Instructions to use logic65/Qwen3.6-Whittle-25B-A3B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use logic65/Qwen3.6-Whittle-25B-A3B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="logic65/Qwen3.6-Whittle-25B-A3B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("logic65/Qwen3.6-Whittle-25B-A3B")
model = AutoModelForCausalLM.from_pretrained("logic65/Qwen3.6-Whittle-25B-A3B", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use logic65/Qwen3.6-Whittle-25B-A3B 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/Qwen3.6-Whittle-25B-A3B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf logic65/Qwen3.6-Whittle-25B-A3B:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf logic65/Qwen3.6-Whittle-25B-A3B:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf logic65/Qwen3.6-Whittle-25B-A3B:Q4_K_M

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/Qwen3.6-Whittle-25B-A3B:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf logic65/Qwen3.6-Whittle-25B-A3B:Q4_K_M

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/Qwen3.6-Whittle-25B-A3B:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf logic65/Qwen3.6-Whittle-25B-A3B:Q4_K_M

Use Docker

docker model run hf.co/logic65/Qwen3.6-Whittle-25B-A3B:Q4_K_M

LM Studio
Jan

vLLM

How to use logic65/Qwen3.6-Whittle-25B-A3B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "logic65/Qwen3.6-Whittle-25B-A3B"
# 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/Qwen3.6-Whittle-25B-A3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/logic65/Qwen3.6-Whittle-25B-A3B:Q4_K_M

SGLang

How to use logic65/Qwen3.6-Whittle-25B-A3B with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "logic65/Qwen3.6-Whittle-25B-A3B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "logic65/Qwen3.6-Whittle-25B-A3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "logic65/Qwen3.6-Whittle-25B-A3B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "logic65/Qwen3.6-Whittle-25B-A3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use logic65/Qwen3.6-Whittle-25B-A3B with Ollama:
```
ollama run hf.co/logic65/Qwen3.6-Whittle-25B-A3B:Q4_K_M
```
Unsloth Desktop

How to use logic65/Qwen3.6-Whittle-25B-A3B with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf logic65/Qwen3.6-Whittle-25B-A3B:Q4_K_M

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/Qwen3.6-Whittle-25B-A3B:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use logic65/Qwen3.6-Whittle-25B-A3B with Docker Model Runner:
```
docker model run hf.co/logic65/Qwen3.6-Whittle-25B-A3B:Q4_K_M
```

Lemonade

How to use logic65/Qwen3.6-Whittle-25B-A3B with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull logic65/Qwen3.6-Whittle-25B-A3B:Q4_K_M

Run and chat with the model

lemonade run user.Qwen3.6-Whittle-25B-A3B-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use logic65/Qwen3.6-Whittle-25B-A3B 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/Qwen3.6-Whittle-25B-A3B:Q4_K_M

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/Qwen3.6-Whittle-25B-A3B:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use logic65/Qwen3.6-Whittle-25B-A3B with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf logic65/Qwen3.6-Whittle-25B-A3B:Q4_K_M

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/Qwen3.6-Whittle-25B-A3B:Q4_K_M" \
  --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"

Qwen3.6-Whittle-25B-A3B

☕ 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.

This repo holds two different things. (1) Whittle-Qwen-3.8-25B-A3B (3 Oct 2026): five GGUFs of the current Whittle flagship's body with its 10 B n-gram memory removed, described in the next section. (2) The Qwen3.6-Whittle-25B-A3B prune+heal body (2 Sep 2026): Qwen3.6-35B-A3B with 30% of its routed experts removed (256 → 180 per layer) and the damage healed by self-distillation from the unpruned weights — the safetensors at the root of this repo and the experimental next/ build, described further down. The two are different models in different formats; the second is the starting point the Whittle-Next line was built from.

Status

Whittle-Qwen-3.8-25B-A3B GGUFs: current (3 Oct 2026). They derive from the current flagship Whittle-Qwen-3.8-35B-A3B (root = Phase-2 step 32010), whose GGUFs with the memory are on Whittle-Qwen-3.8-35B-A3B-GGUF.
Qwen3.6-Whittle-25B-A3B body (2 Sep 2026): the base of the Whittle-Next line. It was converted to the qwen4_exp format and continued as Whittle-Next-26B-A3B → Whittle-Next-27B-A3B → the 35B above. Its own numbers are below.

Whittle-Qwen-3.8-25B-A3B: the 35B body without its memory (3 Oct 2026)

These five files are the Phase-2 step 32010 body of Whittle-Qwen-3.8-35B-A3B with its 10 B n-gram memory removed: the memory table is replaced by an 8-row all-zero table, and every other tensor is byte-identical to the 35B GGUF ladder. They run on stock llama.cpp.

