A TTS model capable of generating ultra-realistic dialogue in one pass.

TTS model capable of streaming conversational audio in realtime.

[ICLR 2025] Official Repository for "Tamper-Resistant Safeguards for Open-Weight LLMs"

What does gpt-oss tell us about OpenAI's training data?

Stuart is simple RAG System, that the Open Data Hub uses as a chatbot to help the Open Data Hub customer care team in solving tickets. Stuart uses as input: the Open Data Hub Wiki, the past tickets history, the readme files and the GitHub issues of all NOI-Techpark repositories.

NVFP4 LLM inference tuned for consumer Blackwell GPUs. Auto-detects your GPU, downloads the model, serves an OpenAI-compatible API — benchmarked and quality-certified.

FastMCP 3.2 server plus Webapp for using the leaked Claude Code locally with ollama and open weight models. Kairos and cat included

No Hopper architecture (RTX 5090, etc.) required! <16 GB VRAM, Windows.

Demonstrator for an open-source speech-to-summary workflow

Open Weight Definition (OWD)

Tracking open-weight LLMs for research, experiments, and inference comparisons.

Autonomous, high-throughput LLM Gateway & Budget Circuit Breaker built for Edge. Prevents cost loops & token spikes for GPT-5.5, Claude Fable 5 & open weights with atomic PID backpressure, Kalman filters & Verifiable Savings Ledger. Wire-compatible with OpenAI/Anthropic SDKs.

Production-ready benchmark scoring conversations on 300+ facets (scalable to 5000+) using open-weights LLMs ≤16B. With per-score confidence, 4 backends, FastAPI + Streamlit UI, and Docker.

Open source claim-to-reference "verification" tool using R, API (Ollama as is) and open weight LLMs. More automatically verify claims to the best of LLMs capability, given available text to verify and corresponding pdf of references.

Notes & comparisons on storage options for large LLMs (Work-in-progress)

See what a prompt costs across 24 models and 5 labs (OpenAI, Anthropic, Google, xAI, open-weight), cheapest first, with each model's limitations next to the price.

Use Cline's OpenAI-compatible models in Copilot Chat. ClinePass: DeepSeek V4, MiMo V2.5, Qwen3.7, Kimi K2.7, GLM 5.2, MiniMax M3. One API key, $9.99/mo, 2-5x rate limits. Bring your own key to native GitHub Copilot Chat.

Notes & experiments on LLMs, open-weight models, multimodal systems, and cloud deployment.

Audit local LLM function calling and agentic reliability. Visual tool-use benchmarking for quantized models on YOUR hardware.

On-prem AI coding agent + LLM cost-router for regulated enterprises — 100% AMD, open-weight, air-gapped, Ed25519 tamper-evident audit. AMD ACT I 1st-place winner.