exact_match float64 0 0 | nem float64 0 0 | token_f1 float64 0.75 1.25 | bleu1 float64 0.44 0.7 | bleu4 float64 0.07 0.16 | rouge_l float64 0.81 1.3 | n_samples int64 100 400 | CAA float64 0 0 | MES float64 0 0.01 | RCS float64 0.5 0.52 | M3VL float64 0.1 0.1 | samples_done int64 100 400 | phase stringclasses 1
value |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
0 | 0 | 0.75 | 0.44 | 0.07 | 0.81 | 100 | 0 | 0 | 0.515 | 0.103 | 100 | baseline |
0 | 0 | 1.15 | 0.64 | 0.14 | 1.18 | 200 | 0 | 0.005 | 0.505 | 0.1025 | 200 | baseline |
0 | 0 | 1.25 | 0.7 | 0.16 | 1.3 | 300 | 0 | 0.0067 | 0.5017 | 0.1023 | 300 | baseline |
0 | 0 | 1.13 | 0.63 | 0.14 | 1.15 | 400 | 0 | 0.005 | 0.5025 | 0.102 | 400 | baseline |
VLM Evaluation Predictions
Prediction outputs from the Zeroshot-DeepVision-24k Bengali Math VQA evaluation pipeline.
Structure
{model_short}/
baseline-testing/ ← predictions BEFORE fine-tuning
post-finetune-testing/ ← predictions AFTER fine-tuning
Generated by the reusable Kaggle evaluation notebook.
- Downloads last month
- 4