mach-kernel/ecu-pilot-q4km

ecu-pilot (GGUF Q4_K_M)

Quantized GGUF of ecu-pilot-fp16 — a fine-tuned Qwen3.5-35B-A3B for structured tool calling against project metadata via MCP.

Quantization

Source	mach-kernel/ecu-pilot-fp16
Method	Q4_K_M via llama.cpp
Size	~20 GB
Architecture	Mixture of Experts (35B total, 3B active per token)

Usage

Ollama

# Create a Modelfile
echo 'FROM ./ecu-pilot-q4_k_m.gguf
PARAMETER temperature 0.2
PARAMETER num_ctx 8192
PARAMETER stop <|im_end|>' > Modelfile

ollama create ecu-pilot -f Modelfile
ollama run ecu-pilot

llama.cpp

llama-cli -m ecu-pilot-q4_k_m.gguf -ngl 99 -cnv

All variants

Format	Repository	Size
FP16	mach-kernel/ecu-pilot-fp16	~67 GB
GGUF Q4_K_M (this repo)	mach-kernel/ecu-pilot-q4km	~20 GB
GGUF Q8_0	mach-kernel/ecu-pilot-q8_0	~35 GB
LoRA adapter	mach-kernel/ecu-pilot-fp16-lora	~4 GB

Why "ecu"

No reason. Just liked how it sounded. Definitely not a Caesar cipher of anything. Don't look into it.