Phi-3.5 (MoE, Mini & Vision) : The NEW BEST Small Model is finally here! (Beats Llama-3.1, Mistral)

[@jackflash6377]

Seems like we can't go 1 week without something spectacular being introduced. What a time to be alive. Thanks for the great breakdown and testing. The MOE is quite impressive. Ollama needs to hurry and make that compatible.

[@ym-xx6kj]

2nd! lets GO, can't wait for things to advance more

[@markantscott]

excellent review, thanks

[@detectivelon8971]

Great video thanks for the explanations.

[@jimlynch9390]

Good test, thanks.

[@hansaireal]

NICE MAN

[@vlee9275]

I don't know why they are comparing Phi-3.5 to smaller models though... They should use some medium models like gemma 2 27b. The architecture is different and uses only 6.6b active parameters at a time but it's still taking a substantial amount of memory.

[@daryladhityahenry]

Hi AICodeKing!! Can you take a look at Falcon Mamba? I already wait for model like that for a looong time since mamba news.. But right now, I don't know how to run it. But it seems pretty cool, I really want to know if possible :D:D. And if it's really working well, I don't think transformers model can compete with mamba because of context length.

[@SansaWood-s7e]

That helps a lot! Thank you. I would like to have a little talk with you, and if it is possible for me to reach out?

[@QHawk7]

Can I have like my local files and environment on my local PC , but use colab, lightningAI or kaggle or other services GPU via terminal? I use Linux for AI stuff . say to make a chatgpt style chatbot with falcon 7b/40b model.

[@MeinDeutschkurs]

These models are strange.I cannot say anything. Have to wait for ollama support.

[@patrickchristianmagtaan5511]

🎉😮

[@dung5990]

where to talk to that model?

[@Aarifshah-A]

Great 👍

[@aivy-aigeneratedmusic6370]

Your questions are almost only math and coding. Maybe mix it up more with new questions ?

[@crushfire2004]

maybe need some creative test like writing poem or fact checking test?

[@techblock7938]

Available with ollama

[@fareastpanda1337]

i can't find the moe. only the mini. do you have the link?

[@techblock7938]

@@fareastpanda1337 ollsma provides currently only the mini modell. On huggingface you can get the moe.

[@dung5990]

2nd=))

[@hazemkhaled2714]

1st