Product library

home page find GPTS Ranking List Build AI Submit market
home page machine learning
machine learning
Updated: 2024-06-21 11:51:46
newest Hot

Thousand Brains Project

AI project for exploring brain intelligence

The THOUSAND BRAINS Project was initiated by Jeff Hawkins and Numenta, aiming to develop a new type of artificial intelligence system by understanding the working principle of the neocortex. Based on the Thought Brains Theory of Intelligence, this project proposes a brain working principle that is fundamentally different from the traditional AI system. The goal of the project is to build an efficient and powerful intelligent system that can realize the intelligent capabilities possessed by human beings. Numenta has opened its research resources, including meeting minutes, code open source, and established a large community around its algorithms. The project has received financial support from the Gates Foundation and others, and encourages global researchers to participate in or join this exciting project.
artificial intelligence neuroscience

ShareGPT4Video

AI model for improving video understanding and generation.

ShareGPT4Video series aims to promote video understanding of large video language models (LVLMs) and video generation of text to video models (T2VMs) through dense and accurate subtitles. This series includes: 1) ShareGPT4Video, 40K GPT4V annotated dense video subtitles, developed through carefully designed data filtering and annotation strategies. 2) ShareCaptioner Video, an efficient and powerful arbitrary video caption model, is annotated with 4.8M high-quality aesthetic video. 3) ShareGPT4Video-8B, a simple but excellent LVLM, has achieved the best performance in three advanced video benchmarks.
Video understanding Text to Video

ComfyUI-LuminaWrapper

A Python wrapper for Lumina model

ComfyUI LuninaWrapper is an open source Python wrapper used to simplify the loading and use of Lumina models. It supports user-defined nodes and workflows, enabling developers to more easily integrate Lumina models into their projects. This plug-in is mainly for developers who want to use Lumina model for deep learning or machine learning in Python environment.
High quality new products
machine learning Deep learning

Promoted

Improve the ranking of market search, recommendation and native advertising

Promoted is a solution focusing on market search, recommendation and native advertising ranking. Through advanced machine learning technology and large language model search relevance AI technology, it significantly improves the conversion rate and advertising quality. It provides a unified search, recommendation and advertising service for the market platform to help enterprises achieve better matching and higher revenue.
Selected from abroad
Market search Recommendation system

L4GM

4D reconstruction model, fast generation of animated objects

L4GM is a 4D large-scale reconstruction model, which can quickly generate animated objects from single view video input. It uses a novel dataset that contains multi view videos that show animated objects rendered in Objaverse. This dataset contains 44K different objects and 110K animations, which are rendered from 48 perspectives and generate 12M videos, including 300M frames in total. L4GM is constructed based on the pre trained 3D large-scale reconstruction model LGM, which can output 3D Gaussian ellipsoids from the multi view image input. L4GM outputs the 3D Gaussian Splatting representation of each frame, and then samples it to a higher frame rate to achieve time smoothing. In addition, L4GM also adds a time self attention layer to help learn the consistency of time, and uses the multi view rendering loss of each time step to train the model.
High quality new products
4D reconstruction Animation generation

Index-1.9B

The lightweight big language model independently developed by BiliBili

Index-1.9B series is a lightweight big language model independently developed by Bilibili Company, including multiple versions, such as base, pure, chat and character. It is suitable for pre training of Chinese and English based language materials, and has excellent performance on multiple evaluation benchmarks. The model supports SFT and DPO alignment, as well as RAG technology for role play customization, and is suitable for dialog generation, role play and other scenarios.
High quality new products
Large language model natural language processing

Meta Chameleon

Advanced machine learning model helps non-commercial research.

Meta Chameleon is a machine learning model developed by Meta Corporation. It provides support for non-commercial research purposes, including research, development, education, processing or analysis. It does not aim at commercial interests or monetary compensation for you or others. The model includes machine learning model code, trained model weight, reasoning enabling code, training enabling code, fine-tuning enabling code, demonstration materials, etc.
machine learning artificial intelligence

multi-token prediction

Multi token prediction model to improve the efficiency and performance of language model

The multi token prediction model is a technology developed by Facebook based on a large language model. It aims to improve the efficiency and performance of the model by predicting multiple future tokens. This technology allows the model to generate multiple tokens in a single forward propagation, thus accelerating the generation speed and possibly improving the accuracy of the model. This model is provided free of charge for non-commercial research purposes, but it should comply with Meta's privacy policy and relevant laws and regulations.
Language model Multi token prediction

Glyph-ByT5-v2

Powerful aesthetic baseline for multilingual visual text rendering

Glyph-ByT5-v2 is a model launched by Microsoft Research Asia for accurate multilingual visual text rendering. It not only supports accurate visual text rendering in 10 different languages, but also significantly improves the aesthetic quality. This model builds a multilingual visual paragraph benchmark by creating high-quality multilingual font text and graphic design data sets, and uses the latest gait awareness preference learning methods to improve the quality of visual aesthetics.
Multilingual Visual text rendering

HelpSteer2

An open source dataset for training high-performance reward models.

