A Developer Compiled a Full RISC-V CPU Into Qwen3.5's Weights
A Qwen3.5-4B variant compiles a 32-bit RISC-V processor directly into its weights, turning Gated DeltaNet layers into addressable RAM.
20 stories
A Qwen3.5-4B variant compiles a 32-bit RISC-V processor directly into its weights, turning Gated DeltaNet layers into addressable RAM.
A 2.8GB distilled reasoning model gets stripped of safety refusals through automated abliteration, running locally on consumer hardware with near-zero refusal rate.
A PhoBERT-based Vietnamese sentence embedding model from BKAI is quietly powering RAG and semantic search apps with over 800K downloads.
A LoRA plus modified VAE for the Anima model that generates clean RGBA images with a real alpha channel, no background removal needed.
A community-tuned drafter for Qwen3.8-27B's NVFP4 build pushes DGX Spark decode from 12.5 to 48.2 tokens per second, bit-exact lossless.
Speridlabs' open image-to-3D model Idol lands in ComfyUI, turning one picture into a million-Gaussian splat on 7.2 GB of VRAM.
A Japanese ASR model from Third Intelligence hits 6.12% CER at roughly 1/70th the cost of GPT-Transcribe, matching commercial APIs.
OpenGVLab released InternVL2-1B, a sub-billion-parameter vision language model pairing InternViT-300M with Qwen2-0.5B for edge-friendly multimodal reasoning.
NVIDIA shipped an NVFP4 quantized DeepSeek V4.1 Flash that trims 75 GB off disk while matching the original on six benchmarks.
A refusal-stripped build of Qwen3.5-4B shrinks to 2.7GB, keeps benchmark scores intact, and runs offline on an 8GB laptop.
A tuned imatrix GGUF repack of Spark-X2.5-4B lands a 4B coder on 6 GB GPUs with usable speed at 131k context.
Amazon's 21M-parameter time series foundation model delivers zero-shot forecasts up to 250x faster than original Chronos while beating specialized baselines.
Liquid AI's LFM3 family packs text, vision, audio, omni, and decision models into six 2B-class checkpoints built from one tri-hybrid backbone.
Sperid Labs released Idol, an Apache 2.0 image-to-3D model that generates up to 1M Gaussian splats in seconds on consumer GPUs.
Pika Labs just dropped open LoRA adapters on top of MiniMax-H3 that generate video and synchronized audio in only six sampling steps.
A 9B reasoning model distilled from DeepSeek-V4-Pro and decensored with Heretic now ships as imatrix GGUFs that run on 4GB GPUs.
A crowd-sourced ternary quantization of Qwen3.8-27B runs 27B parameters at 25 tokens per second on a base M5 MacBook in 13 GB of RAM.
A community Windows desktop app packages Boson AI's Higgs Audio v3 TTS into GGUF quantizations with a local C++/CUDA engine for offline voice cloning.
AutoTrust's open 9B model packs calibrated System 1 decisions and full System 2 reasoning into one Qwen3.5 backbone, matching a closed teacher at a third the latency.
Maya Research dropped open weights for a streaming TTS model covering 11 Indian languages plus English, with 20 voices and zero-shot cloning under Apache-2.0.