Model Anatomy See what's inside open-weight AI models

Can I run Kolibri-1 on an RTX 5090?

Yes, with a GGUF version

The published weights are too big, but the Q2_K file from Hob-forge/Kolibri-1-GGUF (28.6 GB) fits, with room for ≈ 6K tokens of context. That file is under 3 bits per weight, so expect noticeably lower quality.

Full analysis of Kolibri-1 Everything that runs on RTX 5090 Get the Q2_K file on Hugging Face
ollama run hf.co/Hob-forge/Kolibri-1-GGUF:Q2_K

Files and context on RTX 5090

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.

FileSizeOn RTX 5090 Context (16-bit)Context (8-bit)
As published78.8 GBDoesn't fit——
Q8_0Hob-forge/Kolibri-1-GGUF83.1 GBDoesn't fit——
Q6_KHob-forge/Kolibri-1-GGUF64.2 GBDoesn't fit——
Q5_K_MHob-forge/Kolibri-1-GGUF55.6 GBDoesn't fit——
Q4_K_MHob-forge/Kolibri-1-GGUF47.5 GBDoesn't fit——
Q3_K_MHob-forge/Kolibri-1-GGUF37.5 GBDoesn't fit——
Q2_KHob-forge/Kolibri-1-GGUF28.6 GBTight6K14K

Kolibri-1 on other hardware

Popular models that run on RTX 5090

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.