llamafile is incredible

[gavdog7]

106 tokens per second on llama v1.5 7B, all I needed to do was run the .exe with this flag (from CMD): llava-v1.5-7b-q4.llamafile.exe -ngl 999

It's unbelievable that it's this easy! I'm trying mixtral 8x7B next. I wish one of the mega context window open source models was available as a llamafile! Claude or Gemini. Or ideally foundation models, they'd be fun to explore

Thank you so much for making this tech so approachable for beginners!!!!

[jart]

Thanks! I'm glad you've been enjoying it!

[realyukii]

Is there a way to somehow contribute my compute power/computer resources as gratitude to the community? (that don't require server-side or public IP)

[BenedictHW]

Thanks Justine! I'm using it to help kids with learning Python as well as to use Python with LEGO SPIKE education robotics. Running llava-v1.5 locally on my Thinkpad E14 gen 4 gives around 7 tokens per second. Really grateful to you and to all who made contributions to open science.

[jart]

Thank you for sharing! I'm very happy to hear this.