Papers and technical reports from the Ertas AI team, published with the code and models behind them.
QMSum ships without a scorer, so we rescored 15 systems under one implementation. A 406M specialist retrained on the same retrieved transcript spans as our 1.2B system scored 36.33 ROUGE-1 against 35.41, a difference QMSum cannot statistically resolve, while using about a third of the parameters and under half the peak inference memory.
arXiv · GitHub · Hugging Face model
Code, models and data released with our research, free to inspect and reuse under each repository's own terms.
Hugging Face model
ErtasAI/qmsum-summarizer-lfm2.5-1.2b-loraLoRA adapter for LFM2.5-1.2B, the 1.2B summarizer
Hugging Face model
ErtasAI/qmsum-summarizer-segenc-406m-spans406M Fusion-in-Decoder summarizer retrained on retrieved spans
Hugging Face model
ErtasAI/qmsum-locator-minilm-l12-w375Cross-encoder locator over 375-word windows, the promoted configuration
Hugging Face model
ErtasAI/qmsum-locator-minilm-l6-w900Cross-encoder locator over 900-word windows, the protocol-exact configuration