unsloth
Step 3.7 Flash
Step 3.7 Flash is a 197.0B mixture-of-experts step35 model from unsloth, with about 35.0B parameters active per token. Configured context length is 262,144 tokens. License: apache-2.0. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters197B / 35.0B active
Context256K
Attentiongqa
Licenseapache-2.0
45 layers · 64 KV heads · head dim 128 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 367.0 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 195.2 GB | Hugging Face file bytes |
| Q8_0 | 195.0 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 166.4 GB | Hugging Face file bytes |
| UD-Q6_K | 152.1 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 136.4 GB | Hugging Face file bytes |
| UD-Q5_K_M | 136.4 GB | Hugging Face file bytes |
| UD-Q5_K_S | 128.6 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 113.8 GB | Hugging Face file bytes |
| UD-Q4_K_M | 113.7 GB | Hugging Face file bytes |
| UD-Q4_K_S | 106.3 GB | Hugging Face file bytes |
| UD-IQ4_XS | 88.8 GB | Hugging Face file bytes |
| IQ4_NL | 90.6 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 83.3 GB | Hugging Face file bytes |
| Q3_K_M | 83.1 GB | Hugging Face file bytes |
| UD-IQ3_S | 74.5 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 68.5 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 61.3 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 57.5 GB | Hugging Face file bytes |
| UD-IQ1_M | 52.9 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Won't fit | UD-IQ1_M | — |
| GeForce RTX 3060 12GB | Won't fit | UD-IQ1_M | — |
| GeForce RTX 4060 Ti 16GB | Won't fit | UD-IQ1_M | — |
| GeForce RTX 4090 | Won't fit | UD-IQ1_M | — |
| GeForce RTX 5090 | Won't fit | UD-IQ1_M | — |
| Apple M4 Pro | Won't fit | UD-IQ1_M | — |
| Apple M5 | Won't fit | UD-IQ1_M | — |
| CPU only · DDR5-5600 | Won't fit | UD-IQ1_M | — |
Smallest devices that fit it fully
- NVIDIA H100 SXM 80GB — Tight fit, UD-Q2_K_XL, 77–106 tok/s
- NVIDIA H100 PCIe 80GB — Tight fit, UD-Q2_K_XL, 46–63 tok/s
- NVIDIA A100 80GB SXM — Tight fit, UD-Q2_K_XL, 47–64 tok/s
- NVIDIA A100 80GB PCIe — Tight fit, UD-Q2_K_XL, 45–61 tok/s
- Apple M3 Ultra — Tight fit, UD-Q2_K_XL, 25–35 tok/s
- NVIDIA H200 SXM — Tight fit, UD-Q4_K_XL, 78–106 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (UD-IQ1_M).
llama.cpp
llama-server -hf unsloth/Step-3.7-Flash-GGUF:UD-IQ1_M -c 2048 -ngl 0
No matching name in the Ollama library snapshot.
LM Studio
lms get unsloth/Step-3.7-Flash-GGUF # Then load quant UD-IQ1_M