maft-foundation/qmd-query-expansion-1.7B-gguf

QMD Query Expansion 1.7B (GGUF)

Fine-tuned Qwen3-1.7B for search query expansion in QMD (Query Markup Documents).

Model Details

• Base model: Qwen/Qwen3-1.7B
• Fine-tuning: SFT on query expansion pairs using MLX on Apple Silicon
• Format: GGUF (Q4_K_M quantization)
• Size: ~1GB
• Use case: Expanding short search queries into richer search terms for hybrid retrieval

Usage

This model is used by QMD's hybrid search pipeline. Given a short query, it generates expanded terms including lexical keywords, semantic variations, and hypothetical document excerpts (HyDE).

Input: "backup strategy"
Output:
/lex: backup restore restic incremental snapshot retention
/sem: data protection disaster recovery redundancy
/hyde: A comprehensive backup strategy includes regular incremental snapshots...

Training

• Fine-tuned with MLX on Apple Silicon (M4)
• SFT dataset: curated query-expansion pairs
• Exported: MLX → dequantize FP16 → llama.cpp convert → GGUF Q4_K_M

Files

File	Quant	Size	Description
qmd-query-expansion-1.7B-q4_k_m.gguf	Q4_K_M	~1GB	Recommended for production

Configuration

Set QMD_GENERATE_MODEL environment variable:

export QMD_GENERATE_MODEL="hf:maft-foundation/qmd-query-expansion-1.7B-gguf/qmd-query-expansion-1.7B-q4_k_m.gguf"