{ "model_family": "parseq", "model_class": "PARSeq", "framework": "pytorch", "task": "image-to-text", "weights_file": "best.pt", "checkpoint_keys": [ "model_state", "charset", "config", "metrics", "epoch" ], "num_parameters": 95612393, "architecture": { "img_height": 48, "img_width": 512, "patch_size": [ 4, 8 ], "embed_dim": 768, "enc_num_heads": 12, "enc_mlp_ratio": 4, "enc_depth": 12, "dec_num_heads": 12, "dec_mlp_ratio": 4, "dec_depth": 1, "max_label_length": 100, "dropout": 0.1, "decode_ar": true, "refine_iters": 1, "drop_path_rate": 0.05 }, "preprocessing": { "input": "BGR uint8 crop", "resize": "unconditional to (img_width, img_height), aspect ratio not preserved", "interpolation": "INTER_AREA when shrinking, INTER_LINEAR when growing", "channels": "RGB", "scale": "pixel / 255.0", "normalize": { "mean": [ 0.5, 0.5, 0.5 ], "std": [ 0.5, 0.5, 0.5 ] } }, "vocab": { "size": 232, "sha256": "d1b9161ee94eecd37cd643f944e4eb8f02644751ee56a23acac90ad0453ff976", "max_label_length": 100, "token_layout": "[E] at id 0, then the charset in string order, then [B] and [P]; the head emits len(vocab) - 2 logits ([B]/[P] are never predicted)", "charset": "\n !\"#%'()*+,-./0123456789:;<=>?@DFGIJLNQRSUVWYZ[\\]^_bdfghjklmnqrstuvwz{|}~§°±²³·¹º×áýĒřŹžȳʹ́̄̅̈ΓΔΦαδλπφωЁЄІЇАБВГДЕЖЗИЙКЛМНОПРСТУФХЦЧШЩЬЭЮЯабвгдежзийклмнопрстуфхцчшщъыьэюяёєіїўѣҐґ•…‧⁴⁵⁹⁻₀₁₂₃₄₆₉№←↑→↓↔↘↩⇒⇓∆∈∉∗∘∙∛∞∠∥∨∩∪∫≈≠≡≤≥⊆⊕⊗⊥⋅④△♀♂✓⟂" }, "training_val_metrics": { "note": "checkpoint-selection metrics on the training-time validation split; see the model card for the headline evaluation", "epoch": 13, "cer": 0.09525868432708329, "wer": 0.2795733470410589, "accuracy": 0.43191755153029354, "n_samples": 3202.0 }, "base_model": "Hukyl/parseq-b-cyrillic-handwritten" }