Which file

file	size
`Whittle-Qwen-3.8-25B-A3B-Q8_0.gguf`	27.21 GB
`Whittle-Qwen-3.8-25B-A3B-Q6_K.gguf`	21.08 GB
`Whittle-Qwen-3.8-25B-A3B-Q5_K_M.gguf`	18.26 GB
`Whittle-Qwen-3.8-25B-A3B-Q4_K_M.gguf`	15.66 GB
`Whittle-Qwen-3.8-25B-A3B-Q3_K_M.gguf`	12.44 GB

Measured (3 Oct 2026, on the published Q6_K)

Measured on Whittle-Qwen-3.8-25B-A3B-Q6_K.gguf from this repo, served by stock llama.cpp on two T4 GPUs, one sample per item. PopQA-288 and the 40 facts are cloze prompts scored on a greedy (temperature 0) raw-text completion; every other probe goes through the chat endpoint with the card sampler (temperature 0.7, top_p 0.8, top_k 20, repeat_penalty 1.05) and thinking on. The right-hand column is the 35B with its full memory table (its Q6_K), run through the same harness on 2 Oct 2026.

probe	this repo's Q6_K (3 Oct)	35B Q6_K, full memory table (2 Oct)
PopQA-288 cloze, exact match	69/288	71/288
40 plain facts	35/40	35/40
stop/loop battery, 24 prompts	23/24 clean (0 loops, 0 user-turn leaks)	23/24 clean (0 loops, 0 user-turn leaks)
GSM-50	42/50	44/50
strict-JSON hold-out, 72 items	72/72 valid, 71/72 exact	72/72 valid, 70/72 exact
tool-use probes, 96 items	91/96	95/96
of which the ~11k-token context probe (24 items)	19/24	23/24
llama.cpp web page, browser tools: a real chat replayed at two points (20 each)	17/20 and 18/20, 0 invented tools	16/20 and 18/20, 0 invented tools
llama.cpp web page, browser tools: 36 held-out chats finished correctly	28/36	32/36
MATH-60, 6,144-token cap	38/60	not run for the full-table file in that session

Every probe is one sample per item, so differences of a few items between the two columns may be noise.

Run it

llama-server -m <file> -ngl 99 -c 16384 --jinja -fa on

The memory is gone, so -ot per_layer_token_embd=CPU is not needed. Sampler: temperature 0.7, top_p 0.8, top_k 20, repeat_penalty 1.05. Thinking is switched with "chat_template_kwargs": {"enable_thinking": true}.

Caveat

The body depends on the memory on raw text and code completion: cross-entropy on held code collapses without it. These files are for chat, instruction and agent use, not for base-model-style completion.

Qwen3.6-Whittle-25B-A3B: the 2 Sep 2026 prune+heal body (different lineage and format)

Everything below describes the earlier artefact: the qwen3_5_moe-format safetensors at the root of this repo and the next/ build. It is not the model in the GGUFs above.

Qwen3.6-35B-A3B with 30% of its routed experts removed (256 → 180 per layer) and the damage healed by self-distillation from the unpruned weights. Same architecture (qwen3_5_moe), same tokenizer, same 8-routed + 1-shared experts active per token. Runs on stock transformers and stock llama.cpp, no patches.

	Qwen3.6-35B-A3B	Qwen3.6-Whittle-25B-A3B
total parameters (text)	34.7B	25.1B
active per token	~3B	~3B (unchanged)
routed experts / layer	256	180
GSM8K (200 q, no-think, greedy, 512 tok)	88.5%	92.5% (185/200)
held-out CE, corpus text (never seen in calibration/healing)	1.770	1.863 (raw prune 1.926)
held-out CE, chat rows	2.510	1.417
bf16 on disk	70 GB	50 GB

Which files are which

Root: model-00001-of-00015.safetensors … model-00015-of-00015.safetensors + model.safetensors.index.json, config.json, chat_template.jinja, generation_config.json, tokenizer files, gate_act.json — the prune+heal body in qwen3_5_moe format.
eval/ — gsm8k_base_qwen3.6-35b-a3b.json, gsm8k_pruned_healed.json, prune_report.json.
next/ — the experimental qwen4exp build (see below): Qwen3.6-Whittle-25B-A3B-next-EXPERIMENTAL-Q8_0.gguf, sigmoid_settle_step1500.pt, gate_act.json, gsm8k_q4exp_stock_llamacpp.json, gsm8k_sigmoid_transformers.json.
Whittle-Qwen-3.8-25B-A3B-*.gguf — the five files of the section above, a different model.

How it was made

Score every expert on 1M calibration tokens (encyclopaedia, textbooks, maths, code, chat) with a gate-free mean-squared-activation-norm criterion (the task-agnostic winner in the June-2026 one-shot MoE pruning study).
Prune the 76 lowest-scoring experts per layer; router rows sliced to match. Kept experts carried 82% of routed traffic on average (71% in the worst layer).
Heal for 900 steps × 2048 tokens (≈1.8M tokens) against the unpruned model as teacher — the same weights with the mask off, so no second model and no drift: loss = 2·KL(teacher‖student) + 1·CE. Trained: routers, shared experts, rank-8 LoRA on the kept experts (merged into the exported weights). Optimiser: Muon on 2D matrices, AdamW on the rest. Data: 50% raw corpus windows, 50% template-faithful instruction rows.
Export kept experts from the original bf16 shards + LoRA deltas (no quantisation error baked in).

