🔥 phi 4 official released

Local AI
1

a few others:

3 Likes
2

Awesome to see a quantized model released. I’ll have to test the 4 but version later… But I’m really waiting for are smaller and quantized models of DeepSeek v3.

2 Likes
3

bullerwins/DeepSeek-V3-GGUF
Q3, Q4, Q5, Q8, BF16

2 Likes
4

Awesome but damn, like 400GB even with 4bit quantization. No way I’ll be running that but still pretty cool. Might make the model more available and keep the cost down.

And I hadn’t even look in the last few days but these all seem pretty recent.

Thanks.

2 Likes
5

Yeah DeepSeek V3 is just too big haha

I wish they had a 32b/70b parameter version of it or something

6

Thanks for sharing Max! Looks awesome

7

Blockquote
Yeah DeepSeek V3 is just too big haha
I wish they had a 32b/70b parameter version of it or something

same here

the biggest problem imo is that LLMs are trained on too many programming languages, there should be a deepseek-v3-typescript, deepseek-v3-python, …
because i’m quite certain then it should be possible to get to a 70b or even 32b parameter size

3 Likes
8

Great opportunity for a game changer fine tuning project… :smiley: