developmentseed/gazet-model

Gazet Model

LoRA-finetuned Qwen3.5-0.8B for natural-language geocoding over Overture Maps and Natural Earth parquet datasets.

Two tasks:

• Place extraction: Given a user query, extract structured place names with optional country codes and subtypes
• Text-to-SQL: Given a user query and fuzzy-matched candidate entities, generate a DuckDB spatial SQL query

Files

File	Description
ckpt-001.gguf	Q8_0 quantized GGUF (812 MB), ready for llama-server
merged/	Full merged safetensors (for re-quantization or further finetuning)

Usage

Serve with llama-server:

# Download
hf download developmentseed/gazet-model ckpt-001.gguf

# Serve
llama-server -m ckpt-001.gguf --port 9000 --ctx-size 2048

The model exposes /v1/chat/completions on port 9000.

Or use with the full gazet stack via Docker Compose (see gazet repo).

Training

Base model: unsloth/Qwen3.5-0.8B
Method: LoRA (r=16, alpha=32) via Unsloth
Data: developmentseed/gazet-dataset
Hardware: Single A100-80GB on Modal (~3.5 hrs/epoch)
Optimizer: AdamW 8-bit, lr=1e-4, linear schedule
Max sequence length: 2048
Loss: Train on assistant responses only (Unsloth train_on_responses_only)
Full training code: github.com/developmentseed/gazet