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

Can I run clef-flash on an RTX 5070?

Yes, with a GGUF version

The published weights are too big, but the Q8_0 file from bartowski/Cloudflare_clef-flash-GGUF (9.68 GB) fits, with room for ≈ 32K tokens of context.

Full analysis of clef-flash Everything that runs on RTX 5070 Get the Q8_0 file on Hugging Face
ollama run hf.co/bartowski/Cloudflare_clef-flash-GGUF:Q8_0

Files and context on RTX 5070

Each file of clef-flash (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 5070 Context (16-bit)Context (8-bit)
As published18.8 GBDoesn't fit——
BF16ggml-org/Clef-Flash-GGUF18.2 GBDoesn't fit——
Q8_0ggml-org/Clef-Flash-GGUF9.66 GBTight33K65K
Q4_K_Mggml-org/Clef-Flash-GGUF6.49 GBFits127K254K
BF16bartowski/Cloudflare_clef-flash-GGUF18.2 GBDoesn't fit——
Q8_0bartowski/Cloudflare_clef-flash-GGUF9.68 GBTight32K64K
Q6_K_Lbartowski/Cloudflare_clef-flash-GGUF8.32 GBFits72K145K
Q6_Kbartowski/Cloudflare_clef-flash-GGUF7.83 GBFits87K174K
Q5_K_Mbartowski/Cloudflare_clef-flash-GGUF6.98 GBFits112K224K
Q4_K_Lbartowski/Cloudflare_clef-flash-GGUF6.80 GBFits118K236K
Q5_K_Sbartowski/Cloudflare_clef-flash-GGUF6.66 GBFits122K244K
Q4_1bartowski/Cloudflare_clef-flash-GGUF6.07 GBFits139K256K
Q4_K_Mbartowski/Cloudflare_clef-flash-GGUF6.04 GBFits140K256K
Q4_K_Sbartowski/Cloudflare_clef-flash-GGUF5.73 GBFits150K256K
IQ4_NLbartowski/Cloudflare_clef-flash-GGUF5.63 GBFits152K256K
Q4_0bartowski/Cloudflare_clef-flash-GGUF5.60 GBFits153K256K
IQ4_XSbartowski/Cloudflare_clef-flash-GGUF5.40 GBFits159K256K
Q3_K_Lbartowski/Cloudflare_clef-flash-GGUF5.24 GBFits164K256K
Q3_K_Mbartowski/Cloudflare_clef-flash-GGUF5.05 GBFits170K256K
IQ3_Mbartowski/Cloudflare_clef-flash-GGUF4.85 GBFits176K256K
Q3_K_Sbartowski/Cloudflare_clef-flash-GGUF4.80 GBFits177K256K
Q2_Kbartowski/Cloudflare_clef-flash-GGUF4.19 GBFits195K256K

clef-flash on other hardware

Popular models that run on RTX 5070

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.