Instructions to use AtomicChat/Qwen3.8-Flash-Next-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 AtomicChat/Qwen3.8-Flash-Next-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 AtomicChat/Qwen3.8-Flash-Next-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf AtomicChat/Qwen3.8-Flash-Next-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf AtomicChat/Qwen3.8-Flash-Next-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf AtomicChat/Qwen3.8-Flash-Next-GGUF: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 AtomicChat/Qwen3.8-Flash-Next-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf AtomicChat/Qwen3.8-Flash-Next-GGUF: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 AtomicChat/Qwen3.8-Flash-Next-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf AtomicChat/Qwen3.8-Flash-Next-GGUF:Q4_K_M

Use Docker

docker model run hf.co/AtomicChat/Qwen3.8-Flash-Next-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use AtomicChat/Qwen3.8-Flash-Next-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "AtomicChat/Qwen3.8-Flash-Next-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": "AtomicChat/Qwen3.8-Flash-Next-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/AtomicChat/Qwen3.8-Flash-Next-GGUF:Q4_K_M

Ollama
How to use AtomicChat/Qwen3.8-Flash-Next-GGUF with Ollama:
```
ollama run hf.co/AtomicChat/Qwen3.8-Flash-Next-GGUF:Q4_K_M
```
Unsloth Desktop

How to use AtomicChat/Qwen3.8-Flash-Next-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf AtomicChat/Qwen3.8-Flash-Next-GGUF: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": "AtomicChat/Qwen3.8-Flash-Next-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use AtomicChat/Qwen3.8-Flash-Next-GGUF with Docker Model Runner:
```
docker model run hf.co/AtomicChat/Qwen3.8-Flash-Next-GGUF:Q4_K_M
```

Lemonade

How to use AtomicChat/Qwen3.8-Flash-Next-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull AtomicChat/Qwen3.8-Flash-Next-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Qwen3.8-Flash-Next-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use AtomicChat/Qwen3.8-Flash-Next-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 AtomicChat/Qwen3.8-Flash-Next-GGUF: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 AtomicChat/Qwen3.8-Flash-Next-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use AtomicChat/Qwen3.8-Flash-Next-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf AtomicChat/Qwen3.8-Flash-Next-GGUF: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 "AtomicChat/Qwen3.8-Flash-Next-GGUF: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"

AD-4.27bpw on 6GB VRAM (GTX 1660 Ti): numbers, 32k ladder, hiccups + fixes

#20

by TitanMamo10 - opened 5 days ago

Discussion

TitanMamo10

5 days ago

Running AD-4.27bpw on a 6GB card — numbers + hiccups (ik_llama.cpp)

Field report on Qwen3.8-Flash-Next-AD-4.27bpw-Q4_K_M-M64 (33 shards,
SHA-verified). Box: GTX 1660 Ti 6GB (TU116) / i5-9400F / 32GB DDR4 / NVMe.
Runtime: ik_llama.cpp upstream-main + GGML_CUDA_FORCE_MMQ=ON,
GGML_CUDA_NO_PINNED=1, --fit, prefetch-experts, --defer-ple.

Best configs measured with a prefill ladder (accumulating history,
server-reported prompt_n as ground truth — 2k-prompt benches miss the
transient OOM class entirely):

16k ctx (q8 KV, ub1024, m384): tg 7.31, PP 40.63, gen-after 2.84,
quality probes pass. Sweet spot on 6GB.
48k ctx (q4+had KV, ub128): PP 6.31, sanity/gen fine, quality pass.
Slow but real 48k window.
32k experiments: q6/ub512 → 19k ceiling; q4/ub512 → 23k; q6/ub384 → 27k;
q4/ub384 → 30.8k proven (PP ~17). ub1024 does not fit at 32k.

Hiccups, all diagnosed:

The Q8_0 dense stack (4.6GB on GPU) leaves almost no transient
headroom: long re-prefills OOM from ~6.5k tokens up on the 16k config,
and the fit refuses any margin above 512 at 48k (dense alone ≈ 5.1GB).
Unlike small-dense quants, headroom cannot be bought with fit margin
here — only smaller ub chunks and lighter KV move the ceiling (+4k per
KV step: q8 → q6 → q4).
Same ggml_abort fork wedge after fatals as above — fixed upstream in
ikawrakow/ik_llama.cpp PR #2416 (clean deaths since, verified live).
Same hybrid checkpoint illegal-access (ik #1762 class) — checkpoints off.
q6 KV decode tax ~40% on TU116 (prefill unaffected).

Functional quality probes (math/reasoning/explanation checks) pass at every
rung; the 89.5% top-1 claim feels plausible from the seat, though we did
not re-run your harness. Thanks for the M64 shard layout — PLE isolation
streams cleanly from NVMe.

TitanMamo10

5 days ago

Packaged the AD-4.27 numbers from this thread (16k/48k configs, 32k ladder to 30.8k, dense-headroom analysis) into a public repo:

https://github.com/TitanMamo/q38-6gb

Complete ladder tables with per-rung data, exact flags, harness + guard scripts. MIT licensed, corrections welcome.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment