Instructions to use vcruz305/Qwen3.8-2.4T-A95B-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use vcruz305/Qwen3.8-2.4T-A95B-GGUF 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 vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_S
# Run inference directly in the terminal:
llama cli -hf vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_S

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_S
# Run inference directly in the terminal:
llama cli -hf vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_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 vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_S
# Run inference directly in the terminal:
./llama-cli -hf vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_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 vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_S
# Run inference directly in the terminal:
./build/bin/llama-cli -hf vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_S

Use Docker

docker model run hf.co/vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_S

LM Studio
Jan

vLLM

How to use vcruz305/Qwen3.8-2.4T-A95B-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "vcruz305/Qwen3.8-2.4T-A95B-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "vcruz305/Qwen3.8-2.4T-A95B-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_S

Ollama
How to use vcruz305/Qwen3.8-2.4T-A95B-GGUF with Ollama:
```
ollama run hf.co/vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_S
```

Unsloth Studio

How to use vcruz305/Qwen3.8-2.4T-A95B-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for vcruz305/Qwen3.8-2.4T-A95B-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for vcruz305/Qwen3.8-2.4T-A95B-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for vcruz305/Qwen3.8-2.4T-A95B-GGUF to start chatting

How to use vcruz305/Qwen3.8-2.4T-A95B-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_S

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_S"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use vcruz305/Qwen3.8-2.4T-A95B-GGUF with Docker Model Runner:
```
docker model run hf.co/vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_S
```

Lemonade

How to use vcruz305/Qwen3.8-2.4T-A95B-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_S

Run and chat with the model

lemonade run user.Qwen3.8-2.4T-A95B-GGUF-IQ1_S

List all available models

lemonade list

Hermes Agent

How to use vcruz305/Qwen3.8-2.4T-A95B-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_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 vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_S

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use vcruz305/Qwen3.8-2.4T-A95B-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_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 "vcruz305/Qwen3.8-2.4T-A95B-GGUF:IQ1_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"

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Qwen3.8-2.4T-A95B GGUF

Qwen3.8-2.4T-A95B is Qwen's open 2.4T MoE (95B active). 92 layers, 512 routed experts plus one shared expert, ten routed experts per token.

These files are llama.cpp GGUFs of that model, cut to fit a small DGX Spark cluster (128 GB unified memory per box). The k=1024 file is 265 GiB, which is the 3-Spark size. The k=1536 file is 373 GiB and is meant for 4 Sparks. A 3-Spark split of 373 GiB is about 124 GiB per rank and does not leave enough headroom.

What `k` means

k is the FFN width inside each routed expert (n_ff_exp), not the number of experts. Official width is 2048.

File	k	keep	Bytes	GiB
`Qwen3.8-2.4T-A95B-k1024-IQ1_S-Q8.gguf`	1024	50%	284,737,696,224	265.18
`Qwen3.8-2.4T-A95B-k1536-IQ1_S-Q8.gguf`	1536	75%	400,500,486,624	373.01

The router still has all 512 experts. The shared expert is still 2048 wide. Routed expert weights are IQ1_S. Shared expert, attention / Gated DeltaNet, router, and embeddings are Q8_0 / F32. No MTP.

Versus Unsloth's 1-bit packs

Same as these files: Unsloth did not remove experts. unsloth/Qwen3.8-2.4T-A95B-GGUF UD-IQ1_S and UD-Q1_0 still list expert_count=512 and expert_used_count=10. The router is intact.

They also left expert width at the stock 2048 (expert_feed_forward_length=2048). The size cut is Dynamic 2.0 1-bit on that full-width model: UD-Q1_0 is 370 GiB, UD-IQ1_S is 473 GiB.

This repo keeps the same 512 experts and instead narrows each expert's FFN (k), with IQ1_S on the routed tensors and Q8_0 on the always-on trunk. That is why k=1024 is smaller than Unsloth's smallest file without dropping anyone from the committee. There is no head-to-head bench in this card.

Download

Use hf_xet. Do not git clone.

export HF_XET_HIGH_PERFORMANCE=1
hf download vcruz305/Qwen3.8-2.4T-A95B-GGUF \
  --local-dir Qwen3.8-2.4T-A95B-GGUF \
  --include "Qwen3.8-2.4T-A95B-k1536-IQ1_S-Q8.gguf"

Change --include for the k you want.

Run

Needs llama.cpp with qwen35moe. These files will not fully offload a 48 GB GPU.

llama-cli \
  -m Qwen3.8-2.4T-A95B-k1536-IQ1_S-Q8.gguf \
  -ngl 0 -c 4096 -n 64 --temp 0 \
  --jinja --reasoning-format deepseek \
  -p "Say OK."

License

Same terms as the official Qwen3.8-Max / 2.4T-A95B card (license: other, license_name: qwen3.8-max). See Qwen/Qwen3.8-2.4T-A95B-FP8.

Contact

Victor Cruz (vcruz305).

Downloads last month: 29

GGUF

Model size

1.2T params

Architecture

qwen35moe

Hardware compatibility

1-bit

Model tree for vcruz305/Qwen3.8-2.4T-A95B-GGUF

Base model

Qwen/Qwen3.8-2.4T-A95B

Quantized

Qwen/Qwen3.8-2.4T-A95B-FP8

Quantized

(3)

this model

You need to agree to share your contact information to access this model

Qwen3.8-2.4T-A95B GGUF

What k means

Versus Unsloth's 1-bit packs

Download

Run

License

Contact

Model tree for vcruz305/Qwen3.8-2.4T-A95B-GGUF

What `k` means