How much memory do I need to run Step-3.7-Flash (UD-Q4_K_M)?

The file size is 122 GB (measured). With a 8,192-token context, the estimated memory including the KV cache is 123 GB. The formula is published at https://localllm.az-arte.com/en/guide/memory-calculation.

Does Step-3.7-Flash run in 8 GB, 12 GB, 16 GB or 24 GB of VRAM?

8 GB: no quantization fits; 12 GB: no quantization fits; 16 GB: no quantization fits; 24 GB: no quantization fits (estimated at 8,192 tokens with an f16 KV cache, files from the most-downloaded source).

What is the command to run Step-3.7-Flash with Ollama?

Run "ollama run hf.co/unsloth/Step-3.7-Flash-GGUF:UD-Q4_K_M" to use the GGUF on Hugging Face directly. To use another quantization, replace the name at the end.

stepfun-ai

Step-3.7-Flash

Image input

Parameters: 197B
Max context length: 262,144
Architecture: step35
License: Commercial OK

Base model: stepfun-ai/Step-3.7-Flash
License: apache-2.0
llama.cpp: b7964+

Memory needed to run UD-Q4_K_M at 8K tokens

123 GB Estimated

File size 122 GB Measured

VRAM it fits in (rough guide)

8 GB does not fit
12 GB does not fit
16 GB does not fit
24 GB does not fit
32 GB does not fit
48 GB does not fit

See devices and models for 128 GB

The smallest, UD-IQ1_M, needs an estimated 58.1 GB at 8K.

Macs cap how much memory the GPU can use, so leave headroom in unified memory.

Data as of Oct 21, 2026

About Step-3.7-Flash

Step-3.7-Flash is a GGUF model based on stepfun-ai/Step-3.7-Flash on Hugging Face. GGUF files are published by unsloth/Step-3.7-Flash-GGUF. It has about 197B parameters, the step35 architecture, a maximum context length of 262,144 tokens (values written in the GGUF).

There are 22 quantizations, with file sizes from 56.8 GB (UD-IQ1_M) to 394 GB (BF16) (actual sizes on Hugging Face).

Running UD-Q4_K_M (122 GB) with an 8,192-token context needs an estimated 123 GB of memory, and an estimated 125 GB at 32,768 tokens. For 8,192 tokens, a GPU with 128 GB or more of VRAM, or a Mac with 128 GB or more of unified memory, is a rough guide (macOS caps how much memory the GPU can use, so leave some headroom). The smallest, UD-IQ1_M (56.8 GB), needs an estimated 58.1 GB at 8,192 tokens.

The original model is licensed under Apache License 2.0, which allows commercial use (keep the license text and copyright notice when redistributing). Check the original license for the terms of use. An mmproj file for image input is also available.

File size and estimated memory per quantization

Up to 8 GB
Up to 16 GB
Up to 24 GB
Up to 48 GB
Up to 128 GB
More

File sizes are the actual values on Hugging Face. Memory is estimated for each context length (KV cache f16). The formula is in How memory is calculated.

All memory figures are Estimated (KV cache f16). It should run on a setup with at least this much VRAM (or Mac unified memory). Click a value to see the devices and models that fit that amount (Find by GPU / Mac).

Using image input: This model supports image input. To use images, an mmproj file is loaded as well.

About the "quality guide" (how far a quantization drifts from the original)

For each quantization type, how far the output drifts from the original model, grouped into six levels from values llama.cpp measured on Llama-3-8B.

The value is mean KLD (the difference between the original and quantized model in next-token probabilities); smaller means closer to the original. Quantizations made with an imatrix (importance data) come out slightly smaller. Levels use the value without an imatrix (or with one, for types that only have that).

These values were not measured on this model. Larger models and MoE models tend to drift less with the same type. Distributor-specific mixed types (such as UD-Q4_K_XL) cannot be matched by type name alone, so they show "-".

