Kasra Rahjerdi
@jc4p
llama 4 advertised context size: 10 mil tokens llama 4 hosted on groq context size: 131,072 tokens llama 4 hosted on together context size: "you figure it out :)"
2 replies
0 recast
15 reactions
zoo :?
@zoo
what do you think vs gemini?
1 reply
0 recast
1 reaction
Kasra Rahjerdi
@jc4p
i shit you not from my website
1 reply
0 recast
1 reaction
Kasra Rahjerdi
@jc4p
i need to spin up int-4 llama on a single H100 like they brag about and see what max context window i can pull off, otherwise i can test on a 8xH100 at higher quants, or i could test on a GH200 if i do a PR to the quantization library to make it support Grace Hopper
1 reply
0 recast
1 reaction
zoo :?
@zoo
hadnt actually read your site before. gud cv/bio
0 reply
0 recast
1 reaction