Instructions to use poisonxa/Flint-PXQN4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use poisonxa/Flint-PXQN4 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 poisonxa/Flint-PXQN4
# Run inference directly in the terminal:
llama cli -hf poisonxa/Flint-PXQN4

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf poisonxa/Flint-PXQN4
# Run inference directly in the terminal:
llama cli -hf poisonxa/Flint-PXQN4

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 poisonxa/Flint-PXQN4
# Run inference directly in the terminal:
./llama-cli -hf poisonxa/Flint-PXQN4

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 poisonxa/Flint-PXQN4
# Run inference directly in the terminal:
./build/bin/llama-cli -hf poisonxa/Flint-PXQN4

Use Docker

docker model run hf.co/poisonxa/Flint-PXQN4

LM Studio
Jan
Ollama
How to use poisonxa/Flint-PXQN4 with Ollama:
```
ollama run hf.co/poisonxa/Flint-PXQN4
```
Unsloth Desktop

How to use poisonxa/Flint-PXQN4 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf poisonxa/Flint-PXQN4

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": "poisonxa/Flint-PXQN4"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use poisonxa/Flint-PXQN4 with Docker Model Runner:
```
docker model run hf.co/poisonxa/Flint-PXQN4
```

Lemonade

How to use poisonxa/Flint-PXQN4 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull poisonxa/Flint-PXQN4

Run and chat with the model

lemonade run user.Flint-PXQN4-{{QUANT_TAG}}

List all available models

lemonade list

Hermes Agent

How to use poisonxa/Flint-PXQN4 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf poisonxa/Flint-PXQN4

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 poisonxa/Flint-PXQN4

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use poisonxa/Flint-PXQN4 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf poisonxa/Flint-PXQN4

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 "poisonxa/Flint-PXQN4" \
  --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"

Flint: PXQN4

Recursive Flint (39.8B total parameters, about 2.5B active per token, a mixture-of-experts model for coding agents), quantized by PXA Network in our PXQN4 format. The base model is RecursiveSI/flint, released by Recursive under the Apache License 2.0. This file is that model converted and quantized; it was not retrained, fine-tuned or edited. The conversion started from the Q8_0 GGUF in unsloth/flint-GGUF (also Apache-2.0), so a few descriptive fields inside the GGUF (the Unsloth name and calibration file) are inherited from that file.


File	`Flint-PXQN4.gguf`
On-disk size	24.87 GB (24,870,895,296 bytes, 23.16 GiB)
Quantization	PXQN4 tier, with the most sensitive tensors kept at higher precision
Base model	RecursiveSI/flint (Apache-2.0)
Context	up to 262,144 tokens (the model's documented limit); a smaller `-c` needs less memory
Sized for	one 16 GB card (Tesla P100 or V100) with the remaining layers in system RAM, or two 16 GB cards
sha256	`a8d2fb61e1be97c8b8b01bed00f6aa5ef5ee2cf4743c186fcf8b628c69c15992`
Needs	a PXA build with Flint support. Support ships with PXA 4.0, which reaches Valued Supporters on the beta channel first. PXA v3.1 and older cannot load this file.

Quality

Quality is reported as KL divergence (KLD) against the Q8_0 file of the original, on the same text for every file in the table. Lower is better; 0 is identical.

File	Size	KLD vs Q8_0	Same top token	PPL ratio vs Q8_0
Flint-PXQN4 (this file)	24.87 GB	0.326 ± 0.015	86.25 %	1.027
mainline Q4_K_XL	25.05 GB	0.321 ± 0.015	85.90 %	1.031

This file is level with the mainline Q4_K_XL file: the KLD difference (0.326 against 0.321) is well inside the measurement error, the top token matches 0.35 points more often, and the file is 0.18 GB smaller. We do not claim it is better than Q4_K_XL. The numbers come from 8 chunks of text (PXA measurement, 2026-10-24), so small differences are not resolved. Output is coherent in a greedy-decoding check, but this file has not been through a chat-token (assistant-role) KLD run or a long-context needle test yet.

Run it

This file needs a PXA build with Flint support. Support ships with PXA 4.0, which reaches Valued Supporters on the beta channel first; PXA v3.1 (the current public release) and older cannot load it. With such a build installed:

~/.local/share/pxa/current/run-server.sh -m Flint-PXQN4.gguf

The engine places the layers across your cards (and system RAM, if the file does not fit) on its own; add -c <tokens> to set the context size. Or open PXA Control, pick your card(s) and the file, and press Start. Recommended sampling, from the base model: temperature 1.0, top_p 0.95, top_k 20.

License

The base model, Recursive Flint, is released under the Apache License, Version 2.0 (see LICENSE, copied unchanged from RecursiveSI/flint), and its terms still apply to this file. CHANGES.md states exactly what we changed. README.md and CHANGES.md are ours; LICENSE is the source model's.

"Flint" and "Recursive" are Recursive's names, used here only to describe the origin of the weights. PXA Network is not affiliated with Recursive or Unsloth.

Credits

Recursive for Flint.
Unsloth for the Q8_0 GGUF conversion the quantization started from.

Community

Discord: https://discord.pxanetwork.com
Support PXA: https://ko-fi.com/shatteredrealms1

Downloads last month: 94

GGUF

Model size

40B params

Architecture

flint

Hardware compatibility

We're not able to determine the quantization variants.

View all variants

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for poisonxa/Flint-PXQN4

Base model

RecursiveSI/flint

Quantized

(69)

this model