Modelpublic

Pimmetjeoss/tribe-crm-qwen-2b-gguf

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face

1likes14downloads

Model Card

Tribe CRM Agent — Qwen3.5-2B GGUF

GGUF quantized version of `Pimmetjeoss/tribe-crm-qwen-2b-merged` for use with llama.cpp, Ollama, LM Studio, and other GGUF-compatible inference engines.

Files

File	Size	Quantization	Use case
`qwen-tribe-crm-q4_k_m.gguf`	1.2 GB	Q4KM	Recommended for most users (good quality/size balance)

Performance

Measured on the bf16 merged model (GGUF quantization is near-lossless for Q4KM):

Metric	Synthetic heldout	Realistic heldout
Combined accuracy	91.2% (52/57)	98.2% (56/57)

Ollama usage

bash

# Download the GGUF
huggingface-cli download Pimmetjeoss/tribe-crm-qwen-2b-gguf qwen-tribe-crm-q4_k_m.gguf --local-dir ./

# Create a Modelfile
cat > Modelfile <<EOF
FROM ./qwen-tribe-crm-q4_k_m.gguf
TEMPLATE """{{ if .System }}<|im_start|>system
{{ .System }}<|im_end|>
{{ end }}{{ if .Prompt }}<|im_start|>user
{{ .Prompt }}<|im_end|>
<|im_start|>assistant
{{ end }}{{ .Response }}<|im_end|>
"""
PARAMETER temperature 0
EOF

ollama create tribe-crm-qwen -f Modelfile
ollama run tribe-crm-qwen "Zoek Bakkerij De Wit op"

llama.cpp usage

bash

./llama-cli -m qwen-tribe-crm-q4_k_m.gguf \
    --chat-template chatml \
    -p "Zoek Bakkerij De Wit op"

Expected output format

Qwen native XML-style tool calls:

<tool_call>
<function=zoek_contact>
<parameter=naam>
Bakkerij De Wit
</parameter>
</function>
</tool_call>

A Python parser for this format is in agent/parser.py.