GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
Silviase
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
Silviase
Model Tree
Qwen/Qwen3.5-2B
Input Modalities
Output Modalities
Supported Functionality
Full-image OCR uses CC-OCR character P/R/F1 and clipped order-free CER (image means), with NED as a diagnostic. Training set-F1 rewards are not full-image OCR evaluation. Paper metric version: jawildtext-metrics v2 (2026-09-28). Any earlier external exact/soft set-F1 report is superseded by this evaluation.
{ "dev": { "qa": { "n": 500, "mean_reward": 0.106 }, "evidence": { "n": 500, "mean_reward": 0.4219273875771224 }, "ocr": { "n": 500, "mean_reward": 0.808685788053564 } }, "ocr": { "metrics_version": "jawildtext-metrics v2 (2026-09-28)", "images": 580, "capped": 46, "empty": 0, "precision": 0.8812854581378172, "recall": 0.8263092851801185, "f1": 0.8357104020973969, "ned": 0.29902784969843027, "cer": 0.20910484363848766, "character_micro_f1": 0.6171680390888938, "exact_region_recall": 0.429912186962784, "cer_detail": { "clipped_images": 46, "exact_images": 475, "exact_rate": 0.8189655172413793, "mean_gap": 0.0058244401429518285, "median_image_cer": 0.11619964417333842, "reference_characters": 263496, "pooled_cer": null, "substitutions": null, "deletions": null, "insertions": null } }, "qa": { "expected": 1025, "complete": 1025, "capped": 58, "metrics": { "content": { "value": 0.41268292682926827, "ci95": [ 0.3831683168316832, 0.44490034757014024 ], "units": 1025, "images": 745, "excluded_units": 0, "valid_draws": 2000 }, "boxed": { "value": 0.4068292682926829, "ci95": [ 0.3768690220140765, 0.438887063076598 ], "units": 1025, "images": 745, "excluded_units": 0, "valid_draws": 2000 }, "format_valid": { "value": 0.9434146341463414, "ci95": [ 0.9284296879448779, 0.9576771653543307 ], "units": 1025, "images": 745, "excluded_units": 0, "valid_draws": 2000 } }, "scope": "All available outputs for this model; not a paired intersection", "repetitions": 2000, "seed": 42, "unit": "image cluster", "limitation": "Single training seed; intervals exclude training-seed variability" } }