Can I run Swift-1.5-Qwen3.8-Flash-Next-GGUF on an NVIDIA H100?
Yes, just
Its IQ1_S file (70.1 GB) fits in the 72.0 GB an NVIDIA H100 can use, with room for ≈ 71K tokens of context. That file is under 3 bits per weight, so expect noticeably lower quality.
Full analysis of Swift-1.5-Qwen3.8-Flash-Next-GGUF
Everything that runs on NVIDIA H100
Get the IQ1_S file on Hugging Face
llama-server -hf ukisai/Swift-1.5-Qwen3.8-Flash-Next-GGUF:IQ1_SFiles and context on NVIDIA H100
Each file of Swift-1.5-Qwen3.8-Flash-Next-GGUF (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) |
|---|---|---|---|---|
| Q8_0 | 188 GB | Doesn't fit | — | — |
| Q6_K | 168 GB | Doesn't fit | — | — |
| Q5_K_L | 151 GB | Doesn't fit | — | — |
| Q4_K_L | 139 GB | Doesn't fit | — | — |
| Q5_K_M | 135 GB | Doesn't fit | — | — |
| Q5_K_S | 128 GB | Doesn't fit | — | — |
| Q4_K_M | 120 GB | Doesn't fit | — | — |
| Q3_K_XL | 119 GB | Doesn't fit | — | — |
| Q4_K_S | 113 GB | Doesn't fit | — | — |
| Q4_1 | 111 GB | Doesn't fit | — | — |
| Q2_K_L | 107 GB | Doesn't fit | — | — |
| Q4_0 | 101 GB | Doesn't fit | — | — |
| IQ4_NL | 100 GB | Doesn't fit | — | — |
| IQ4_XS | 97.7 GB | Doesn't fit | — | — |
| IQ3_M | 93.2 GB | Doesn't fit | — | — |
| Q3_K_L | 93.2 GB | Doesn't fit | — | — |
| Q3_K_M | 92.0 GB | Doesn't fit | — | — |
| IQ3_XS | 91.9 GB | Doesn't fit | — | — |
| Q3_K_S | 89.4 GB | Doesn't fit | — | — |
| IQ3_XXS | 88.0 GB | Doesn't fit | — | — |
| Q2_K | 80.9 GB | Doesn't fit | — | — |
| IQ2_M | 80.4 GB | Doesn't fit | — | — |
| IQ2_S | 77.8 GB | Doesn't fit | — | — |
| IQ2_XS | 77.7 GB | Doesn't fit | — | — |
| IQ2_XXS | 75.2 GB | Doesn't fit | — | — |
| IQ1_M | 72.0 GB | Doesn't fit | — | — |
| IQ1_S | 70.1 GB | Tight | 71K | 141K |