HelpSteer2 is an open source dataset released by NVIDIA, which aims to support training to align models to make them more helpful, factual and coherent, and has adjustability in response complexity and redundancy. This dataset was created in cooperation with Scale AI. When used with the Llama 370B basic model, it achieved 88.8% performance on RewardBench, which is one of the best reward models as of June 12, 2024.
Open source dataset Reward model

VideoTetris

Innovative framework for text to video generation

VideoTetris is a novel framework, which implements text to video generation, and is particularly suitable for processing complex video generation scenes containing multiple objects or dynamic changes in the number of objects. The framework accurately follows the complex text semantics through the spatial and temporal combination diffusion technology, and is implemented by operating and combining the spatial and temporal attention maps of the denoising network. In addition, it also introduces a new reference frame attention mechanism to improve the consistency of autoregressive video generation. VideoTetris has achieved impressive qualitative and quantitative results in combining text to video generation.
High quality new products
Text to Video Video generation

Samba

Official Implementation of Efficient Infinite Context Language Model

Samba is a simple and powerful hybrid model with unlimited context length. Its architecture is very simple: Samba=Mamba+MLP+sliding window attention+hierarchical MLP stacking. The Samba-3.8B model has trained 3.2 trillion tokens on the Phi3 dataset, and its performance on major benchmark tests (such as MMLU, GSM8K, and HumanEval) has greatly exceeded that of Phi3-mini. Samba can also achieve perfect long context retrieval capability with minimal instruction adjustment, while maintaining linear complexity with sequence length. This makes Samba-3.8B-instrument perform well on downstream tasks (such as long context summary).
High quality new products
natural language processing machine learning

Depth Anything V2

Advanced monocular depth estimation model

Depth Anything V2 is an improved monocular depth estimation model, which provides more precise and robust depth prediction than the previous version by using synthetic images and a large number of unlabeled real images for training. This model has significantly improved in efficiency and accuracy, and is more than 10 times faster than the latest model based on Stable Diffusion.
Depth estimation Monocular vision

YaFSDP

An efficient distributed data parallel framework designed for large language models.

YaFSDP is a distributed data parallel framework designed for good cooperation with transformer like neural network structures. It is 20% faster than traditional FSDP when pre training large language models (LLMs), and performs better under high memory pressure. YaFSDP is designed to reduce the overhead of communication and memory operations.
Distributed Computing Data parallelism

transformers.js

Run the advanced machine learning model directly in the browser.

Transformers.js is a JavaScript library designed to provide advanced machine learning capabilities for web pages. It allows users to run the pre trained Transformers model directly in the browser without server support. The library uses the ONNX Runtime as the backend, and supports the conversion of PyTorch, TensorFlow, or JAX models to the ONNX format. Transformers.js is equivalent to the transformation Python library of Hugging Face, and provides similar APIs, enabling developers to easily migrate existing code to the web side.
machine learning Transformers

MASA

A general model for object matching in cross domain video frames.

MASA is an advanced model for object matching in video frames, which can handle multiple object tracking (MOT) in complex scenes. MASA does not rely on the annotation video dataset of a specific domain, but learns the correspondence at the instance level through the rich object segmentation of the Segment Anything Model (SAM). MASA has designed a universal adapter, which can be used together with the basic segmentation or detection model to achieve zero sample tracking capability, and can perform well even in complex fields.
High quality new products
computer vision target tracking

sherpa-onnx

Open source project supporting multiple speech recognition and speech synthesis functions

Sherpa onnx is a speech recognition and speech synthesis project based on the next generation Kaldi. It uses onnxruntime for reasoning and supports a variety of voice related functions, including speech to text (ASR), text to speech (TTS), speaker recognition, speaker verification, language recognition, keyword detection, etc. It supports multiple platforms and operating systems, including embedded systems Android、iOS、Raspberry Pi、RISC-V、 Server, etc.
speech recognition speech synthesis

Stable Audio Open 1.0

AI model for generating variable length stereo audio based on text prompts.

Stable Audio Open 1.0 is an AI model that uses self encoder, T5 based text embedding and transformer based diffusion model to generate stereo audio of up to 47 seconds. It generates music and audio through text prompts, supports research and experiments, and explores the current capabilities of generative AI models. The model is trained on the datasets of Freesound and Free Music Archive (FMA) to ensure data diversity and copyright legitimacy.
High quality new products
AI music generation Audio processing

Mamba-2

New development of efficient sequence models

Mamba-2 is a new sequence model developed by Goomba AI Lab, which aims to improve the efficiency and performance of sequence models in the machine learning community. It combines the advantages of the state space model (SSM) and attention mechanism through the structured state space dual (SSD) model, providing a more efficient training process and a larger state dimension. Mamba-2's design allows the model to use matrix multiplication during training, thus improving hardware efficiency. In addition, Mamba-2 performs well in tasks such as multiple query associative memory (MQAR), showing its potential in complex sequence processing tasks.
High quality new products
Sequential model State space model

AutoCoder

The new model of code generation task has higher test accuracy than GPT-4 Turbo.

AutoCoder is a new model designed specifically for code generation tasks. Its test accuracy on the HumanEval benchmark dataset exceeds GPT-4 Turbo (April 2024) and GPT-4o. Compared with the previous open source model, AutoCoder provides a new function: it can automatically install the required package and try to run the code when the user wants to execute the code until it is confirmed that there is no problem.
machine learning Text generation

© 2024     AIbase    Record No.: Min ICP Bei 08105208 No.14

Privacy Policy

User Agreement

Feedback Site Map