_zh.md * fix pre-commit * fix * update mmbench prompt * fix bugs * fix bugs * update docs * update * update * Update README.md
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Docs: https://github.com/LZHgrla/xtuner/blob/lzh/llava/xtuner/configs/llava/README.md
Features
TODO
exclude_frozen_parameters
forDeepSpeedStrategy
open-mmlab/mmengine#1415exclude_frozen_parameters
forDeepSpeedStrategy
'sresume
open-mmlab/mmengine#1424module_state_dict
microsoft/DeepSpeed#4587Table.__getstate__
andTable.__setstate__
huggingface/datasets#6444Enable paramwise_cfg for DeepSpeedOptimWrapper, [Fix] Delete frozen parameters when usingparamwise_cfg
open-mmlab/mmengine#1441LLaVA_XTuner
models open-compass/VLMEvalKit#17datasets