Instructions to use majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K
# Run inference directly in the terminal:
llama cli -hf majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K
# Run inference directly in the terminal:
llama cli -hf majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K
# Run inference directly in the terminal:
./llama-cli -hf majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K
# Run inference directly in the terminal:
./build/bin/llama-cli -hf majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K

Use Docker

docker model run hf.co/majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K

LM Studio
Jan

vLLM

How to use majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K

Ollama
How to use majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K with Ollama:
```
ollama run hf.co/majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K
```

Unsloth Studio

How to use majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K to start chatting

How to use majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Docker Model Runner
How to use majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K with Docker Model Runner:
```
docker model run hf.co/majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K
```

Lemonade

How to use majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K:Q2_K

Run and chat with the model

lemonade run user.Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K-Q2_K

List all available models

lemonade list

majentik commited on Apr 15

Commit

1196d7d

verified ·

1 Parent(s): 03ea283

Update model card with accurate fork requirements and ecosystem status

Browse files

Files changed (1) hide show

README.md +110 -41

README.md CHANGED Viewed

@@ -1,5 +1,7 @@
 ---
-library_name: gguf
 base_model: nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16
 tags:
   - gguf
@@ -11,75 +13,142 @@ tags:
   - hybrid
   - llama-cpp
   - quantized
-license: other
-license_name: nvidia-open-model-license
-license_link: https://developer.download.nvidia.com/licenses/nvidia-open-model-license-agreement-june-2024.pdf
 ---
 # Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K
-GGUF Q2_K weight-quantized variant of [nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16) with **RotorQuant** KV cache compression for efficient inference with llama.cpp, Ollama, and LM Studio. Features a dense hybrid Mamba-2 + Attention architecture with 4B parameters and up to 262K context length. Maximum compression for memory-constrained environments.
 ## Overview
-This model combines two compression techniques:
-- **GGUF Q2_K weight quantization** -- reduces model size from ~8 GB to ~1.5 GB
-- **RotorQuant KV cache compression** -- block-diagonal rotations (Clifford algebra) for 3-bit KV cache, 5.3x faster prefill
 ## Quickstart
-### llama.cpp
 ```bash
-llama-cli -m Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K.gguf \
-  --cache-type-k planar3 --cache-type-v iso3 \
   -p "Explain quantum computing"
 ```
-### Ollama
 ```bash
-ollama run majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K
 ```
-### LM Studio
-Download the GGUF file and load in LM Studio. Enable RotorQuant KV cache in advanced settings.
 ## Specifications
 | Property | Value |
 |----------|-------|
-| Base Model | nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16 |
-| Parameters | 4B (dense, Mamba-2 + Attention hybrid) |
-| Context Length | 262,144 tokens (262K) |
-| Weight Quantization | GGUF Q2_K |
-| KV Cache | RotorQuant 3-bit (planar/iso) |
-| File Size | ~1.5 GB |
-| License | NVIDIA Open Model License (commercial use OK) |
-| Compatible | llama.cpp, Ollama, LM Studio, koboldcpp |
 ## What is RotorQuant?
-RotorQuant applies block-diagonal rotations (Clifford algebra) for KV cache compression. When used with llama.cpp's `--cache-type-k planar3 --cache-type-v iso3` flags:
-| Metric | RotorQuant | TurboQuant |
-|--------|-----------|-----------|
-| Prefill Speed | 3,822 tok/s | 722 tok/s |
-| Decode Speed | 119 tok/s | 93 tok/s |
-| Perplexity | 6.91 | 7.07 |
-## GGUF Quant Variants
-| Quant | File Size | Quality | Variant |
-|-------|-----------|---------|---------|
-| **Q2_K** | **~1.5 GB** | **Lowest (max compression)** | **This model** |
-| Q3_K_M | ~1.8 GB | Low | [Q3_K_M](https://huggingface.co/majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q3_K_M) |
-| IQ4_XS | ~2 GB | Medium-Low | [IQ4_XS](https://huggingface.co/majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-IQ4_XS) |
-| Q4_K_M | ~2.5 GB | Medium | [Q4_K_M](https://huggingface.co/majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q4_K_M) |
-| Q5_K_M | ~3 GB | Medium-High | [Q5_K_M](https://huggingface.co/majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q5_K_M) |
-| Q8_0 | ~4.5 GB | High | [Q8_0](https://huggingface.co/majentik/Nemotron-3-Nano-4B-RotorQuant-GGUF-Q8_0) |
 ## See Also
-- [nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16) -- Base model
-- [majentik/Nemotron-3-Nano-4B-RotorQuant](https://huggingface.co/majentik/Nemotron-3-Nano-4B-RotorQuant) -- RotorQuant KV-cache (transformers)
-- [majentik/Nemotron-3-Nano-4B-RotorQuant-MLX-2bit](https://huggingface.co/majentik/Nemotron-3-Nano-4B-RotorQuant-MLX-2bit) -- MLX 2-bit variant
 - [RotorQuant GitHub](https://github.com/scrya-com/rotorquant)

