Instructions to use PollardWeights/K2-Horizon-3.7B-Pollard with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use PollardWeights/K2-Horizon-3.7B-Pollard 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 PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S
# Run inference directly in the terminal:
llama cli -hf PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S
# Run inference directly in the terminal:
llama cli -hf PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S

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 PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S
# Run inference directly in the terminal:
./llama-cli -hf PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S

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 PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S
# Run inference directly in the terminal:
./build/bin/llama-cli -hf PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S

Use Docker

docker model run hf.co/PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S

LM Studio
Jan

vLLM

How to use PollardWeights/K2-Horizon-3.7B-Pollard with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "PollardWeights/K2-Horizon-3.7B-Pollard"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "PollardWeights/K2-Horizon-3.7B-Pollard",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S

Ollama
How to use PollardWeights/K2-Horizon-3.7B-Pollard with Ollama:
```
ollama run hf.co/PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S
```
Unsloth Desktop

How to use PollardWeights/K2-Horizon-3.7B-Pollard with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use PollardWeights/K2-Horizon-3.7B-Pollard with Docker Model Runner:
```
docker model run hf.co/PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S
```

Lemonade

How to use PollardWeights/K2-Horizon-3.7B-Pollard with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S

Run and chat with the model

lemonade run user.K2-Horizon-3.7B-Pollard-IQ3_S

List all available models

lemonade list

Hermes Agent

How to use PollardWeights/K2-Horizon-3.7B-Pollard with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S

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 PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use PollardWeights/K2-Horizon-3.7B-Pollard with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S

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 "PollardWeights/K2-Horizon-3.7B-Pollard:IQ3_S" \
  --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"

K2-Horizon-3.7B — Pollard

Pollard shrank this model: 10.1 GB (f16) → 2.75 GB — 73% smaller, 3.7× down.

The smallest rung here; larger, higher-fidelity rungs are listed below.

format this model's size

f16 10.1 GB

Q8_0 ~5.4 GB

Q6_K ~4.1 GB

Q4_K_M ~2.9 GB

PollardMix (this repo's IQ3_S) 2.75 GB

format	this model's size
f16	10.1 GB
Q8_0	~5.4 GB
Q6_K	~4.1 GB
Q4_K_M	~2.9 GB
PollardMix (this repo's IQ3_S)	2.75 GB

Pollard builds of IFM/K2-Horizon-3.7B made with Pollard Weights — a ladder of measured-allocation quants (bits placed by per-layer sensitivity, not a uniform crush).

Standard GGUF. The k2-horizon architecture is not in stock llama.cpp yet — these files need a build of the MBZUAI-IFM/llama.cpp fork. Older builds report unknown architecture 'k2-horizon'; they will run in stock llama.cpp / Ollama / LM Studio once the arch is merged upstream.

Available files (wikitext-2 test, ctx 512)

f16 reference PPL 11.2798.

file	PPL	size	Mean KLD	notes
`K2-Horizon-3.7B-Pollard-IQ3_S.gguf`	12.5164	2.75 GB	—	smallest (+1.24)
`K2-Horizon-3.7B-Pollard-IQ4_XS.gguf`	11.4211	3.33 GB	—	recommended default (+0.14)
`K2-Horizon-3.7B-Pollard-Q6_K.gguf`	11.3603	4.16 GB	—	near-lossless (+0.08 over f16)

Usage

llama-cli -m K2-Horizon-3.7B-Pollard-IQ4_XS.gguf -p "Explain why the sky is blue." --temp 0.7
ollama run hf.co/PollardWeights/K2-Horizon-3.7B-Pollard

Errata

Measured allocation places bits by per-layer sensitivity under a size budget, on a Calib 3.0 imatrix — not a uniform crush.
PPL was measured on these exact uploaded files, re-downloaded from this repo, on Apple Metal.
Building on Windows needs a one-line fix. MSVC's std::regex rejects the \p{L} escapes in K2's pre-tokenizer, so an unpatched Windows build of the fork fails to load any K2 GGUF with regex_error(error_escape); macOS is unaffected. Patch: notes/k2-horizon-msvc-regex.patch.
The base GGUF trips a special_eos_id is not in special_eog_ids warning from the fork. That is upstream tokenizer metadata, harmless for perplexity and generation.
Every rung was rebuilt from scratch on a second machine (macOS/Metal + CPU imatrix vs Windows/CUDA + GPU imatrix) and came out byte-for-byte the same size; the two machines' PPLs agree to within their error bars (Q6_K 11.3603 vs 11.3780). The allocation is reproducible, not machine-specific.

Built with Pollard Weights — frontier models, small hardware, no compromise.

Downloads last month: -

GGUF

Model size

5B params

Architecture

k2-horizon

Hardware compatibility

3-bit

4-bit

6-bit

Model tree for PollardWeights/K2-Horizon-3.7B-Pollard

Base model

IFM/K2-Horizon-3.7B

Quantized

(10)

this model

PollardWeights
/

K2-Horizon-3.7B-Pollard

K2-Horizon-3.7B — Pollard

Pollard shrank this model: 10.1 GB (f16) → 2.75 GB — 73% smaller, 3.7× down.

Available files (wikitext-2 test, ctx 512)

Usage

Errata

Model tree for PollardWeights/K2-Horizon-3.7B-Pollard