Pinned Loading

1. llm2vec-8gb-offload-lossless-batching llm2vec-8gb-offload-lossless-batching Public

   Run LLM2Vec (Llama-3-8B) on an 8 GB GPU via RAM offload, and batch it without changing the embeddings (padding shifts them: cosine 0.76 vs 0.999956)

   Python

2. vergilius vergilius Public

   Assistente AI locale per capire cosa succede nel mondo, ovunque: modelli piccoli, macchine modeste, nessun fornitore esterno. Odysseus + ShadowBroker.

   Python

3. blackwell6000-qwen3.8-flash-next blackwell6000-qwen3.8-flash-next Public

   Blackwell 6000 + qwen 3.8 flash next: Swift 1.5 Qwen3.8 Flash-Next NVFP4 on one RTX PRO 6000 with vLLM, 4 users x 227k context, ~100 tok/s each, exact kernels + determinism fix

   Cuda

4. bonsai-webgpu-tiled-vae-fix bonsai-webgpu-tiled-vae-fix Public

   Tiled VAE fix: run PrismML's Bonsai-Image 4B WebGPU at 512px+ on GPUs without shader-f16 (GTX 10xx / Pascal & older). Root cause analysis + 1.4KB patch, no quality loss on f16 devices.

   JavaScript 1

5. bonsai2-ternary-pascal bonsai2-ternary-pascal Public

   Ternary Bonsai 2 9B on a GTX 1080 (sm_61): +51% decode with bit-identical output (PrismML llama.cpp fork, PQ2_0 + MTP)

   Python 1

6. Partido-De-What/claude-plugins Partido-De-What/claude-plugins Public

   Claude Code and Codex plugins by ArtyomITA - chatroom: put your agents in one room and watch them work (live group chat, task board, 3D voxel office)

   JavaScript