EvolvingLMMs-Lab/LLaVA-OneVision-2

类库 › LLaVA-OneVision-2

EvolvingLMMs-Lab/LLaVA-OneVision-2

LLaVA-OneVision-2是一个完全开源的多模态训练框架，旨在推动多模态技术的普及。它提供下一代多模态模型、对齐视觉编码器及专用数据集（如视频描述和空间理解），支持用户进行高效的多模态模型训练与开发。

986 72 986 58

在 GitHub 上查看

技术栈

aiak_megatron/megatron/core python

查看全部依赖 (2)

依赖

packaging torch

aiak_megatron/requirements/pytorch_24.01 python

测试

pytest

查看全部依赖 (15)

依赖

einops flask-restful nltk nvidia-modelopt pytest-cov pytest-random-order pytest_asyncio pytest_mock sentencepiece tensorstore tiktoken triton wandb wrapt zarr

aiak_megatron/requirements/pytorch_24.07 python

测试

pytest

查看全部依赖 (15)

依赖

einops flask-restful nltk nvidia-modelopt nvidia-resiliency-ext pytest-cov pytest-random-order pytest_asyncio pytest_mock sentencepiece tensorstore tiktoken wandb wrapt zarr

aiak_megatron/requirements/pytorch_24.10 python

测试

pytest

查看全部依赖 (16)

依赖

einops flask-restful nltk nvidia-modelopt nvidia-resiliency-ext pytest-cov pytest-random-order pytest_asyncio pytest_mock sentencepiece tensorstore tiktoken torch wandb wrapt zarr

根目录 python

查看全部依赖 (23)

依赖

accelerate datasets einops ftfy hydra-core jsonlines mdformat mdformat-gfm mdformat-pyproject megatron-energon nltk omegaconf pandarallel pre-commit protobuf py-cpuinfo qwen_vl_utils sentencepiece tiktoken timm transformers wandb wrapt

EvolvingLMMs-Lab/LLaVA-OneVision-2