Instructions to use Achilles1089/fable-coder-35B-A3B-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 Achilles1089/fable-coder-35B-A3B-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 Achilles1089/fable-coder-35B-A3B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Achilles1089/fable-coder-35B-A3B-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 Achilles1089/fable-coder-35B-A3B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Achilles1089/fable-coder-35B-A3B-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 Achilles1089/fable-coder-35B-A3B-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf Achilles1089/fable-coder-35B-A3B-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 Achilles1089/fable-coder-35B-A3B-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Achilles1089/fable-coder-35B-A3B-GGUF:Q4_K_M

Use Docker

docker model run hf.co/Achilles1089/fable-coder-35B-A3B-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use Achilles1089/fable-coder-35B-A3B-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Achilles1089/fable-coder-35B-A3B-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": "Achilles1089/fable-coder-35B-A3B-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Achilles1089/fable-coder-35B-A3B-GGUF:Q4_K_M

Ollama
How to use Achilles1089/fable-coder-35B-A3B-GGUF with Ollama:
```
ollama run hf.co/Achilles1089/fable-coder-35B-A3B-GGUF:Q4_K_M
```

Unsloth Studio

How to use Achilles1089/fable-coder-35B-A3B-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 Achilles1089/fable-coder-35B-A3B-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 Achilles1089/fable-coder-35B-A3B-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Achilles1089/fable-coder-35B-A3B-GGUF to start chatting

How to use Achilles1089/fable-coder-35B-A3B-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Achilles1089/fable-coder-35B-A3B-GGUF:Q4_K_M

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": "Achilles1089/fable-coder-35B-A3B-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

OpenClaw new

How to use Achilles1089/fable-coder-35B-A3B-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Achilles1089/fable-coder-35B-A3B-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 "Achilles1089/fable-coder-35B-A3B-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"

Docker Model Runner
How to use Achilles1089/fable-coder-35B-A3B-GGUF with Docker Model Runner:
```
docker model run hf.co/Achilles1089/fable-coder-35B-A3B-GGUF:Q4_K_M
```

Lemonade

How to use Achilles1089/fable-coder-35B-A3B-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Achilles1089/fable-coder-35B-A3B-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.fable-coder-35B-A3B-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use Achilles1089/fable-coder-35B-A3B-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 Achilles1089/fable-coder-35B-A3B-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 Achilles1089/fable-coder-35B-A3B-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

Q4_K_M fails to load: missing tensor 'blk.40.ssm_conv1d.weight' -- is this a known issue or something on my end?

by G-R-A-V-I-T-Y - opened 10 days ago

Discussion

G-R-A-V-I-T-Y

10 days ago

Trying to run fable-coder-35B-A3B-Q4_K_M.gguf and it fails to load with:

llama_model_loader: loading model tensors, this can take a while... (mmap = true, direct_io = false)
llama_model_load: error loading model: missing tensor 'blk.40.ssm_conv1d.weight'
llama_model_load_from_file_impl: failed to load model

Setup:

llama-server build b9075-4f331667d
--n-gpu-layers 999 --n-cpu-moe 32 (same flags I use successfully for other Qwen3.6-35B-A3B GGUFs)
Downloaded via hf download, verified the local file's SHA256 matches the repo's LFS hash exactly (f88b25416bff442627fcbcc04d3191b83922ac394575c66f3c0813bc6fe06b92), so it's not a corrupted/incomplete download.

For context: I run the base Qwen3.6-35B-A3B architecture (hybrid MoE + SSM/Gated-DeltaNet layers) all the time on this exact same llama-server build with no issues, so my binary does have working hybrid-SSM support. This looks like layer 40's ssm_conv1d weight may have been dropped somewhere in the LoRA merge / GGUF conversion pipeline for this specific model.

Is this:

Something I'm doing wrong on my end (missing flag, wrong build, etc.)?
A known issue with this repo's conversion that's already being looked at?
A new one for you to dig into?

Happy to share more logs/repro details if useful. Thanks for the model!

Achilles1089

Owner 10 days ago

Hey my man i appreciate the question

Achilles1089

Owner 10 days ago

On the missing-tensor theory: blk.40 is the MTP / next-token-prediction block, and it's attention-style by design — the stock Qwen3.6-35B-A3B GGUF has no ssm_conv1d there either. I read the tensor index straight out of the uploaded Q4_K_M and diffed it against the base:

	base Qwen3.6-35B-A3B	this repo's Q4_K_M
total tensors	753	753
`blk.40` tensors	20	20, identical names
`blk.40.ssm_conv1d`	absent	absent
`block_count`	41	41
`nextn_predict_layers`	1	1
SSM blocks	30 (`0,1,2,4,5,6,…`)	30, same pattern

The hybrid pattern puts full-attention layers at blocks 3, 7, 11 … 39, with block 40 as the MTP block on top of the 40 real layers. So nothing was dropped in the LoRA merge or the GGUF conversion — a reasonable hypothesis, just not what happened. Build 9075 appears to type block 40 as a regular hybrid layer and go looking for ssm_conv1d; the newer build handles it.

That also explains why the base works for you: most base GGUF uploads ship with the MTP block stripped (block_count=40, nextn_predict_layers=0), so there's no block 40 to mis-type. This repo keeps it, which needs a newer loader. If you want to confirm, on the base that works for you:

python -c "from gguf import GGUFReader as G; r=G('base.gguf'); \
print({k: v.contents() for k, v in r.fields.items() if 'block_count' in k or 'nextn' in k})"

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