Instructions to use cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0 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 cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0
# Run inference directly in the terminal:
llama cli -hf cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0
# Run inference directly in the terminal:
llama cli -hf cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0

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 cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0
# Run inference directly in the terminal:
./llama-cli -hf cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0

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 cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0
# Run inference directly in the terminal:
./build/bin/llama-cli -hf cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0

Use Docker

docker model run hf.co/cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0

LM Studio
Jan

vLLM

How to use cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0

Ollama
How to use cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0 with Ollama:
```
ollama run hf.co/cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0
```
Unsloth Desktop

How to use cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0

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": "cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0 with Docker Model Runner:
```
docker model run hf.co/cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0
```

Lemonade

How to use cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0

Run and chat with the model

lemonade run user.Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0-Q8_0

List all available models

lemonade list

Hermes Agent

How to use cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0

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 cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0

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 "cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0:Q8_0" \
  --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-Flash-Next — 5.33 BPW PLEQ8_0

Download

Model Details

Property	Value
File size	109.80 GiB
Total BPW	5.330
Embedding	Q8_0
BPW without Embedding	3.9936
MoE avg BPW	3.8403

Tensor Distribution *

* excluding Non-MoE and n-gram embedding

Recommended llama.cpp configurations

Tested configurations for 8 logical cores, 64 GB of RAM and 12 GB of VRAM:

Headless, full context:

./llama-server \
  -fit off \
  -t 8 \
  -np 1 \
  -lzm on \
  -ctk q8_0 \
  -ctv q8_0 \
  -ngl 99 \
  -cmoe \
  -c 0 \
  --reasoning-preserve \
  --reasoning-effort xhigh \
  --temp 0.6 \
  --top-k 20 \
  --top-p 0.95 \
  --min-p 0.1 \
  --presence-penalty 0.0 \
  -m Qwen3.8-Flash-Next.gguf

Desktop, with vision but half context to save some ram for the desktop environment:

./llama-server \
  -fit off \
  -t 8 \
  -np 1 \
  -lzm on \
  -b 512 \
  -ub 512 \
  -kvu \
  -ctk q8_0 \
  -ctv q8_0 \
  -ngl 99 \
  -ncmoe 46 \
  -cram 0 \
  --ctx-checkpoints 8 \
  --checkpoint-min-step 1024 \
  -c 131072 \
  --reasoning-preserve \
  --reasoning-effort xhigh \
  --temp 0.6 \
  --top-k 20 \
  --top-p 0.95 \
  --min-p 0.1 \
  --presence-penalty 0.0 \
  --image-min-tokens = 1024 \
  --no-mmproj-offload \
  --mmproj mmproj-Qwen3.8-Flash-Next-Q8_0.gguf \
  -m Qwen3.8-Flash-Next.gguf

Imatrix

imatrix_unsloth.gguf_file

Quantization Recipe

^per_layer_token_embd\.weight$=q8_0
^blk\.\d+\.attn_k_norm\.weight$=f32
^blk\.\d+\.attn_q_norm\.weight$=f32
^blk\.\d+\.ffn_gate_inp\.weight$=f32
^blk\.\d+\.ffn_gate_inp_shexp\.weight$=f32
^blk\.\d+\.ssm_a$=f32
^blk\.\d+\.ssm_conv1d\.weight$=f32
^blk\.\d+\.ssm_dt\.bias$=f32
^blk\.\d+\.ssm_norm\.weight$=f32
^blk\.\d+\.ssm_alpha\.weight$=f32
^blk\.\d+\.ssm_beta\.weight$=f32
^blk\.\d+\.hc_attn_norm\.weight$=f32
^blk\.\d+\.hc_ffn_norm\.weight$=f32
^blk\.\d+\.hc_attn_inject\.weight$=f32
^blk\.\d+\.hc_ffn_inject\.weight$=f32
^blk\.\d+\.indexer\.q_norm\.weight$=f32
^blk\.\d+\.indexer\.k_norm\.weight$=f32
^blk\.\d+\.ple_norm_conv\.weight$=f32
^blk\.\d+\.ple_norm_key\.weight$=f32
^blk\.\d+\.ple_norm_query\.weight$=f32
^blk\.\d+\.ple_conv1d\.weight$=f32
^output_hc_norm\.weight$=f32
^blk\.\d+\.ffn_down_exps\.weight$=iq4_nl
^blk\.47\.ffn_up_exps\.weight$=q4_k
^blk\.(1|2|3|4|5|43|44|45|46)\.ffn_up_exps\.weight$=iq4_xs
^blk\.(0|31|32|33|34|35|36|37|38|39|40|41|42)\.ffn_up_exps\.weight$=iq3_xxs
^blk\.(6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30)\.ffn_up_exps\.weight$=iq3_s
^blk\.47\.ffn_gate_exps\.weight$=q4_k
^blk\.(1|2|3|4|5|43|44|45|46)\.ffn_gate_exps\.weight$=iq4_xs
^blk\.(0|31|32|33|34|35|36|37|38|39|40|41|42)\.ffn_gate_exps\.weight$=iq3_xxs
^blk\.(6|7|8|9|10|11|12|13|14|15|16|17|18|19|20|21|22|23|24|25|26|27|28|29|30)\.ffn_gate_exps\.weight$=iq3_s

Credits to:

Thireus since it's taking inspiration from Qwen3.5-122B-A10B's recipe.
ggml-org for the mmproj and llama.cpp .
Unsloth for the BF16 and the imatrix.

Downloads last month: 1,345

GGUF

Model size

177B params

Architecture

qwen4exp

Hardware compatibility

We're not able to determine the quantization variants.

View all variants

Model tree for cmh/Qwen3.8-Flash-Next-5.33bpw-PLEQ8_0

Base model

Qwen/Qwen3.8-Flash-Next

Quantized

(250)

this model