Instructions to use moontato/Swift-Qwen3.8-27B-Dynamic-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use moontato/Swift-Qwen3.8-27B-Dynamic-GGUF 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 moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
llama cli -hf moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL

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 moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./llama-cli -hf moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL

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 moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL
# Run inference directly in the terminal:
./build/bin/llama-cli -hf moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL

Use Docker

docker model run hf.co/moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL

LM Studio
Jan
Ollama
How to use moontato/Swift-Qwen3.8-27B-Dynamic-GGUF with Ollama:
```
ollama run hf.co/moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL
```
Unsloth Desktop

How to use moontato/Swift-Qwen3.8-27B-Dynamic-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL

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": "moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use moontato/Swift-Qwen3.8-27B-Dynamic-GGUF with Docker Model Runner:
```
docker model run hf.co/moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL
```

Lemonade

How to use moontato/Swift-Qwen3.8-27B-Dynamic-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL

Run and chat with the model

lemonade run user.Swift-Qwen3.8-27B-Dynamic-GGUF-UD-Q4_K_XL

List all available models

lemonade list

Hermes Agent

How to use moontato/Swift-Qwen3.8-27B-Dynamic-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL

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 moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use moontato/Swift-Qwen3.8-27B-Dynamic-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL

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 "moontato/Swift-Qwen3.8-27B-Dynamic-GGUF:UD-Q4_K_XL" \
  --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"

Swift-Qwen3.8-27B UD-Q4_K_XL GGUF

A custom mixed-precision GGUF quantization of ukisai/Swift-Qwen3.8-27b.

This build combines:

Swift-Qwen3.8-27B fine-tuned weights from UkisAI
Bartowski's Swift-specific importance matrix
the UD-Q4_K_XL per-tensor quantization layout derived from Unsloth's Qwen3.8-27B dynamic quantization recipe

The goal was to retain roughly Q4-class model size and memory bandwidth requirements while selectively preserving sensitive tensors at substantially higher precision.

Model file

File	Size
`Swift-Qwen3.8-27B-UD-Q4_K_XL.gguf`	17,559,177,376 bytes
Approximate binary size	16.35 GiB
Quantizer-reported average	5.14 BPW

Tensor distribution

Quantization type	Tensor count
Q5_K	191
Q8_0	110
IQ4_XS	70
Q4_K	69
Q6_K	56
IQ4_NL	6
Q3_K	3
IQ3_S	1
Total quantized tensors	506

The source GGUF also contains 360 F32 tensors, which remain F32.

The exact per-tensor assignment is included as:

tensor_types_UD-Q4_K_XL.txt

Importance matrix

This build used:

ukisai_Swift-Qwen3.8-27b-imatrix.gguf

from:

bartowski/ukisai_Swift-Qwen3.8-27b-GGUF

The quantizer reported 496 importance-matrix entries computed from 582 calibration chunks.

The imatrix itself is not redistributed here.

So this build is:

Swift-Qwen3.8-27B weights
+ Bartowski Swift-specific imatrix
+ Unsloth UD-Q4_K_XL tensor allocation
= this quant

Quantization details

Source model:

ukisai/Swift-Qwen3.8-27b

Observed source GGUF properties:

Architecture:     qwen35
Blocks:           65
Context length:   262144
Embedding size:   5120
FFN size:         17408
Total tensors:    866
F32 tensors:      360
BF16 tensors:     506

Quantized with llama.cpp:

build:  10990
commit: 72b590d65

Reproduction

llama-quantize \
  --imatrix ukisai_Swift-Qwen3.8-27b-imatrix.gguf \
  --tensor-type-file tensor_types_UD-Q4_K_XL.txt \
  Swift-Qwen3.8-27B-BF16.gguf \
  Swift-Qwen3.8-27B-UD-Q4_K_XL.gguf \
  Q4_K_M

All 506 BF16 tensors are explicitly assigned by tensor_types_UD-Q4_K_XL.txt, so the output uses the mixed UD-Q4_K_XL allocation rather than ordinary Q4_K_M tensor selection.

Validation

The GGUF has been tested successfully with llama.cpp for:

normal chat
coding
OpenCode
tool calling
multi-file coding-agent workflows
structured code edits

It has behaved well in practical OpenCode use so far.

This is not a claim of benchmark equivalence to BF16, Q5_K_M, or Q6_K.

llama.cpp usage

llama-cli \
  -m Swift-Qwen3.8-27B-UD-Q4_K_XL.gguf \
  -cnv

Server example:

llama-server \
  -m Swift-Qwen3.8-27B-UD-Q4_K_XL.gguf \
  --host 0.0.0.0 \
  --port 8080

Multimodal / vision use

This repository contains the language-model GGUF only.

A compatible Qwen3.8-27B multimodal projector (mmproj) is required for image input and is not included here.

Checksums

SHA-256 hashes are provided in SHA256SUMS.

Verify them with:

sha256sum -c SHA256SUMS

Credits

Swift model

ukisai/Swift-Qwen3.8-27b

Base model

Qwen/Qwen3.8-27B

Swift importance matrix

bartowski/ukisai_Swift-Qwen3.8-27b-GGUF

UD-Q4_K_XL allocation

The tensor allocation is based on Unsloth's UD-Q4_K_XL layout for Qwen3.8-27B.

The tensor map used to reproduce that allocation was obtained through:

ajgazin/Qwen3.8-27B-Heretic-Dynamic-GGUF

License

This repository contains a quantized derivative of ukisai/Swift-Qwen3.8-27b.

The Swift fine-tuned weights are distributed under the Swift Open License v1.0. Review the original Swift model repository and license before redistribution or commercial use.

Disclaimer

This is an independently produced community quantization and is not an official UkisAI, Qwen, Bartowski, or Unsloth release.

Downloads last month: 27

GGUF

Model size

27B params

Architecture

qwen35

Hardware compatibility

4-bit

Inference Providers NEW

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

Model tree for moontato/Swift-Qwen3.8-27B-Dynamic-GGUF

Base model

Qwen/Qwen3.8-27B

Finetuned

ukisai/Swift-Qwen3.8-27b

Quantized

(25)

this model