Instructions to use ayourtch/Qwen3.8-27B-shoehorn-fits with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use ayourtch/Qwen3.8-27B-shoehorn-fits 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 ayourtch/Qwen3.8-27B-shoehorn-fits
# Run inference directly in the terminal:
llama cli -hf ayourtch/Qwen3.8-27B-shoehorn-fits

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf ayourtch/Qwen3.8-27B-shoehorn-fits
# Run inference directly in the terminal:
llama cli -hf ayourtch/Qwen3.8-27B-shoehorn-fits

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 ayourtch/Qwen3.8-27B-shoehorn-fits
# Run inference directly in the terminal:
./llama-cli -hf ayourtch/Qwen3.8-27B-shoehorn-fits

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 ayourtch/Qwen3.8-27B-shoehorn-fits
# Run inference directly in the terminal:
./build/bin/llama-cli -hf ayourtch/Qwen3.8-27B-shoehorn-fits

Use Docker

docker model run hf.co/ayourtch/Qwen3.8-27B-shoehorn-fits

LM Studio
Jan

vLLM

How to use ayourtch/Qwen3.8-27B-shoehorn-fits with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ayourtch/Qwen3.8-27B-shoehorn-fits"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ayourtch/Qwen3.8-27B-shoehorn-fits",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ayourtch/Qwen3.8-27B-shoehorn-fits

Ollama
How to use ayourtch/Qwen3.8-27B-shoehorn-fits with Ollama:
```
ollama run hf.co/ayourtch/Qwen3.8-27B-shoehorn-fits
```
Unsloth Desktop

How to use ayourtch/Qwen3.8-27B-shoehorn-fits with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ayourtch/Qwen3.8-27B-shoehorn-fits

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": "ayourtch/Qwen3.8-27B-shoehorn-fits"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use ayourtch/Qwen3.8-27B-shoehorn-fits with Docker Model Runner:
```
docker model run hf.co/ayourtch/Qwen3.8-27B-shoehorn-fits
```

Lemonade

How to use ayourtch/Qwen3.8-27B-shoehorn-fits with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull ayourtch/Qwen3.8-27B-shoehorn-fits

Run and chat with the model

lemonade run user.Qwen3.8-27B-shoehorn-fits-{{QUANT_TAG}}

List all available models

lemonade list

Hermes Agent

How to use ayourtch/Qwen3.8-27B-shoehorn-fits with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ayourtch/Qwen3.8-27B-shoehorn-fits

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 ayourtch/Qwen3.8-27B-shoehorn-fits

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use ayourtch/Qwen3.8-27B-shoehorn-fits with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ayourtch/Qwen3.8-27B-shoehorn-fits

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 "ayourtch/Qwen3.8-27B-shoehorn-fits" \
  --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"

ayourtch commited on 13 days ago

Commit

f6543a0

verified ·

1 Parent(s): b089299

Add model card and per-leg benchmark results

Browse files

Model card for the four shoehorn budget-exact fits, plus the six-leg accuracy benchmark: summary JSON, per-case CSV and the exact server config for each leg. Raw model outputs are deliberately excluded (they quote benchmark questions verbatim).

Files changed (1) hide show

README.md +2 -2

README.md CHANGED Viewed

@@ -154,7 +154,7 @@ So the defensible claim is **no measurable accuracy loss against Q8_0 down to
 4.920 bpw**, and explicitly *not* that the fits are better than Q8_0. If you
 want a ranking among those five, this benchmark cannot give you one.
-### fit-17g is different, and the reason matters
 The 2.876 bpw file is the one comparison here that clears the bar. Against
 Q8_0: 19-vs-4 discordant, p = 0.003, power 0.88. Against fit-256k: 20-vs-2,
@@ -180,7 +180,7 @@ A larger token budget would likely recover part of that 16 points. We have not
 tested that, and until someone does, treat 67.4% as "this configuration, at
 32768 tokens" rather than as the model's ceiling.
-### Failure follows the question, not the quantization
 Across all six legs, 58 exhaustion events land on only 22 distinct questions,
 and 4 questions exhaust in **all six** legs — including Q8_0. Under

 4.920 bpw**, and explicitly *not* that the fits are better than Q8_0. If you
 want a ranking among those five, this benchmark cannot give you one.
+### What happens at 2.876 bpw
 The 2.876 bpw file is the one comparison here that clears the bar. Against
 Q8_0: 19-vs-4 discordant, p = 0.003, power 0.88. Against fit-256k: 20-vs-2,
 tested that, and until someone does, treat 67.4% as "this configuration, at
 32768 tokens" rather than as the model's ceiling.
+### The same questions fail in every leg
 Across all six legs, 58 exhaustion events land on only 22 distinct questions,
 and 4 questions exhaust in **all six** legs — including Q8_0. Under