On-device document-summarization engine for the PaperSummary app. A supervised fine-tune (LoRA) of Qwen2.5-1.5B-Instruct, distilled from Qwen2.5-72B-Instruct, aligned to the app output contract: structured Title / Description / Summary markers, faithful markdown, level/format control, 29 languages, and map-reduce merge for long documents.

Format: GGUF, Q4_K_M (runs via llama.cpp / llama.rn, Metal/GPU), n_ctx 4096
File: papersummary-1.5b-v2-q4_k_m.gguf (986,048,000 bytes)
Improvements over stock Qwen2.5-1.5B (measured): output-marker compliance 71%→100%, cross-lingual output language obeyed, map-reduce merge without hallucinated figures.
License: Apache-2.0 (inherits Qwen2.5). Fine-tune data generated synthetically.

Intended as a drop-in engine download for the app; not a general chat model.

Probees1492/papersummary-1.5b-v2-gguf

PaperSummary 1.5B (fine-tuned, Q4_K_M GGUF)

Probees1492/
papersummary-1.5b-v2-gguf