Quantization	Quality guide	KLD (no imatrix)	KLD (imatrix)	Size (Llama-3-8B)
Q8_0	Nearly identical	0.0014	-	7.96 GiB
Q6_K	Very small	0.0055	-	6.14 GiB
Q5_K_M	Very small	0.0108	-	5.33 GiB
Q5_K_S	Small	0.0166	-	5.21 GiB
Q5_1	Small	0.0180	-	5.65 GiB
Q5_0	Small	0.0222	-	5.21 GiB
Q4_K_M	Small	0.0313	0.0282	4.58 GiB
Q4_K_S	Medium	0.0431	0.0320	4.37 GiB
IQ4_NL	Small	-	0.0357	4.35 GiB
IQ4_XS	Small	-	0.0363	4.14 GiB
Q4_1	Medium	0.0717	-	4.78 GiB
Q4_0	Medium	0.0719	-	4.34 GiB
Q3_K_L	Medium	0.0872	0.0731	4.03 GiB
Q3_K_M	Medium	0.1019	0.0844	3.74 GiB
IQ3_M	Medium	-	0.1025	3.53 GiB
IQ3_S	Medium	-	0.1113	3.42 GiB
IQ3_XS	Large	-	0.1387	3.28 GiB
IQ3_XXS	Large	-	0.1836	3.05 GiB
Q3_K_S	Large	0.2282	0.1998	3.41 GiB
IQ2_M	Large	-	0.3260	2.74 GiB
Q2_K	Very large	0.4451	0.3322	2.96 GiB
Q2_K_S	Very large	-	0.4034	2.96 GiB
IQ2_S	Very large	-	0.4392	2.56 GiB
IQ2_XS	Very large	-	0.5463	2.43 GiB
IQ2_XXS	Very large	-	0.8120	2.24 GiB
IQ1_M	Very large	-	1.3931	2.01 GiB
IQ1_S	Very large	-	2.2546	1.88 GiB

Source: llama.cpp (MIT), measured on Llama-3-8B against FP16, Wikitext-2, revision f364eb6f. Original table · Reading and choosing quantizations

GGUF source

unsloth/Step-3.7-Flash-GGUF

Converted and published by a third party About this label

Last checked: Oct 21, 2026

Architecture: step35; llama.cpp b7964 and later
Parameters: 197.0B
Max context length: 262,144
License: apache-2.0 Commercial OK

File size and memory per quantization for Step-3.7-Flash (unsloth/Step-3.7-Flash-GGUF)
Quantization	File size Measured	Quality guide	4,096	8,192	16,384	32,768	65,536	131,072	262,144 max
UD-IQ1_M (3 parts)	56.8 GB	-	57.9 GB	58.1 GB	58.5 GB	59.3 GB	60.9 GB	64.1 GB	70.6 GB
UD-IQ2_XXS (3 parts)	61.7 GB	-	62.8 GB	63.0 GB	63.4 GB	64.3 GB	65.9 GB	69.1 GB	75.5 GB
UD-IQ2_M (3 parts)	61.8 GB	-	63.0 GB	63.2 GB	63.6 GB	64.4 GB	66.0 GB	69.2 GB	75.6 GB
UD-Q2_K_XL (3 parts)	65.8 GB	-	66.9 GB	67.1 GB	67.5 GB	68.3 GB	69.9 GB	73.2 GB	79.6 GB
UD-IQ3_XXS (3 parts)	73.6 GB	-	74.7 GB	74.9 GB	75.3 GB	76.1 GB	77.8 GB	81.0 GB	87.4 GB
UD-IQ3_S (3 parts)	80.0 GB	-	81.2 GB	81.4 GB	81.8 GB	82.6 GB	84.2 GB	87.4 GB	93.9 GB
UD-Q3_K_M (3 parts)	89.3 GB	-	90.4 GB	90.6 GB	91.0 GB	91.8 GB	93.4 GB	96.6 GB	103 GB
UD-Q3_K_XL (3 parts)	89.4 GB	-	90.5 GB	90.7 GB	91.1 GB	91.9 GB	93.6 GB	96.8 GB	103 GB
UD-IQ4_XS (3 parts)	95.3 GB	-	96.5 GB	96.7 GB	97.1 GB	97.9 GB	99.5 GB	103 GB	109 GB
UD-IQ4_NL (3 parts)	97.3 GB	-	98.5 GB	98.7 GB	99.1 GB	99.9 GB	101 GB	105 GB	111 GB
UD-Q4_K_S (4 parts)	114 GB	-	115 GB	116 GB	116 GB	117 GB	118 GB	122 GB	128 GB
MXFP4_MOE (4 parts)	118 GB	-	119 GB	120 GB	120 GB	121 GB	122 GB	126 GB	132 GB
UD-Q4_K_M (4 parts)	122 GB	-	123 GB	123 GB	124 GB	125 GB	126 GB	129 GB	136 GB
UD-Q4_K_XL (4 parts)	122 GB	-	123 GB	124 GB	124 GB	125 GB	126 GB	130 GB	136 GB
UD-Q5_K_S (4 parts)	138 GB	-	139 GB	139 GB	140 GB	141 GB	142 GB	145 GB	152 GB
UD-Q5_K_M (4 parts)	146 GB	-	148 GB	148 GB	148 GB	149 GB	151 GB	154 GB	160 GB
UD-Q5_K_XL (4 parts)	146 GB	-	148 GB	148 GB	148 GB	149 GB	151 GB	154 GB	160 GB
UD-Q6_K (5 parts)	163 GB	-	164 GB	165 GB	165 GB	166 GB	168 GB	171 GB	177 GB
UD-Q6_K_XL (5 parts)	179 GB	-	180 GB	180 GB	180 GB	181 GB	183 GB	186 GB	193 GB
Q8_0 (6 parts)	209 GB	Nearly identical	211 GB	211 GB	211 GB	212 GB	214 GB	217 GB	223 GB
UD-Q8_K_XL (6 parts)	210 GB	-	211 GB	211 GB	211 GB	212 GB	214 GB	217 GB	223 GB
BF16 (8 parts)	394 GB	Original precision	395 GB	395 GB	396 GB	397 GB	398 GB	401 GB	408 GB

