florianleuerer

AI and Machine Learning Enthusiast | Deep Learning | Large Language Models 🧠

[florianleuerer @florianleuerer]

@BenjaminDEKR @JackWoth98 I‘m pasting multiline text all the time and never had an issue once?

[florianleuerer @florianleuerer]

@OfficialLoganK @OfficialLoganK make 2.5 GA and bring it to Vertex AI EU regions PPLLLEEEAAASSSEEE 🙏

[florianleuerer @florianleuerer]

@OfficialLoganK PLEASE just make a Model better than Gemini 1.5 Pro GA on vertex AI (no 2.0 Flash is not better for long context tasks).

We are stuck on a 6 month old model …

[florianleuerer @florianleuerer]

@OfficialLoganK Do we still get Gemini 2.0 Pro GA some time or even better 2.5 Pro GA soon?

[florianleuerer @florianleuerer]

@jxnlco So it is basically still RAG but with predefined questions?

[florianleuerer @florianleuerer]

@yoheinakajima Have you ever heard of ABAP?

[florianleuerer @florianleuerer]

I think @mattshumer_ is quite happy about the o1 release 😉

[florianleuerer @florianleuerer]

@airesearch12 Or was he tricked by @csahil28? Matt didn‘t know what LORA was and in one of his early posts he mentioned, that benchmarks were done by sahil?

[florianleuerer @florianleuerer]

@TheXeophon Oh no the weights got mixed up again! 🙃

[florianleuerer @florianleuerer]

@ArtificialAnlys @mattshumer_ How do you run the MMLU benchmarks? The default implementation is logprob and not „generation“. I guess you run it as a generation benchmark and prompt the models to genrate the letter of the correct answer?

[florianleuerer @florianleuerer]

@cognitivecompai @FernandoNetoAi @maximelabonne I wonder how that would work for eg MMLU with lm-eval-harness?! MMLU is a logprob Benchmark which does not actually generate the response / answer? In my opinion there is no way to benchmark the Reflection model on MMLU and compare the results?!

[florianleuerer @florianleuerer]

@itsSandraKublik And its finally available in Azure ❤️

[florianleuerer @florianleuerer]

@Nils_Reimers @aidangomez @cohere I've been doing that regularly since the Cohere announcement, obviously without success so far 😐

[florianleuerer @florianleuerer]

@HamelHusain It‘s not a course it‘s a movement!

[florianleuerer @florianleuerer]

@HamelHusain Finally no longer GPU poor!

[florianleuerer @florianleuerer]

@HamelHusain Just delete the huggingface cache 🤗

[florianleuerer @florianleuerer]

@seb_ruder Will Cohere Rerank also be available on Azure?

[florianleuerer @florianleuerer]

@TheZachMueller And it‘s actually pretty easy 😄 the only problem is the auto-update that might break things when installing the drivers that way.

[florianleuerer @florianleuerer]

@gblazex @SebastianB929 Bard (Gemini Ultra)?