open-weight-models

Easily fine-tune, evaluate and deploy Qwen, Gemma, or any open weight LLM!

The Best TypeScript framework for fine-tuning

Curated list of open-weight AI models with commercially exploitable licenses, verified benchmarks, and no EU restrictions.

주제별 핵심 개념 이해부터 직접 실행까지 다루는 가이드·튜토리얼·실습 모음

Fine-tune small open models into reliable JSON tool callers. Reproducible, end to end, on a single GPU.

“A clean, from-scratch implementation of the OLMo architecture with KV caching, RoPE, and an efficient autoregressive inference pipeline. Designed as a minimal yet extensible foundation for post-training research, including RLHF, preference optimization, and reasoning-focused systems.”

Benchmark how well any model performs real-world, long-horizon coding tasks across coding agents (harnesses) -- Claude Code and pi. Run models three ways -- Anthropic on Amazon Bedrock, open-weight on Bedrock via a LiteLLM proxy, or self-hosted on EC2 with vLLM -- score them with an LLM judge, and plot the cost/quality Pareto frontier.

Benchmarking open-weight LLM coding agents as SCOUT delegates: model comparison experiments with pre-registered protocols, blind scoring, and full data.

An independent, evidence-graded read on how much you own the open models and inference providers you rely on.

Frontier-Grade Open Weights ── フロンティア級のオープンウェイトモデルは、開かれたのか。Kimi K3 / GLM-5.3 を継続追跡する定点観測型OSS書籍 / They matched the frontier. But no one can hold them. Continuously updated.

Tamper Attack Resistance (BlueDot Impact Technical AI Safety Project)

Policy-enforcing inference for open-weight LLMs. Self-hosted and OpenAI-compatible: enforces your policy at the residual-stream level, so it survives obfuscation, role-play and jailbreak wrappers. Byte-for-byte no-op on in-policy requests.

Local AI lab for llama.cpp model routing, workload aliases, benchmarking, and evaluation through one OpenAI-compatible gateway.

Run a POC on any open-weight model and validate your GPU needs in one session. Sizes VRAM, provisions the right instance, benchmarks it on vLLM, and tears everything down so you know what fits, how fast it runs, and what it costs before you commit.

Klepon: Taste of Indonesia

Anthropic-style emotion-vector geometry, on any open-weight LLM, in one command. Frozen corpus + unified pipeline + statistical rigor + 5-model reference results.

Experimental framework for cross-subtask malicious intent detection in stateless AI agents — research into open-weight model safety

OpenGauntlet — independent conversational-EQ rankings and sourced voice-AI research for locally hosted models

Companion code for "Clinical Text De-identification with Locally Deployed Open-Weight Large Language Models: A Real-World Evaluation of Discharge Notes". Submitting to Frontiers in Digital Health.

Workspace-lens audit cards and training-workflow scaffolds for open-weight language models.