The mmproj file for image input (F16) is 3.97 GB Measured . It is not included in the memory above, so add it if you use images.

Ollama  ollama run hf.co/unsloth/Step-3.7-Flash-GGUF:UD-Q4_K_M

llama.cpp  llama-server -hf unsloth/Step-3.7-Flash-GGUF:UD-Q4_K_M

License and commercial use

Apache License 2.0 Commercial OK

Allows commercial use, modification and redistribution. When redistributing, keep the license text and copyright notice, and note changes in modified files.

Based on the license on the original model's Hugging Face page. Original license text (checked Oct 18, 2026)

The summary states facts this site read from the original text and is not legal advice. Always check the original license and its latest terms before use. All licenses

How to run

Mainline llama.cpp can load this model's format (step35) from b7964 (Feb 6, 2026) onward. Ollama and LM Studio bundle llama.cpp internally, so an older app fails to load it with errors such as "unknown model architecture". If that happens, update the app.

Example commands using UD-Q4_K_M. To use another quantization, replace the name with one from the table above. How to run local LLMs

Ollama

Run it by pointing directly at the GGUF on Hugging Face.

ollama run hf.co/unsloth/Step-3.7-Flash-GGUF:UD-Q4_K_M

llama.cpp

Pass the repository and quantization name to llama-server (or llama-cli) with -hf.

llama-server -hf unsloth/Step-3.7-Flash-GGUF:UD-Q4_K_M

LM Studio

Paste unsloth/Step-3.7-Flash-GGUF into the search box, download it, and choose the quantization from the file list.

Frequently asked questions

How much memory do I need to run Step-3.7-Flash (UD-Q4_K_M)?: The file size is 122 GB (measured). With a 8,192-token context, the estimated memory including the KV cache is 123 GB. The formula is published at https://localllm.az-arte.com/en/guide/memory-calculation.
Can Step-3.7-Flash be used commercially?: The original model's license is Apache License 2.0 (Commercial use allowed).Allows commercial use, modification and redistribution. When redistributing, keep the license text and copyright notice, and note changes in modified files. Check the original license text before deciding.
Does Step-3.7-Flash run in 8 GB, 12 GB, 16 GB or 24 GB of VRAM?: 8 GB: no quantization fits; 12 GB: no quantization fits; 16 GB: no quantization fits; 24 GB: no quantization fits (estimated at 8,192 tokens with an f16 KV cache, files from the most-downloaded source).
What is the command to run Step-3.7-Flash with Ollama?: Run "ollama run hf.co/unsloth/Step-3.7-Flash-GGUF:UD-Q4_K_M" to use the GGUF on Hugging Face directly. To use another quantization, replace the name at the end.