Instructions to use PollardWeights/Qwen3-30B-A3B-Pollard with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use PollardWeights/Qwen3-30B-A3B-Pollard with Trellis:

# No code snippets available yet for this library.

# To use this model, check the repository files and the library's documentation.

# Want to help? PRs adding snippets are welcome at:
# https://github.com/huggingface/huggingface.js

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use PollardWeights/Qwen3-30B-A3B-Pollard 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 PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S
# Run inference directly in the terminal:
llama cli -hf PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S
# Run inference directly in the terminal:
llama cli -hf PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S

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 PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S
# Run inference directly in the terminal:
./llama-cli -hf PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S

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 PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S
# Run inference directly in the terminal:
./build/bin/llama-cli -hf PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S

Use Docker

docker model run hf.co/PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S

LM Studio
Jan

vLLM

How to use PollardWeights/Qwen3-30B-A3B-Pollard with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "PollardWeights/Qwen3-30B-A3B-Pollard"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "PollardWeights/Qwen3-30B-A3B-Pollard",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S

Ollama
How to use PollardWeights/Qwen3-30B-A3B-Pollard with Ollama:
```
ollama run hf.co/PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S
```
Unsloth Desktop

How to use PollardWeights/Qwen3-30B-A3B-Pollard with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S

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": "PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use PollardWeights/Qwen3-30B-A3B-Pollard with Docker Model Runner:
```
docker model run hf.co/PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S
```

Lemonade

How to use PollardWeights/Qwen3-30B-A3B-Pollard with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S

Run and chat with the model

lemonade run user.Qwen3-30B-A3B-Pollard-IQ3_S

List all available models

lemonade list

Hermes Agent

How to use PollardWeights/Qwen3-30B-A3B-Pollard with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S

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 PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use PollardWeights/Qwen3-30B-A3B-Pollard with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S

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 "PollardWeights/Qwen3-30B-A3B-Pollard:IQ3_S" \
  --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"

Qwen3-30B-A3B — Pollard

Pollard shrank this model: 56.9 GB (f16) → 7.02 GB — 88% smaller, 8.1× down, and under half the size of NVFP4 (~16.2 GB).

The 1-bit-class MoE flagship (IQ1_KT mixed-precision), still beating uniform 1-bit. Want more quality? The IQ3_S / IQ4_XS / Q6_K rungs below are larger, higher-fidelity options.

format this model's size

f16 56.9 GB

Q8_0 ~32.4 GB

Q6_K ~23.4 GB

Q4_K_M / NVFP4 ~18.5 / ~16.2 GB

PollardMix (this repo's IQ1_KT) 7.02 GB

format	this model's size
f16	56.9 GB
Q8_0	~32.4 GB
Q6_K	~23.4 GB
Q4_K_M / NVFP4	~18.5 / ~16.2 GB
PollardMix (this repo's IQ1_KT)	7.02 GB

A memory-fit, mixed-precision build of Qwen3-30B-A3B (a 30.5B-param MoE, 3.3B active, 128 experts) made with Pollard Weights. The cold expert bulk (ffn_gate/up_exps) is crushed to 1-bit trellis (IQ1_KT); the router, the residual writer (ffn_down_exps), the shared experts, attention, and the first/last blocks are protected. It sits in the 1-bit size class but reads closer to 2-bit.

This is a standard GGUF — the trellis *_KT files need ik_llama.cpp; the K-quants (Q6_K) run in stock llama.cpp / Ollama / LM Studio.

The numbers (WikiText-2 raw, ctx 2048, 145 chunks; KLD vs Q6_K base)

build	role	PPL	size	bpw	Mean KLD	Median KLD	top-1
uniform IQ2_KT	2-bit ceiling	7.28	8.34 GB	2.19	0.134	0.059	84.81%
PollardMix	this model	8.57	7.02 GB	1.84	0.310	0.140	77.81%
uniform IQ1_KT	1-bit baseline	9.01	6.57 GB	1.73	0.360	0.174	75.47%

PollardMix beats the uniform 1-bit trellis quant on every metric — PPL −4.9%, Mean KLD −14%, Median KLD −20%, top-1 +2.3 pts — at +6.9% size, under the 2-bit ceiling. Same clean sweep as the 7B/14B dense cards, now reproduced on a MoE — the automap policy generalizes (crush cold experts, protect the router / ffn_down_exps / shared experts / attention).

Allocation (the surgery)

tensor role	atom
cold expert bulk (`ffn_gate/up_exps`)	`IQ1_KT`	crushed
`ffn_down_exps` (residual writer)	`IQ2_KT`	protected
`ffn_gate_inp` (router)	`Q6_K`	kept high
shared experts	`IQ2_KT` / `IQ3_KT`	protected
attention q, output	`IQ2_KT`	protected
attention k, v	`IQ1_KT`	crushed
first-2 / last-2 blocks	`IQ2_KT`	protected
token embeddings / output head	`Q4_K` / `Q6_K`	kept

Available files

file	size	notes
`…-Q6_K.gguf`	23.4 GB	near-lossless
`…-IQ4_XS.gguf`	~15 GB	recommended default
`…-IQ3_S.gguf`	~13 GB	smaller
`…-IQ1_KT.gguf`	7.02 GB	flagship — 1-bit mixed-precision MoE trellis

Usage

# ik_llama.cpp (for the IQ1_KT trellis flagship)
llama-cli -m Qwen3-30B-A3B-Pollard-IQ1_KT.gguf -p "Explain why the sky is blue." --temp 0.7 --repeat-penalty 1.15

# Ollama (Q6_K / K-quant rungs)
ollama run hf.co/PollardWeights/Qwen3-30B-A3B-Pollard

Errata (honest scope)

*_KT are ik_llama.cpp trellis quants; build ik_llama.cpp for them (they load in stock llama.cpp too). Q6_K runs anywhere.
Chat at the 1-bit tier benefits from --repeat-penalty 1.15.
MoE imatrix note: ffn_gate_exps shares its input with ffn_up_exps (SwiGLU), so its importance was taken from up_exps for full expert coverage.
Single machine (32 GB RAM / 16 GB GPU); replication invited — pollard --gguf model.gguf --benchmark reproduces the 3-bar board.

Built with Pollard Weights — frontier models, small hardware, no compromise.

Downloads last month: -

GGUF

Model size

31B params

Architecture

qwen3moe

Hardware compatibility

3-bit

4-bit

6-bit

View +1 variant

Model tree for PollardWeights/Qwen3-30B-A3B-Pollard

Base model

Qwen/Qwen3-30B-A3B-Base

Finetuned

Qwen/Qwen3-30B-A3B

Quantized

(145)

this model

PollardWeights
/

Qwen3-30B-A3B-Pollard

Qwen3-30B-A3B — Pollard

Pollard shrank this model: 56.9 GB (f16) → 7.02 GB — 88% smaller, 8.1× down, and under half the size of NVFP4 (~16.2 GB).

The numbers (WikiText-2 raw, ctx 2048, 145 chunks; KLD vs Q6_K base)

Allocation (the surgery)

Available files

Usage

Errata (honest scope)

Model tree for PollardWeights/Qwen3-30B-A3B-Pollard