Lmst

Ex-Microsoft AI team WizardLM resurfaces at Tencent's Hunyuan, dropping their new Hunyuan-TurboS 0416 model, touted to beat Google's Gemma 3 series. Major AI moves in China. #AI #Tencent #WizardLM

Microsoft's WizardLM AI Team: Tencent Calls After Past Drama

You may have *briefly* seen an announcement from Microsoft's WizardLM team about the release of WizardLM v2 LLM.

I say briefly because WizardLM just yanked all v2 models & repositories from Huggingface & Github.

An announcement from the team on Twitter/X stated that it was temporarily withdrawn because it hadn't yet passed a mandatory "Red Teaming & Safety Assurance" validation process - a new requirement since their last release.

However, it's important to note...

#LLMs #WizardLM #Microsoft

Ich sammle gerade #opensource #LLMs, mit denen man gut auf Deutsch kommunizieren kann👀

Mit Phi1.5 oder Finetunings von OpenLlama(v2) & MBT klappt das bislang eher nicht.

Gute Erfahrungen habe ich v.a. mit #Llama2-Finetunings wie #WizardLM & german-assistant-v4 gemacht (in zweiter Linie auch mit Tulpar und Openorca-Platypus2).

Kennt ihr weitere Modelle, die man sich in diesem Zusammenhang mal ansehen sollte?

I’m totally blown away by #Llama2 and most of its fine-tunings,
but why did Meta itself decide to build a chat model and put use in languages other than English out of scope?
Unlike #WizardLM, this model keeps switching to English even if prompted in German 😟

https://huggingface.co/meta-llama/Llama-2-70b-chat-hf