Factorial Capital is a venture capital firm that specializes in early-stage investments in technology companies. The firm focuses on backing founders who are building tools and infrastructure that will empower the next generation of creators and companies.

Factorial Capital

Factorial

Excited to share that Flower Labs  has raised $20m from Felicis led by Niki Pezeshki to expand their federated machine learning platform. https://lnkd.in/gx83u5WW

Playing with a basic, fully-local and open-source speech-to-LLM-to-speech pipeline on my mac

Less than 120 lines of code to chain local whisper + Zephyr (in LM studio) + an Openvoice TTS: https://lnkd.in/enbK2gfd

End to end latency is 1.5-2.5 sec on my M3. Aalready quite impressed how all these local models run with such a low latency even without specific optimizations.

This 2 s of latency is my baseline. Now excited to see what a more optimized fully local OSS assistant could reach, looking at you Candle, LAION BUD-E, metavoice, Openvoice, suno, Llama.cpp :)

Big news 🚨 Today, I’m excited to announce our partnership with Hugging Face on Enterprise Scenarios Leaderboard, the first LLM leaderboard for real world use cases! 

After lots of conversations with enterprise leaders, we learned how hard it is to evaluate, benchmark, and choose LLMs for real world scenarios. For example, how do we know Llama-2 is better than Mistral-7B on answering financial questions? It’s even harder to pick the right LLM when new LLMs are released every week…

Enter: the Enterprise Scenarios Leaderboard 🚀

The leaderboard ranks open source LLMs on 6 enterprise use cases: finance, legal, customer support dialogue, enterprise PII, toxicity, and creative writing. We curated each of these evaluation datasets by hand, and some of these went viral in the past few months 🔥 Example: FinanceBench, which was covered by CNBC and Fortune in December!

Here’s the best part: it’s not easy to game this leaderboard! 😎 We kept some of the evaluation datasets closed source. This means LLM developers can’t just train on these evaluation datasets to artificially get their LLMs ranked at the top. Say no to test set leakage 🙅🚫

It was awesome collaborating with Clémentine Fourrier from Hugging Face on this! And big thanks to Selvan Sunitha Ravi from Patronus AI for leading this ✨ 

The best is yet to come 🚀

The Enterprise Scenarios Leaderboard: https://lnkd.in/ghXxQfhB
Read Hugging Face's blog post here: https://lnkd.in/ge5SYkSZ

COM:FACTORIALCAP

Factorial Capital