Rising in Models & inference
Open models, inference engines, fine-tuning and evaluation.
2,836 repos tracked in this topic · RSS
- deepseek-ai/DeepInfer
Open-source offline inference framework for DeepSeek models, offering optimized kernels, MoE expert and attention parallelism, speculative decoding, asynchronous scheduling, deterministic sampling and
- AndreSzymanski15/ai-jailbreaks
Collection of 35 copy-paste prompts (system prompts, persona definitions, AGENTS.md files, Ollama Modelfiles) that aim to alter how 25+ LLMs respond, including refusal behavior. Documented for AI secu
- Fractal-Compute-Orchestrations/FractalAndroid
Android client that runs on-device federated training with TFLite and partitioned batch training for the Fractal network, gated by battery, temperature, Wi-Fi and memory telemetry. Uploads parameter c
- chenglou/handwired
A research project that compiles ordinary programs with if-else branches, loops and state into neural networks built only from matrix multiplication and ReLU over floats. The weights are written by ha
- OpenMOSS/MOSS-Image
Official code for MOSS-Image, a 5B-parameter diffusion transformer for text-to-image generation. It supports text rendering inside images, JSON-based layout control of element placement, prompts up to
- mikroverk/esp32sim
An ESP32 emulator written in Rust that runs unmodified firmware, including the mask ROM, bootloader and app, on a computer or in a browser. It models board peripherals such as displays, touch, LEDs, a
- HazyResearch/ncm
Research code for Next-Concept Modeling, a two-stage framework for parallel text generation. A multiscale residual VQ tokenizer encodes a sequence as a hierarchy of discrete codes, and a transformer p
- MemTensor/MemT-1.0
A research model family, MemT-1.0, that adds trainable native memory blocks alongside transformer attention so models retain historical information across turns without re-reading the original context
- mistralai/timestral
Standalone inference code for Timestral, a family of time series foundation models for multivariate probabilistic forecasting. Supports past-only, known-future and categorical covariates, outputs 21 q
- datalab-to/indra
Indra OCR is a Python OCR tool with a 1M-parameter recognizer that runs on ONNX Runtime on CPU, without PyTorch. It detects lines and layout blocks with reading order, reads digital PDF text layers, a
- cobanov/PS5LM
Runs llama.cpp on a jailbroken PS5, using the console GPU through Vulkan to load GGUF models. Serves OpenAI and Anthropic compatible APIs over the local network, plus a chat page, a decision model end
- MiaAI-Lab/DeepSeek-v4.1-Flash-1017-2x-DGX-Sparks
Serves DeepSeek-V4.1-Flash-1017 across two NVIDIA DGX Sparks through an OpenAI-compatible API, using TensorFold's native Zig engine and a 2.9 bpw EXL3 quantization. Supports long context up to 1,048,5
- StartLuxLabs/StartLux-Decision
Open-weights decision models from 0.8B to 27B parameters that return a calibrated probability for every option of each typed question (choice, yes/no, or scale level) in one pass, with an optional sho
- flybrainlabs/miner
Native and WebAssembly simulation engines for deterministic integer leaky integrate-and-fire (LIF) networks, with browser workers, persistent model and result storage, three-stage mining, and cancella
- inclusionAI/Ming-3.0-tts
Multilingual text-to-speech models (0.5b dense and tiny MoE) with voice cloning, voice design from text, built-in voices, natural-language delivery instructions and inline paralinguistic tags such as
- corrinehu/dsh-workbuddy-connect
Connects the models available in the logged-in WorkBuddy and WorkBuddy AI desktop apps to DeepSeek Harness as a plugin. It reuses the app's login and credits without manual configuration and shows a s
- 0814wdwd/solo_jev
Python library that reorders rows and fields of tabular or JSON inputs so a prefix-caching LLM backend can reuse more computation, then returns decisions in the original row order. Includes a client f
- 0c-biggo/0c-ime
免選字的注音輸入法：在本機用 Transformer 看前文選字。支援 Linux（Fcitx5／IBus）、macOS、Windows。
- YeonwooSung/llm-architecture-from-scratch
A collection of from-scratch PyTorch implementations of modern LLM architectures, including GPT-2 XL, Llama variants, DeepSeek V3, Qwen3, and Gemma 3. A comparison table lists each model's normalizati
- linguo2625469/workbuddy2api-panel
Go gateway that exposes CodeBuddy accounts as an OpenAI-compatible /v1/chat/completions endpoint. It rotates across a pool of accounts, applies cooldowns and circuit breaking on upstream errors, keeps
- Moxin-Tech/moworld_base
A research codebase for MoWorld, a flash world model that generates real-time interactive video from a first frame, a text prompt and camera controls. It covers training on Ascend NPUs with few-step a
- Edge0-AI/Edge0
Open-source streaming MoE inference framework that runs large sparse-MoE models on consumer hardware by streaming expert weights from SSD on demand. A trained prerouter predicts expert routing one ste
- BaruqueRodrigues/llm-legislativo
Builds a large corpus of Brazilian federal legislative documents from the Chamber of Deputies and Senate archives, with deduplication and family-based train/test splits. Trains a causal Transformer la
- jasonxu114514/opencode2api
A Go gateway that exposes OpenAI Chat Completions, Responses and Anthropic Messages endpoints in front of OpenCode Zen and Zen Go. It translates between the protocols and manages upstream keys and pro
- Niko1221/Strata
Strata is a C++ inference engine that runs the Qwen3.8-Flash-Next model on consumer hardware with NVIDIA, AMD, Intel or Apple Silicon GPUs. It exposes OpenAI and Anthropic compatible APIs on localhost
- Indras-Mirror/Xolotl
A C++ inference engine and OpenAI/Anthropic-compatible server, forked from Strata, that runs large MoE models such as DeepSeek-V4.1-Flash (552B) on one 24 GB NVIDIA GPU by tiering expert weights acros
- mw00/project-maya
A C++ engine, server and dashboard for running the GLM-5.3-Flash and DeepSeek-V4.1-Flash models locally on one to sixteen NVIDIA or AMD GPUs, on Linux or Windows. It offers OpenAI- and Anthropic-compa
- timescale/pg_vectorsearch
A PostgreSQL extension providing the PRISM approximate nearest neighbor index, an IVF-style index with a hierarchical centroid tree and RaBitQ quantization. Supports vec32 and vec16 types and pgvector
- zjwfufu/SMD
A training and inference codebase for few-step diffusion distillation using an on-policy method. It provides configurations for distilling text-to-image, image editing, text-to-video, audio-video and
- Panchovix/nvidia-vbios-vix-editor
Offline editor for NVIDIA Blackwell, Ada, Ampere and Turing VBIOS images, with a GTK window and a JSON command line. It edits power limits, voltage ceilings, clock offset ranges, fan settings and rela
See all 2,836 repos in this topic
Get alerts for repos like this
We'll email you when something in Models & inference starts rising. Free, once a week, only what matches.
18,267 GitHub repos watched and ranked every hour. Figures as of 20 October 2026.
What arrives: one email on Mondays with the repos that match, never more. See last week's. Privacy