Can I run Kolibri-1 on an NVIDIA H100?

Yes, with a GGUF version

The published weights are too big, but the Q6_K file from Hob-forge/Kolibri-1-GGUF (64.2 GB) fits, with room for ≈ 256K tokens of context.

Full analysis of Kolibri-1 Everything that runs on NVIDIA H100 Get the Q6_K file on Hugging Face

llama-server -hf Hob-forge/Kolibri-1-GGUF:Q6_K

Files and context on NVIDIA H100

Each file of Kolibri-1 (and of GGUF versions the site has analyzed), and how much context fits next to it, with the context cache in 16-bit or 8-bit.

File	Size	On NVIDIA H100	Context (16-bit)	Context (8-bit)
As published	78.8 GB	Doesn't fit	—	—
Q8_0Hob-forge/Kolibri-1-GGUF	83.1 GB	Doesn't fit	—	—
Q6_KHob-forge/Kolibri-1-GGUF	64.2 GB	Tight	256K	256K
Q5_K_MHob-forge/Kolibri-1-GGUF	55.6 GB	Fits	256K	256K
Q4_K_MHob-forge/Kolibri-1-GGUF	47.5 GB	Fits	256K	256K
Q3_K_MHob-forge/Kolibri-1-GGUF	37.5 GB	Fits	256K	256K
Q2_KHob-forge/Kolibri-1-GGUF	28.6 GB	Fits	256K	256K

Kolibri-1 on other hardware

Popular models that run on NVIDIA H100

Assumes 90% of the card's memory is usable and at least 4K tokens of context for one conversation, with the context cache in 16-bit unless noted. Fits leaves at least 20% of that memory for context; Tight leaves less.