Run it

# transformers ≥ 5.16
from transformers import AutoModelForCausalLM, AutoTokenizer
m = AutoModelForCausalLM.from_pretrained("logic65/Qwen3.6-Whittle-25B-A3B", dtype="bfloat16", device_map="auto")
# llama.cpp (stock): GGUF Q4_K_M in this repo
llama-server -m Qwen3.6-Whittle-25B-A3B-Q4_K_M.gguf -ngl 99 -c 8192 --jinja

Recommended sampling as the parent: temperature 0.7, top_p 0.8, top_k 20, repeat_penalty 1.05. Thinking mode works as in the parent.

Note (3 Oct 2026): Qwen3.6-Whittle-25B-A3B-Q4_K_M.gguf, named in the block above, is not in this repo's current file listing. The GGUFs in the repo are the Whittle-Qwen-3.8-25B-A3B ladder (a different model, see above) and the experimental next/ build.

Why is GSM8K higher than the parent?

Eight more correct answers out of 200; sampling noise at this size is about ±2 points, so treat it as "at least parity". Because the heal is also an SFT pass. Half of the healing batches were template-faithful instruction rows, and those include OpenR1-Math reasoning traces, so for 900 steps the model was distilled from its unpruned self and trained on worked maths answers in exactly the no-think, step-by-step format GSM8K is scored in. The parent never had that pass. It is a real gain on this task, not a general one: on raw corpus text the pruned model still sits 0.09 nats of cross-entropy above the parent (1.863 vs 1.771), which is the honest cost of removing 30% of the experts. Expect the same pattern elsewhere: strong on instruction-style tasks close to the healing data, slightly weaker on long-tail knowledge.

What to expect

Twelve heal steps in were enough for "hello" → "Hello! How can I help you today?", one-sentence physics, 17+25=42, and two-turn name recall. The pruned model scores lower CE than the parent on chat-formatted text (calibration and healing both contain chat data) and slightly higher on raw corpus text — the honest damage number is the corpus one. Expect small regressions on long-tail knowledge relative to the 35B; experts that fired rarely on the calibration mix are the ones removed.

`next/` — experimental Qwen3.8-Next-format build (work in progress)

next/Qwen3.6-Whittle-25B-A3B-next-EXPERIMENTAL-Q8_0.gguf is the same pruned model converted to the qwen4exp architecture: GDN output gate retrained from silu to sigmoid (progressive, back to front, self-distilled), identity hyper-connections, an inert sparse-attention indexer, no n-gram memory yet. It loads and runs on unmodified upstream llama.cpp as qwen4exp. GSM8K on it: 86.5% through llama.cpp (Q8_0), 82.0% through transformers (bf16) — the gate conversion still costs a few points versus the 92.5% of the silu model above; conversation quality is unchanged. This file is the base the hyper-connections and the n-gram memory are being trained into; treat it as a preview, not a release. next/sigmoid_settle_step1500.pt holds the trainable state that produced it.

Support this work

Whittle runs on one hobbyist's grocery budget and rented GPU hours. If this research is useful to you: ko-fi.com/davida81328 ☕

Authors

David Aylward (logic65) & Claude (Anthropic) — designed, debugged and verified together in one day on a single rented GPU.

Provenance

Parent: Qwen/Qwen3.6-35B-A3B (Apache-2.0). Method references: REAP (Cerebras, arXiv 2510.13999) and "How to Score Experts for One-Shot MoE Expert Pruning" (arXiv 2606.15716). Scripts: logic65/mini-next-a100-kit/colab/ (prune_qwen36.py, heal_qwen36.py, run_h100.sh). Built on one RTX PRO 6000 Blackwell in about five hours. Part of the Whittle project by logic65.

The Whittle-Qwen-3.8-25B-A3B GGUFs above inherit the 35B's lineage: parent logic65/Whittle-Next-27B-A3B, teacher Qwen/Qwen3.8-27B, base lineage Qwen3.6-35B-A3B pruned 256 → 180 experts. All Apache-2.0.

Downloads last month: 1,648

Safetensors

Model size

25B params

Tensor type

BF16

Model tree for logic65/Qwen3.6-Whittle-25B-A3B

Base model

Qwen/Qwen3.6-35B-A3B

Quantized

(863)

this model

Quantizations

2 models

Qwen3.6-Whittle-25B-A3B

☕ Support this work

Status

Whittle-Qwen-3.8-25B-A3B: the 35B body without its memory (3 Oct 2026)

Which file

Measured (3 Oct 2026, on the published Q6_K)

Run it

Caveat

Qwen3.6-Whittle-25B-A3B: the 2 Sep 2026 prune+heal body (different lineage and format)

Which files are which

How it was made

Run it

Why is GSM8K higher than the parent?

What to expect

next/ — experimental Qwen3.8-Next-format build (work in progress)

Support this work

Authors

Provenance

Model tree for logic65/Qwen3.6-Whittle-25B-A3B

`next/` — experimental Qwen3.8-Next-format build (work in progress)