潜力新星（90天）

Run DeepSeek-V4-Flash 304B on a single NVIDIA GB10 / DGX Spark - 2-bit MoE planes, FP4 quality recovery, speculative decoding

Move files between devices with a screen and a camera — offline optical transfer of RVF cognitive containers and WASM artifacts over animated QR codes. Mobile-first, no network, no accounts, SHA-256 verified.

MCP server that gives text-only LLM coding agents vision — analyze images via any multimodal model (mimo, Claude, Gemini, OpenAI-compatible). Works with Claude Code, Codex, Kimi, opencode, PI.

DeepSeek Harness 插件：DeepSeek Pro 大脑 + 自动识图。GUI 附加图片默认经官方 deepseek-v4-flash-vision-exp 原生识图，转译成文字后交给 DeepSeek 作答（纯文本的 V4-Pro 也能看图）；支持百炼/智谱/OpenRouter 等任意 OpenAI 兼容 VLM，无 key 自动探测本地 Ollama；安装时有一问式确认

Open-source local AI server configs, GFX906 runtime maintenance, reproducible benchmarks, and QC methods for affordable AI research infrastructure.

Indonesian news summarization with IndoBART / mT5 (headline + paragraph summary).

🧪 SynthForge — toolkit for generating high-quality synthetic datasets for LLM fine-tuning. Multi-backend, filterable, GPU-accelerated. Early-stage experimental project.

基于Qwen2.5-VL-3B + QLoRA的智能驾驶场景结构化理解系统

NeuroLens is a premium, state-of-the-art Document Retrieval-Augmented Generation (RAG) system. It combines an immersive, holographic, sci-fi-themed React frontend with a high-performance FastAPI backend.

An out-of-tree vLLM plugin for Mobilint NPU runtime integration.

Context Runtime — a database query planner for LLM context. Decides what a model sees before it answers; plans it, runs it through reused substrate, and learns from the outcome.

Enterprise multi-source RAG platform with multimodal embedding, hybrid vector/BM25 retrieval, MCP & RBAC support

GLM-5.2 on 4x DGX Spark with adaptive MTP K2/K4/K5, FULL CUDA graphs, DCP2, 520K context, and a downloadable ARM64 runtime image.

Detect and resolve missing models in ComfyUI workflows — active-graph analysis, multi-source resolution (Civitai / Hugging Face), and checksum-verified downloads.

DeepSeek-V4-Flash-0731 tuned on 4x DGX Spark (GB10) at TP=4 — 123.13 tok/s decode. Every variable measured one at a time, with the values that lost.

Run Ornith, Qwen, Nemotron & DeepSeek optimized on AMD Strix Halo — unified OpenAI-compatible server with tuned ROCmFP4 quants, validated 262K context, vision support & hot-swap model zoo (iGPU + NPU)

Extend LLM context windows beyond GPU memory limits with disk-backed KV cache.

Tracking exactly what happens to the internal "circuitry" (induction heads) of a 2-layer attention-only Transformer when forced to undergo domain adaptation from prose to structured Python code.

Kubernetes-native control plane for scale-to-zero serving of long-tail LLMs

The holographic core for leOS