Skip to content
localllm
日本語

nvidia

NVIDIA-Nemotron-3.5-Transcribe-30B-A4B

Speech / audio

Parameters
32.9B
Max context length
1,048,576
Architecture
nemotron_h_moe
License
Commercial OK
Base model
nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B
License
openmdw-1.1
llama.cpp
b7418+

About NVIDIA-Nemotron-3.5-Transcribe-30B-A4B

NVIDIA-Nemotron-3.5-Transcribe-30B-A4B is a GGUF model based on nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B on Hugging Face. GGUF files are published by 2 sources (AtomicChat/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-GGUF, Jmica/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-GGUF). It has about 32.9B parameters, the nemotron_h_moe architecture, a maximum context length of 1,048,576 tokens (values written in the GGUF).

There are 9 quantizations, with file sizes from 1.67 GB (F16) to 35.0 GB (Q8_0) (actual sizes on Hugging Face).

Running F16 (1.67 GB) with an 8,192-token context needs an estimated 2.58 GB of memory, and an estimated 2.73 GB at 32,768 tokens. For 8,192 tokens, a GPU with 4 GB or more of VRAM (such as Arc A310, Arc A370M), or a Mac with 4 GB or more of unified memory, is a rough guide (macOS caps how much memory the GPU can use, so leave some headroom).

The original model is licensed under OpenMDW License 1.1, 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).

2 sources publish GGUF files of this model. Quantization types and file sizes differ slightly between sources, so there is a table for each source below.

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

AtomicChat/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-GGUF

Converted and published by a third party About this label

Last checked: Nov 11, 2026

Architecture
nemotron_h_moe
llama.cpp b7418 and later
Parameters
32.9B
Max context length
1,048,576
License
openmdw-1.1 Commercial OK
File size and memory per quantization for NVIDIA-Nemotron-3.5-Transcribe-30B-A4B (AtomicChat/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-GGUF)
Quantization File size Measured Quality guide 4,096 8,192 16,384 32,768 65,536 131,072 262,144 1,048,576 max
F16 1.67 GB Original precision 2.55 GB 2.58 GB 2.63 GB 2.73 GB 2.93 GB 3.33 GB 4.14 GB 8.97 GB
Q2_0 NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-AD-Q2_0_M 13.6 GB - 14.5 GB 14.5 GB 14.5 GB 14.6 GB 14.8 GB 15.2 GB 16.0 GB 20.9 GB
Q2_0 NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-AD-Q2_0_L 15.5 GB - 16.4 GB 16.5 GB 16.5 GB 16.6 GB 16.8 GB 17.2 GB 18.0 GB 22.8 GB
NVFP4 19.7 GB - 20.6 GB 20.6 GB 20.6 GB 20.7 GB 20.9 GB 21.3 GB 22.1 GB 27.0 GB
IQ4_NL NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-AD-IQ4_NL 19.7 GB Small 20.6 GB 20.6 GB 20.6 GB 20.7 GB 20.9 GB 21.3 GB 22.1 GB 27.0 GB
IQ4_NL NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-AD-IQ4_NL_L 22.6 GB Small 23.4 GB 23.5 GB 23.5 GB 23.6 GB 23.8 GB 24.2 GB 25.0 GB 29.8 GB
Q5_0 29.3 GB Small 30.1 GB 30.2 GB 30.2 GB 30.3 GB 30.5 GB 30.9 GB 31.7 GB 36.6 GB
Q8_0 35.0 GB Nearly identical 35.9 GB 35.9 GB 36.0 GB 36.1 GB 36.3 GB 36.7 GB 37.5 GB 42.3 GB 
Ollama  ollama run hf.co/AtomicChat/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-GGUF:F16
llama.cpp  llama-server -hf AtomicChat/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-GGUF:F16

GGUF source

Jmica/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-GGUF

Converted and published by a third party About this label

Last checked: Nov 11, 2026

Architecture
nemotron_h_moe
llama.cpp b7418 and later
License
openmdw-1.1 Commercial OK
File size and memory per quantization for NVIDIA-Nemotron-3.5-Transcribe-30B-A4B (Jmica/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-GGUF)
Quantization File size Measured Quality guide 4,096 8,192 16,384 32,768 65,536 131,072 262,144 1,048,576
Q2_K 13.9 GB Very large 14.8 GB 14.9 GB 14.9 GB 15.0 GB 15.2 GB 15.6 GB 16.4 GB 21.2 GB
Q3_K_S 17.3 GB Large 18.2 GB 18.2 GB 18.3 GB 18.4 GB 18.6 GB 19.0 GB 19.8 GB 24.6 GB
IQ4_NL 20.4 GB Small 21.3 GB 21.3 GB 21.3 GB 21.4 GB 21.6 GB 22.0 GB 22.8 GB 27.7 GB
Q4_K_S 21.7 GB Medium 22.6 GB 22.6 GB 22.6 GB 22.7 GB 22.9 GB 23.3 GB 24.1 GB 29.0 GB

The mmproj file for image input (F16) is 1.67 GB Measured . It is not included in the memory above, so add it if you use images. 

Ollama  ollama run hf.co/Jmica/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-GGUF:Q2_K
llama.cpp  llama-server -hf Jmica/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-GGUF:Q2_K

License and commercial use

OpenMDW License 1.1 Commercial OK

Allows use without restriction, including commercial use. When redistributing, keep the license text and copyright notice. There are no restrictions on using outputs.

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 (nemotron_h_moe) from b7418 (Dec 16, 2025) 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 F16. To use another quantization, replace the name with one from the table above. Commands for files from other sources are in each source's section. How to run local LLMs

Ollama

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

ollama run hf.co/AtomicChat/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-GGUF:F16

llama.cpp

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

llama-server -hf AtomicChat/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-GGUF:F16

LM Studio

Paste AtomicChat/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-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 NVIDIA-Nemotron-3.5-Transcribe-30B-A4B (F16)?
The file size is 1.67 GB (measured). With a 8,192-token context, the estimated memory including the KV cache is 2.58 GB. The formula is published at https://localllm.az-arte.com/en/guide/memory-calculation.
Can NVIDIA-Nemotron-3.5-Transcribe-30B-A4B be used commercially?
The original model's license is OpenMDW License 1.1 (Commercial use allowed).Allows use without restriction, including commercial use. When redistributing, keep the license text and copyright notice. There are no restrictions on using outputs. Check the original license text before deciding.
Does NVIDIA-Nemotron-3.5-Transcribe-30B-A4B run in 8 GB, 12 GB, 16 GB or 24 GB of VRAM?
8 GB: up to F16 (est. 2.58 GB); 12 GB: up to F16 (est. 2.58 GB); 16 GB: up to Q2_0 (est. 16.5 GB); 24 GB: up to IQ4_NL (est. 23.5 GB) (estimated at 8,192 tokens with an f16 KV cache, files from the most-downloaded source).
What is the command to run NVIDIA-Nemotron-3.5-Transcribe-30B-A4B with Ollama?
Run "ollama run hf.co/AtomicChat/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-GGUF:F16" to use the GGUF on Hugging Face directly. To use another quantization, replace the name at the end.