Overview of Recent Open‑Source AI Models and Tools (November 2023)

Content Overview

Recently announced open‑source projects span multiple AI categories, including natural‑language processing, multimodal processing, intelligent agents, recommendation systems, and model training/acceleration technologies.

In natural‑language processing, two notable projects are highlighted: the OrionStar‑Yi‑34B‑Chat model, a bilingual chat model fine‑tuned from Yi‑34B with over 150 K high‑quality corpora, and the 15‑billion‑parameter SQLCoder, which converts natural‑language queries into SQL statements for easier database interaction.

For multimodal processing, several projects are mentioned: the Monkey model, which raises input resolution to 896 × 1344 px without pre‑training; Google DeepMind’s GraphCast, a global weather‑forecast model that predicts ten‑day weather maps in one minute; and Alibaba Cloud’s Qwen‑Audio, a large‑scale audio‑language model supporting diverse audio and text inputs and outputting text.

Additional projects address intelligent agents and recommendation systems, such as Lumos (an open‑source LLM agent with unified data formats and modular design) and Agent4Rec, a large‑model‑based recommendation system simulator that mimics real user interactions. ProAgent combines large‑model agents with workflow automation to assist humans in building automated pipelines.

Model training and acceleration techniques are also featured. S‑LoRA introduces a novel tensor‑parallel strategy and highly optimized custom CUDA kernels for heterogeneous batch processing of LoRA computations. LCM‑LoRA provides a generic Stable Diffusion acceleration module for faster image generation.

Project Details

Creation Date

Project Name

Project Description

Project URL

2023‑11‑16

OrionStar‑Yi‑34B‑Chat

Open‑source bilingual chat model fine‑tuned from Yi‑34B using over 150 K high‑quality corpora.

HuggingFace

2023‑11‑16

SQLCoder

15‑billion‑parameter model that translates natural‑language questions into SQL queries.

HuggingFace

2023‑11‑16

Monkey

Multimodal large model that raises input resolution to 896 × 1344 px without pre‑training.

Github

2023‑11‑16

Lumos

Open‑source LLM agent with unified data format and modular design.

Github

2023‑11‑15

GraphCast

Google DeepMind’s global weather‑forecast model that predicts 10‑day weather in 1 minute.

Github

2023‑11‑15

Qwen‑Audio

Alibaba Cloud’s large‑scale audio‑language model supporting multiple audio and text inputs, outputting text.

Github

2023‑11‑15

Agent4Rec

Large‑model agent‑based recommendation system simulator that mimics real user interactions.

Github

2023‑11‑14

ProAgent

Intelligent‑agent workflow automation combining large‑model agents to help humans build pipelines.

Github

2023‑11‑14

mPLUG‑Owl2

Multimodal large language model from DAMO Academy achieving SOTA on text and multimodal datasets.

HuggingFace

Github

2023‑11‑14

dolphin‑2_2‑yi‑34b

Yi‑based model fine‑tuned with a 16 k context window.

HuggingFace

2023‑11‑14

Nous‑Capybara‑34B

Yi‑34B model extended to 200 K context length and trained on the Capybara dataset.

HuggingFace

2023‑11‑13

S‑LoRA

Novel tensor‑parallel strategy and custom CUDA kernels for heterogeneous batch LoRA computation.

Github

2023‑11‑13

MathVista

Multimodal math‑reasoning benchmark dataset.

HuggingFace

Github

2023‑11‑13

GLaMM

Multimodal dialogue model that generates natural‑language answers tightly coupled with object segmentation masks.

Github

2023‑11‑13

LCM‑LoRA

General Stable Diffusion acceleration module for faster image generation.

HuggingFace

Github