Nemotron-3-Super-120B-A12B-TurboQuant-GGUF-Q3_K_M

GGUF Q3_K_M weight-quantized variant of nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 optimised for use with TurboQuant KV cache compression via a dedicated llama.cpp fork.

Important: TurboQuant KV cache types (planar3, iso3) are not available in upstream llama.cpp, standard Ollama, or LM Studio. They require a specific llama.cpp fork. The GGUF file itself is a standard GGUF and works with any llama.cpp-compatible runtime using normal KV cache types (f16, q8_0, q4_0, etc.).

Hardware compatibility

Device	VRAM / RAM	Recommendation
CPU host with ≥58 GB RAM	~58.1 GB	works via llama.cpp; slower than GPU but no accelerator required
Apple Silicon (Metal)	~63.4 GB	llama.cpp Metal backend; fast on M-series unified memory
NVIDIA GPU (partial offload)	split between GPU + RAM	offload as many layers as VRAM allows; rest on CPU

Overview

This model combines two independent compression techniques:

Technique	What it does	Requirement
GGUF Q3_K_M weight quantization	Reduces model size from ~240 GB (BF16) to ~52.8 GB	Any llama.cpp-compatible runtime
TurboQuant KV cache compression — random rotation + Lloyd-Max scalar quantization (--cache-type-k planar3 --cache-type-v planar3)	Block-diagonal rotations / random rotation for compressed KV cache	llama-cpp-turboquant fork only

Quickstart

Option A — With TurboQuant KV cache (fork required)

You must build from the TurboQuant-enabled llama.cpp fork:

# Clone and build the fork
git clone https://github.com/johndpope/llama-cpp-turboquant.git
cd llama-cpp-turboquant && git checkout feature/planarquant-kv-cache

# CUDA (Windows/Linux)
cmake -B build -DGGML_CUDA=ON -DCMAKE_BUILD_TYPE=Release && cmake --build build -j

# Metal (Apple Silicon)
cmake -B build -DGGML_METAL=ON -DGGML_METAL_EMBED_LIBRARY=ON -DCMAKE_BUILD_TYPE=Release && cmake --build build -j

# Run with TurboQuant KV cache
./build/bin/llama-cli -m Nemotron-3-Super-120B-A12B-TurboQuant-GGUF-Q3_K_M.gguf \
  --cache-type-k planar3 --cache-type-v planar3 \
  -ngl 99 -fa \
  -p "Explain quantum computing"

# Or run as a server
./build/bin/llama-server -m Nemotron-3-Super-120B-A12B-TurboQuant-GGUF-Q3_K_M.gguf \
  --cache-type-k planar3 --cache-type-v planar3 \
  -ngl 99 -fa --jinja

Option B — With standard llama.cpp / LM Studio / Ollama

The GGUF works as a normal quantised model. You won't get TurboQuant-specific KV cache benefits, but standard KV cache quantization (q8_0, q4_0) still reduces VRAM significantly.

llama.cpp (upstream)

llama-cli -m Nemotron-3-Super-120B-A12B-TurboQuant-GGUF-Q3_K_M.gguf \
  --cache-type-k q8_0 --cache-type-v q8_0 \
  -ngl 99 -fa \
  -p "Explain quantum computing"

LM Studio

1. Download the GGUF file and load in LM Studio.
2. Enable Developer Mode (Settings → Developer).
3. In the model loader's advanced settings, set Flash Attention to ON.
4. Set K Cache Quantization and V Cache Quantization to q8_0 (or q4_0 for more aggressive VRAM savings).
5. Note: LM Studio does not currently support TurboQuant's planar3 cache types. Track this feature request for updates.

Ollama

# Standard Ollama does not support TurboQuant cache types.
# Use with default or q8_0 KV cache via OLLAMA_KV_CACHE_TYPE=q8_0
OLLAMA_KV_CACHE_TYPE=q8_0 OLLAMA_FLASH_ATTENTION=1 ollama run majentik/Nemotron-3-Super-120B-A12B-TurboQuant-GGUF-Q3_K_M

Specifications

Property	Value
Base Model	nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16
Architecture	Mamba-2 + Transformer hybrid Sparse MoE
Parameters	120B total, 12B active per token
Context Length	1M
Weight Quantization	GGUF Q3_K_M (compact 3-bit, slight quality loss)
Original Size (BF16)	~240 GB
Quantized File Size	~52.8 GB
KV Cache (TurboQuant)	3-bit via --cache-type-k planar3 --cache-type-v planar3 (fork only)
KV Cache (standard)	q8_0, q4_0, f16, etc. (any llama.cpp runtime)
License	other
Modalities	Text only
Compatible Runtimes	llama.cpp, LM Studio, Ollama, koboldcpp

What is TurboQuant?

TurboQuant (ICLR 2026) is a KV cache compression method that applies a random orthogonal rotation followed by optimal scalar quantization. Bit-identical prefill logits at 4-bit on tested models, with up to 4-8× memory savings for long sequences.

Benchmarks from the TurboQuant repository (Llama 3.1 8B, RTX 5090 — results will vary by model and hardware):

Metric	TurboQuant (4-bit)	Standard q4_0
Quality	Bit-identical prefill	Lossy
KV Compression	~4× vs FP16	~4× vs FP16
Speedup (Apple Silicon)	1.4–1.7×	—

Note: These benchmarks are from the TurboQuant repository using Llama 3.1 8B on an RTX 5090. Performance on Nemotron-3-Super-120B-A12B will differ. Independent benchmarks for this specific model are welcome — please open a discussion if you have results to share.

Current Status of TurboQuant in the Ecosystem

Runtime	TurboQuant Support	Standard KV Quant
llama.cpp (upstream)	❌ Not merged	✅ q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1
llama-cpp-turboquant fork	✅ planar3	✅ All standard types
LM Studio	❌ Requested	✅ Via advanced settings
Ollama	❌ Not supported	✅ Via OLLAMA_KV_CACHE_TYPE
koboldcpp	❌ Not supported	✅ Standard types

Recommended Settings

For VRAM-constrained setups, standard q8_0 KV cache quantization already halves KV cache memory with negligible quality impact. Flash Attention should always be enabled — it is required for V cache quantization and improves memory efficiency regardless.

VRAM	Suggested Configuration
24 GB (RTX 4090)	Q3_K_M + q8_0 KV cache + Flash Attention, 8K–16K context
16 GB	Q3_K_M + q4_0 KV cache + Flash Attention, 4K–8K context
48+ GB	Q3_K_M + f16 KV cache, full 32K+ context

See Also

• RotorQuant GitHub
• llama-cpp-turboquant fork
• TurboQuant llama.cpp discussion
• TurboQuant paper (arXiv: 2504.19874)
• Base model: nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16
• Nemotron-3-Super-120B-A12B announcement