 ---
+license: other
+license_name: nvidia-open-model-license
+license_link: https://developer.download.nvidia.com/licenses/nvidia-open-model-license-agreement-june-2024.pdf
 base_model: nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16
 tags:
   - gguf
   - hybrid
   - llama-cpp
   - quantized
+library_name: gguf
+pipeline_tag: text-generation
 ---
 # Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K
+GGUF Q2_K weight-quantized variant of [nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16) optimised for use with **RotorQuant** KV cache compression via a dedicated llama.cpp fork.
+> **Important:** RotorQuant KV cache types (`planar3`, `iso3`) are **not** available in upstream llama.cpp, standard Ollama, or LM Studio.
+> They require a [specific llama.cpp fork](https://github.com/johndpope/llama-cpp-turboquant/tree/feature/planarquant-kv-cache).
+> 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.).
 ## Overview
+This model combines two independent compression techniques:
+| Technique | What it does | Requirement |
+|-----------|-------------|-------------|
+| **GGUF Q2_K weight quantization** | Reduces model size from ~8 GB (BF16) to ~1.4 GB | Any llama.cpp-compatible runtime |
+| **RotorQuant KV cache compression** — block-diagonal Clifford-algebra rotors for 3-bit KV cache (`--cache-type-k iso3 --cache-type-v iso3`) | Block-diagonal rotations / random rotation for compressed KV cache | [llama-cpp-turboquant fork](https://github.com/johndpope/llama-cpp-turboquant/tree/feature/planarquant-kv-cache) only |
 ## Quickstart
+### Option A — With RotorQuant KV cache (fork required)
+You must build from the RotorQuant-enabled llama.cpp fork:
 ```bash
+# 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 RotorQuant KV cache
+./build/bin/llama-cli -m Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K.gguf \
+  --cache-type-k iso3 --cache-type-v iso3 \
+  -ngl 99 -fa \
   -p "Explain quantum computing"
+# Or run as a server
+./build/bin/llama-server -m Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K.gguf \
+  --cache-type-k iso3 --cache-type-v iso3 \
+  -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 RotorQuant-specific KV cache benefits, but standard KV cache quantization (q8_0, q4_0) still reduces VRAM significantly.
+**llama.cpp (upstream)**
 ```bash
+llama-cli -m Nemotron-3-Nano-4B-RotorQuant-GGUF-Q2_K.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 RotorQuant's `iso3` cache types. Track [this feature request](https://github.com/lmstudio-ai/lmstudio-bug-tracker/issues/1719) for updates.
+**Ollama**
+```bash
+# Standard Ollama does not support RotorQuant 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-Nano-4B-RotorQuant-GGUF-Q2_K
+```
 ## Specifications
 | Property | Value |
 |----------|-------|
+| Base Model | [nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16) |
+| Architecture | Mamba-2 + Transformer hybrid (dense) |
+| Parameters | 4B (dense hybrid) |
+| Context Length | 262K |
+| Weight Quantization | GGUF Q2_K (aggressive 2-bit, noticeable quality drop) |
+| Original Size (BF16) | ~8 GB |
+| Quantized File Size | ~1.4 GB |
+| KV Cache (RotorQuant) | 3-bit via `--cache-type-k iso3 --cache-type-v iso3` (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 RotorQuant?
+[RotorQuant](https://github.com/scrya-com/rotorquant) is a KV cache compression method based on Clifford algebra (Cl(3,0)) rotors. It was developed as a faster, more parameter-efficient alternative to Google's [TurboQuant](https://arxiv.org/abs/2504.19874) (ICLR 2026).
+Instead of applying a dense d×d random orthogonal rotation matrix (as TurboQuant does), RotorQuant uses lightweight block-diagonal rotations — independent 2D/4D rotations per pair/quartet — achieving O(d) complexity instead of O(d log d), fully parallelisable with no inter-element dependencies.
+**Benchmarks from the RotorQuant repository** (Llama 3.1 8B, RTX 5090 — results will vary by model and hardware):
+| Metric | RotorQuant (iso3) | TurboQuant | Standard q4_0 |
+|--------|-------------------|------------|---------------|
+| Prefill Speed | 3,822 tok/s | 722 tok/s | — |
+| Decode Speed | 119 tok/s | 93 tok/s | — |
+| Perplexity (PPL) | 6.91 | 7.07 | — |
+| KV Compression | ~5× vs FP16 | ~5× vs FP16 | ~4× vs FP16 |
+| Rotation Parameters | 4 per rotor | 16,384 per matrix | N/A |
+> **Note:** These benchmarks are from the RotorQuant repository using Llama 3.1 8B on an RTX 5090. Performance on Nemotron-3-Nano-4B will differ. Independent benchmarks for this specific model are welcome — please open a discussion if you have results to share.
+## Current Status of RotorQuant in the Ecosystem
+| Runtime | RotorQuant 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, iso3 | ✅ All standard types |
+| LM Studio | ❌ [Requested](https://github.com/lmstudio-ai/lmstudio-bug-tracker/issues/1719) | ✅ 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) | Q2_K + q8_0 KV cache + Flash Attention, 8K–16K context |
+| 16 GB | Q2_K + q4_0 KV cache + Flash Attention, 4K–8K context |
+| 48+ GB | Q2_K + f16 KV cache, full 32K+ context |
 ## See Also
 - [RotorQuant GitHub](https://github.com/scrya-com/rotorquant)
+- [llama-cpp-turboquant fork](https://github.com/johndpope/llama-cpp-turboquant/tree/feature/planarquant-kv-cache)
+- [TurboQuant llama.cpp discussion](https://github.com/ggml-org/llama.cpp/discussions/20969)
+- [TurboQuant paper (arXiv: 2504.19874)](https://arxiv.org/abs/2504.19874)
+- [Base model: nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16)
+- [Nemotron-3-Nano-4B announcement](https://huggingface.co/blog/nvidia/nemotron-3-nano-4b)