Instructions to use logic65/Qwen3.8-Whittle-v40-16.8B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use logic65/Qwen3.8-Whittle-v40-16.8B 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 logic65/Qwen3.8-Whittle-v40-16.8B
# Run inference directly in the terminal:
llama cli -hf logic65/Qwen3.8-Whittle-v40-16.8B

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf logic65/Qwen3.8-Whittle-v40-16.8B
# Run inference directly in the terminal:
llama cli -hf logic65/Qwen3.8-Whittle-v40-16.8B

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 logic65/Qwen3.8-Whittle-v40-16.8B
# Run inference directly in the terminal:
./llama-cli -hf logic65/Qwen3.8-Whittle-v40-16.8B

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 logic65/Qwen3.8-Whittle-v40-16.8B
# Run inference directly in the terminal:
./build/bin/llama-cli -hf logic65/Qwen3.8-Whittle-v40-16.8B

Use Docker

docker model run hf.co/logic65/Qwen3.8-Whittle-v40-16.8B

LM Studio
Jan

vLLM

How to use logic65/Qwen3.8-Whittle-v40-16.8B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "logic65/Qwen3.8-Whittle-v40-16.8B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "logic65/Qwen3.8-Whittle-v40-16.8B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/logic65/Qwen3.8-Whittle-v40-16.8B

Ollama
How to use logic65/Qwen3.8-Whittle-v40-16.8B with Ollama:
```
ollama run hf.co/logic65/Qwen3.8-Whittle-v40-16.8B
```
Unsloth Desktop

How to use logic65/Qwen3.8-Whittle-v40-16.8B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf logic65/Qwen3.8-Whittle-v40-16.8B

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": "logic65/Qwen3.8-Whittle-v40-16.8B"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use logic65/Qwen3.8-Whittle-v40-16.8B with Docker Model Runner:
```
docker model run hf.co/logic65/Qwen3.8-Whittle-v40-16.8B
```

Lemonade

How to use logic65/Qwen3.8-Whittle-v40-16.8B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull logic65/Qwen3.8-Whittle-v40-16.8B

Run and chat with the model

lemonade run user.Qwen3.8-Whittle-v40-16.8B-{{QUANT_TAG}}

List all available models

lemonade list

Hermes Agent

How to use logic65/Qwen3.8-Whittle-v40-16.8B with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf logic65/Qwen3.8-Whittle-v40-16.8B

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 logic65/Qwen3.8-Whittle-v40-16.8B

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use logic65/Qwen3.8-Whittle-v40-16.8B with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf logic65/Qwen3.8-Whittle-v40-16.8B

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 "logic65/Qwen3.8-Whittle-v40-16.8B" \
  --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"

Qwen3.8-Whittle-v40 (16.8B) best measured cut

☕ Support this work

Whittle is built by one person on a grocery budget and rented GPU hours. If this research is useful to you, or you want to see it finished: ko-fi.com/davida81328. Every hour of GPU time goes straight into the next checkpoint, and every checkpoint, table and log lands in these repos.

40 layers of Qwen3.8-27B, kept in the exact [gdn, gdn, gdn, attention] pattern llama.cpp requires: 30 gated-DeltaNet layers, 10 full-attention stations, a true 3:1 ratio. Which layers survive was decided by a per-layer identity map (cosine between each layer's input and output residual stream over 80 probes): within every span between surviving attention stations, the three most load-bearing recurrent layers are kept and the rest are dropped.

Status

Earlier line (Aug 2026); the current Whittle models are the Whittle-Next / Whittle-Qwen-3.8 line: Whittle-Qwen-3.8-35B-A3B.

Research preview, no training: smaller than every other cut in the family and better than all of them on the knowledge-atlas measurement below; chosen by measurement, not by assumption. It needs post training and is published as the record of a compression method and its measurements rather than as a finished assistant. Further healing and instruction tuning runs are planned over time, and these checkpoints will improve as those land.

Which weights are which

path	what it is
`q38_v40_q8.gguf`	the model, Q8_0 GGUF (the only weights in this repo)
`recipe/q38_build_keep_hf.py`, `recipe/variant_drops.json`	the build script and the variant layer-drop table that produced it

Measured

Measured with an 80-probe, 8-domain knowledge atlas, scored as the mean fraction of the intact model's next-token probability retained:

shape	params	layers	retention
v40 (this)	16.8B	40	0.648
44L cut (not currently published)	19.2B	44	0.408
same size, attention crowded early	16.8B	40	0.379
same rule, 36 layers	15.1B	36	0.441

Two lessons are in that table. Choosing layers by measurement instead of removing contiguous bands is worth more than 2 billion parameters. And where the attention stations sit matters more than how many there are: three variants at identical size and identical selection rule score 0.648, 0.544 and 0.379 purely by station placement.

Against the intact 27B, this cut is at or above the original on agentic control (0.0922 vs 0.0885), Python (0.0030 vs 0.0023) and science (0.0035 vs 0.0034), holds 90% on math, and keeps 58% of multilingual recall where the 44L cut kept 3%. Its end-of-turn probability at a finished answer, the quantity that governs looping, stays at 0.46 and 0.44 versus the original's 0.53 and 0.33.

How it was made

No training. The cut is the identity-guided layer selection described at the top of this card, built with the recipe in recipe/.

Run it

Anti-loop sampling is still recommended: --dry-multiplier 0.8 --dry-base 1.75 --dry-allowed-length 4 --repeat-penalty 1.15

Caveats

No repair training has been done. It is a research preview: expect dulled confidence on hard recall and the family's known rough edges. C++ recall is the weakest domain, being distributed across layers no cut preserves.

Support this work

Independent research on consumer hardware. Every donation becomes A100 hours, and every A100 hour ends up as a public model or a public measurement. ☕ ko-fi.com/davida81328