# Synthetic Minds > Synthetic Minds is an open-source AI research and engineering lab in London, United Kingdom. We build tools to train, fine-tune and run open models on Tenstorrent Blackhole and Galaxy Neural Accelerators. Our first project is Infall-TT (planned Python package name: infall-tt), an open-source framework for quantization, LoRA fine-tuning and inference on Tenstorrent hardware, currently in development. Key facts: - Legal entity: SYNTHETIC MINDS LTD, company number 17593682, London, United Kingdom. - Founder and CEO: Tudor Iustin. - Infall-TT is in developer preview and is NOT yet available to the public: the infall-tt package is not on PyPI and the source code is not yet published on GitHub. Its waitlist is open. Code and results shown in announcements are examples and may change before release. - Tenstorrent hardware and software power our infrastructure and frameworks. We are not affiliated with Tenstorrent beyond building on its open software stack. - We publish no speed or throughput numbers until measured on real silicon. Accuracy results so far come from Tenstorrent's bit-exact Blackhole simulator (ttsim). - The core of what we build is open source under the Apache 2.0 license. - Contact: hello@syntheticmindslab.com (general, sponsorships, careers, collaboration). ## Main pages - [Home](https://syntheticmindslab.com/): Mission, why we started, why we chose Tenstorrent, how we work. - [Infall-TT](https://syntheticmindslab.com/infall-tt): Waitlist for the Infall-TT beta. - [Introducing Infall-TT](https://syntheticmindslab.com/news/introducing-infall-tt): Announcement with highlights, quantization accuracy charts and example code. - [Why we started Synthetic Minds](https://syntheticmindslab.com/news/why-synthetic-minds): The case for moving from GPUs to Neural Accelerators. - [Models](https://syntheticmindslab.com/models): Catalog of open models quantized by Synthetic Minds. - [Qwen3.8-27B-BFP4-Infall-TT](https://syntheticmindslab.com/models/qwen3-8-27b-bfp4-infall-tt): Tenstorrent-native BFP4 (bfloat4_b) quantization of Qwen3.8-27B made with Infall-TT (GPTQ on the BFP4 grid): +4.0% WikiText-2 perplexity vs +11.0% for Tenstorrent's default rounding, about 15 GB in BFP4, for tt-metal on Blackhole. Not yet run on hardware; no speed claims. Published on Hugging Face. - [Inkling-2-Mini-Infall_TT GGUF](https://syntheticmindslab.com/models/inkling-2-mini-infall-tt-gguf): 7 GGUF quantizations (18.7 GB to 27.8 GB) of Thinking Machines Lab's Inkling-2-Mini (26.4B total, 3.4B active, Apache 2.0) with KL divergence, top-1 agreement and perplexity versus Q8_0, plus llama.cpp and Ollama commands. Published on Hugging Face. - [News](https://syntheticmindslab.com/news): All announcements. - [About](https://syntheticmindslab.com/about): Company and founder. - [Careers](https://syntheticmindslab.com/careers): Values, how we work and what we offer. - [Open roles](https://syntheticmindslab.com/careers/search): Founding AI Engineer, Founding Developer Advocate, Commercial GTM Lead, Marketing Manager. ## Elsewhere - [GitHub](https://github.com/syntheticminds-lab) - [Hugging Face](https://huggingface.co/synthetic-minds) - [X](https://x.com/syntheticminds2) - [LinkedIn](https://www.linkedin.com/company/synthetic-minds-lab/) ## Optional - [Full text of this site for language models](https://syntheticmindslab.com/llms-full.txt) - [Sitemap](https://syntheticmindslab.com/sitemap.xml)