Pinned Loading

1. pulsar pulsar Public

   SSD-streaming inference engine for giant MoE models (Rust + CUDA). GLM 5.2 743B at 2 tok/s and Hy3 295B at 7 tok/s on two consumer 16GB GPUs. Zero-config multi-GPU: measures PCIe bandwidth, places …

   Rust 205 25

2. neutronstar neutronstar Public

   Forked from antirez/ds4

   Giant MoE models on a single consumer GPU by streaming experts from SSD. CUDA fork of antirez/ds4: runs GLM-5.2 (743B), Tencent Hy3 (295B), and DeepSeek 4 Flash, with io_uring expert streaming, LFU…

   C 40 3

3. kokoro-tts-mcp kokoro-tts-mcp Public

   MCP server for Kokoro text-to-speech, with adjustable voices/speed and an optional OpenAI-compatible (kokoro-fastapi) backend.

   Python 10 4

4. agora agora Public

   Rust

5. encoder-free-speech encoder-free-speech Public

   Conversational speech generation on a frozen, encoder-free multimodal LLM: QLoRA adapters + CSM-style speech head on a single 16GB GPU. Paper + full pipeline.

   Python

6. protostar protostar Public

   134M LLM trained from scratch on two consumer GPUs + accretion: replay-gated memory retaining 77% of learned facts across sequential consolidations (vs 12.5% without)

   Python