Havuz

:sunrise: A comprehensive, cutting-edge open-source framework for next-generation Transformers/LLM, with compute-matched benchmarking and a

多模态检索实战：意图识别、Prompt 工程、Qwen LoRA、EmbeddingGemma 2 与 InfoNCE 对比学习。含真实演示、自有数据接入、训练评测及 RAG 检索组件学习指南。

Local AI BGM and sound effects for Claude Code and any MCP client. Stable Audio 3 on Apple Silicon (MLX) and Windows NVIDIA GPUs (PyTorch).

🚀🚀 A multimodal System One decision model that gives calibrated answers to typed questions about screens, photos, video and text in one forw

AuK, tek bir doğal dil talimat arayüzüyle konuşma üretimi ve düzenleme görevlerini gerçekleştiren açık ağırlıklı bir speech modelidir.

Neden yükseliyor · Yeni yayımlanmış bir model olarak kısa sürede 1308 yıldız almış; son 28 gündeki 30 commit ve 5 farklı yazar da aktif geliştirmeye işaret ediyor. README ayrıca vLLM-Omni, SGLang-Omni ve audio.cpp desteğini, modelin farklı inference ekosistemlerine taşınmasını sağlayan somut ilgi göstergeleri olarak listeliyor.

Phonon: open speech recognition models (Phonon-2, Phonon-1) — CLI, CPU and CUDA images

🔥🔥 Accelerating decision inference with exact state forking

A small chat model trained from scratch: tokenizer → pretraining → SFT → RL → multimodal. 0.78B params, CORE 0.248, tool calling, vLLM. Weig

This project is for learning: it shows how an agent loop works underneath, in a way more intuitive and visual than reading about it. Behind

Bilingual (中文+EN) ML / LLM / diffusion / agent interview cheat sheets for AI 秋招 — generated by ARIS /interview-cheatsheet, rendered by /rend

Free Course : Data, Machine Learning, Deep Learning and AI Engineering, the long way round: the math, the code, the systems and the security

结合预训练编码器、BiLSTM 与 GlobalPointer 的实体关系联合抽取模型，支持标注校验、训练、评估及三元组预测。

OpenMythos, Claude Mythos’un gerçek mimarisini temsil ettiği kanıtlanmayan, araştırma literatürüne dayalı varsayımsal bir Recurrent-Depth Transformer uygulamasıdır.

Neden yükseliyor · Claude Mythos’a doğrudan atıf yapan adı ve açıklaması, varsayımsal mimariyi incelemek isteyenlerin ilgisini çekmiş olabilir; 162 günde 14.904 yıldız bu ilgiyi gösteriyor. Bu, benimseme veya teknik doğrulama kanıtı değil: son 28 günde commit ve farklı yazar sayısı sıfır.

Halo is an open-source framework built by White Circle for training large language and multimodal models

Automated alignment adjustment for LLMs — direct steering, LoRA, and MoE expert-granular abliteration, optimized via multi-objective Optuna

Full-featured DAW, DJ, and VJ app interoperable w/Ableton, Reaper, Resolume & more. Stable Audio 3, Magenta RT2, Suno API, Chimera track fus

🔥 AI agent that ports CUDA code to AMD ROCm using local LLMs — FastAPI + React + Ollama | Live: rocmporter-agent.vercel.app

From PyTorch model to end-to-end TensorRT inference experience in two commands—AI-native, cross-platform, and built for the best possible us

Iris-3B pixel-space text-to-image on Apple Silicon GPUs: Metal-optimized engine, native macOS app, CLI, Python and local HTTP API

Best AI Workload Orchestrator for Kubernetes Python 2026

A self-hosted studio for generating images, video, and audio with diffusion models.

Implementation of LatentMoE,Toward Optimal Accuracy per FLOP and Parameter in Mixture of Experts (Elango et al., NVIDIA 2026) — in Pytorch.

Lev: Thinking Fast and Deep with Looped Decision Models. One checkpoint, multiple depths: System 1.5 decision models that think in loops, no

Qwen-Image-2.1 ile Apple Silicon üzerinde yerel görsel üretimi ve çoklu görsel düzenleme sağlayan bir CLI ile helmstudio’ya bağlı web arayüzü sunuyor.

Neden yükseliyor · Beş günlük yaşında 55 yıldız ve son 28 günde 12 commit, iki farklı yazar görünüyor. Bu ilk ilgi ve geliştirme hareketliliğine işaret ediyor; ancak kısa geçmiş ve sınırlı yazar sayısı benimsenme ya da kalıcı bakım kanıtı değil.

Confucius4-TTS: a Multilingual and Cross-Lingual Zero-Shot TTS Engine

Jacobian-Brainwash : A manual alignment tool for large language models built on Anthropic's Jacobian Lens. Results are exportable.

Self-tuning multi-agent AI trading system. 8-source signal fusion, Bull/Bear/Judge debate on Claude Opus 4.7, Kelly + ATR position sizing. P

Transformers for decision models: fine-tune, load and serve calibrated decsion-model on any Hugging Face model.

Build world models from scratch — from perception and latent dynamics to planning, JEPA, VLA, and 3D/4D intelligence.

[ICML 2026] NanoQuant: Efficient Sub-1-Bit Quantization of Large Language Models

Train JEV-style System 1 decision models on Qwen3.5 and Qwen3.8-27B: 24-slot decision head, KL distillation, calibration and serving. 在 Qwen

Official implementation of "Unifying Distributional Training for One-Step Visual Generation" (MGFlow).

Single-GPU LLM fine-tuning preflight: memory estimates, runnable LoRA/QLoRA recipes, and local measurements.

To know what models don't say out loud.

Self-Tuning Multi-Agent AI Trading System 2026: 8-Source Signal Fusion & Kronos Model

PyTorch 기반 한국어 LLM 구현 및 학습 프로젝트

Tiny local text-to-24x24 pixel art model, trained on roughly 200K samples in 30 minutes on an RTX 5090.

First open-source implementation of Google TurboQuant (ICLR 2026) -- near-optimal KV cache compression for LLM inference. 5x compression wit

Production-ready, bleeding-edge, upstream-first vLLM Docker image for the NVIDIA DGX Spark (GB10 / sm_121a), verified against real models on

End-to-end pipeline for seeing how LLMs actually process your prompts. Capture attention across every layer, render heatmaps and cooking cur

This repo is a alternative solution for standard self-attention. It uses tree routing with learnable functions for cheaper self-attention wh

Recomputation-free, context-independent KV caching for large language models

Open, calibrated System 1 decision models: typed choice / score / yes-no answers with a probability for every option. 400M to 80B; CPU, NVID

Unofficial fairseq-free PyTorch implementation of UTMOS (v1, 2022), matching the original system.

Building an Image-to-Video (I2V) Model from Scratch

Taiwanese Hokkien (Taigi) speech-to-text transcriber - MediaTek Breeze-ASR-26 with faster-whisper, tuned for RTX 3050 4GB low-VRAM GPUs. Gra

Local text-to-speech: 28 voices, 10 languages, voice cloning, SDKs for Python, Swift, Go, Rust and TypeScript

Tiny-MoE is a lightweight Mixture-of-Experts language model built entirely from scratch in native PyTorch and trained end-to-end on Kaggle u

PyTorch-native framework for fault-tolerant distributed training of foundation models on AI clusters