Instructions to use mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF with Transformers:

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-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 mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-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 mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-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 mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-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 mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF:Q4_K_M

Use Docker

docker model run hf.co/mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF:Q4_K_M

LM Studio
Jan
Ollama
How to use mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF with Ollama:
```
ollama run hf.co/mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF:Q4_K_M
```
Unsloth Desktop

How to use mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-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": "mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF with Docker Model Runner:
```
docker model run hf.co/mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF:Q4_K_M
```

Lemonade

How to use mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.qwen3.8-flash-coder-85gb-bf16-i1-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-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 mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-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 mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-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 "mradermacher/qwen3.8-flash-coder-85gb-bf16-i1-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"

Seems to have a bug and is generally not strong

by Mike0365 - opened 3 days ago

Discussion

Mike0365

3 days ago

Hi folks,

I tested "mradermacher/qwen3.8-flash-coder-85gb-bf16-GGUF" Q8_0 (45.4GB) today.

It seems to have a bug:
With terse instructions like "answer only with..." (reproduced in both English and German), the model dumps its entire response into the reasoning_content channel and leaves content empty. Any standard harness reading only content sees a blank answer. Needed to patch my benchmark scripts with a fallback to reasoning_content to measure the model's actual capability rather than this problem.

Test Results:
Light coding suite: 17/30 - where other models based on qwen3.6 had 30/30
Hard coding suite: 4/32 - worse than EVERY other model I have tested in 2026
Trap prompts : 0/2 -
Currency: wrong conversion chain and the arithmetic itself was wrong.
Elevator riddle: completely derailed — treated a classic lateral-thinking puzzle as a binary-search algorithm problem

The thin post-pruning recalibration (308 steps) appears to have damaged both output-channel discipline and general reasoning quality.
The model's self-reported benchmark claims (67% Pass@1) don't hold up.

Sorry for no better feedback. Would have loved to have a strong qwen3.8 model
Regards, Mike

RichardErkhov

3 days ago

Would have loved to have a strong qwen3.8 model

well, try another model then =)

and I cant guarrantee the quality, since you know, we are not the author, we are quanting =)

Leonerdo60

3 days ago

I asked qwen3.8-flash-coder-85gb-i1 to do the code review of some of my scripts using Hermes Agent. Unfortunately, it falls in endless loop while thinking. Maybe I'm doing something wrong, but now I'm not happy with it.

RichardErkhov

3 days ago

•

edited 3 days ago

Not sure guys, it might simply be llamacpp not being sure how to handle this model and routing through nullified experts. Im not sure, maybe that's the reason, since the architecture was modified (removed like 2/3 of experts), so quite interesting it was even been able to quant. Could anyone try to check the temperature and other settings? Maybe setup needs to be different on inference stage

redaihf

3 days ago

It may need some finetuning first.

kapha

3 days ago

I asked qwen3.8-flash-coder-85gb-i1 to do the code review of some of my scripts using Hermes Agent. Unfortunately, it falls in endless loop while thinking. Maybe I'm doing something wrong, but now I'm not happy with it.

Try another coding agent like Pi or whatever.
Then pass the results to Claude or ChatGPT. Let the LLM figure whats wrong. Then learn from it.

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