AI Audio2023
Unifies text-to-speech, singing voice synthesis, voice conversion, and text-to-audio/music in one PyTorch framework with shared vocoders and a common evaluation pipeline. Ships recipes, pretrained checkpoints, and visualizations of classic models.
AI Client2023
Hands-free voice-first companion with a Live2D avatar for real-time conversations with LLMs. Cross-platform web and desktop clients, runs locally or via cloud APIs, supports local ASR/TTS and modular customization for personas and models.
AI Train2023
Collection of runnable model implementations — LLaMA, Mistral, Stable Diffusion, Whisper, CLIP, plus LoRA fine-tuning — ported to the MLX array framework so they run natively on Apple silicon's unified memory rather than CUDA.
AI Infra2023
Provides a NumPy-like array framework for building and training ML on Apple silicon, with Python, C/C++, and Swift APIs plus PyTorch-style higher-level modules. Features lazy evaluation, composable AD/vectorization, and a unified-memory multi-device model so arrays can be used on CPU and GPU without explicit copies.
AI Leaderboard2023
Runs one-command evaluation of vision-language models across 80+ multimodal benchmarks, handling data download, inference, and metric scoring in a single pass. Supports 220+ LMMs; adding a new model means writing one generate_inner() function.
AI Agent2023
Builds realtime voice AI agents that run as server-side participants in WebRTC rooms — mix STT, LLM, and TTS providers or use one realtime model. Adds semantic turn detection, SIP telephony, multi-agent handoffs, and an LLM-judge test harness.
AI Video2023
Automates uploading and scheduled publishing of videos to major Chinese and international social platforms (Douyin, Bilibili, Xiaohongshu, Kuaishou, WeChat Video Channel, TikTok, etc.). Offers a CLI, platform-specific uploader modules, headless/browser automation and agent-skill integration for scripted cross-posting workflows.
AI Image2023
Provides a diffusion-model studio for image, video, audio-video, editing, LoRA, and full training workflows so many model families share one inference and training framework.
AI Model2023
Estimates and tracks 6D poses of novel objects without per-object fine-tuning — supports both model-based (CAD) and model-free (few reference images) setups. Trained on large-scale synthetic data with a transformer-based architecture and contrastive learning; CVPR 2024 highlight with demos and pretrained weights.
AI Infra2023
Triton kernels and PyTorch layers for linear-attention, state-space, and sparse-attention token mixers (GLA, RWKV, Mamba2, GSA) as drop-in replacements for multihead attention. Runs on NVIDIA, AMD, and Intel GPUs with Hugging Face support.
AI Infra2023
Traces how Transformer LLMs route information from input to output, attributing each block's effect to individual attention heads and feed-forward neurons. Click any edge to see what a head promotes or suppresses in vocabulary space.
AI Audio2023
Builds real-time voice and multimodal AI agents as composable streaming pipelines. Vendor-neutral: swap among 20+ STT, 20+ LLM and 30+ TTS providers over WebRTC or WebSockets, and compose multi-agent systems with handoff and parallel workers.