langleu/qmd-query-expansion-lfm2.5-1.2b-base-v2

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.2BQuant:BF16Context Size:32kPublished:Jul 25, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

The langleu/qmd-query-expansion-lfm2.5-1.2b-base-v2 is a 1.2 billion parameter language model, fine-tuned from LiquidAI/LFM2.5-1.2B-Base using LoRA. It is specifically optimized for QMD (Query-Managed Documents) query expansion, leveraging a v2 data recipe. This model excels at generating expanded search queries, supporting directives like /only:lex, /only:vec, and /only:hyde, and achieves a 95.17% average QMD reward.

Loading preview...

Overview

This model, langleu/qmd-query-expansion-lfm2.5-1.2b-base-v2, is a 1.2 billion parameter language model derived from LiquidAI/LFM2.5-1.2B-Base. It has been fine-tuned with LoRA specifically for QMD query expansion using a v2 data recipe. The repository provides both a merged BF16 Transformers checkpoint and QMD-ready GGUF quantizations (Q8_0, Q5_K_M, Q4_K_M).

Key Capabilities

  • Specialized Query Expansion: Designed to expand search queries for QMD, supporting specific directives such as /only:lex, /only:vec, and /only:hyde.
  • High Performance: Achieves an average QMD reward of 95.17% and 100% format compliance, with strong entity preservation (98.31%).
  • GGUF Quantizations: Offers various GGUF formats for efficient deployment with llama.cpp and QMD, with the q5_k_m variant showing a slight performance improvement over the BF16 baseline.
  • Prompt Adherence: Follows a specific prompt structure (/no_think Expand this search query: {query}) and emits lines prefixed with lex:, vec:, or hyde:.

Good For

  • Developers and systems requiring accurate and structured query expansion within a QMD framework.
  • Applications needing a compact yet performant model for information retrieval and search enhancement.
  • Integration into systems utilizing llama.cpp for efficient inference with its provided GGUF quantizations.