9 days ago
-
Add viewer-compatible benchmark CSV
-
Enable Dataset Viewer for KultuTR benchmark
Adds explicit default/train CSV configuration and a Viewer-compatible CSV export while retaining the original JSONL benchmark. -
Update Qwen3.5 temperature scores in benchmark table
Adds both Qwen3.5-0.8B temperatures to the main results table and clearly distinguishes strict semantic scoring from answer-mention scoring. -
Add Qwen3.5 temperature 0.7 run
Adds the Qwen3.5-0.8B temperature 0.7 predictions and documents a consistent answer-inclusion comparison against the temperature 0.4 run. -
Add Qwen3.5-0.8B KultuTR results
Adds 100 Qwen3.5-0.8B predictions and updates the results table, evaluation notes, Kaggle run record, and model references. -
Add Qwen3-0.6B KultuTR results
Adds the Qwen3-0.6B 100-question prediction file and updates the Turkish model-results table, evaluation notes, and Kaggle run record. -
Update README.md
-
Add KultuTR benchmark and four model evaluations
-
initial commit
