{ "model_id": "MatterWaveAI/grt-30m-image", "architecture": "GRTImageForConditionalGeneration", "base_model": "MatterWaveAI/grt-30m", "optimizer": "HuygensAdam (Circular Order Parameter R_p Splay Torque) (#1 RVQ Val Loss in 8-Way Ablation)", "total_parameters": 29371239, "language_parameters": 29305498, "image_head_parameters": 65741, "rvq_codec_parameters": 32772, "num_rvq_codebooks": 4, "rvq_codebook_size": 64, "nn_linear_layers": 0, "nn_conv2d_layers": 0, "image_size": 224, "patch_size": 14, "spatial_merge_size": 2, "patches_per_token": 4, "colors": [ "red", "blue", "green", "gold", "cyan", "magenta" ], "shapes": [ "circle", "square", "triangle", "ring", "stripe", "cross" ], "layouts": [ "top-left", "center", "bottom-right", "diagonal" ], "num_concepts": 144, "before_alignment": { "val_loss": 4.3965, "val_rvq_loss": 4.3965, "val_ppl": 81.1658, "val_rvq_ppl": 81.1658, "val_rvq_mean_acc": 0.0148, "rvq_token_acc": 0.0148, "val_rvq_cb0_acc": 0.0134, "rvq_cb0_acc": 0.0134, "rvq_codebook_accs": [ 0.0134, 0.0203, 0.0097, 0.0156 ], "text_conditioning_margin": -0.0074, "mismatched_text_rvq_loss": 4.3891, "decoder_val_loss": 2.9498, "val_mse": 0.2922, "patch_polar_cosine_sim": 0.1004, "image_polar_cosine_sim": 0.0855, "spectral_magnitude_cosine": 0.1782, "spectral_snr_db": -8.2, "psnr_db": 11.38, "spectral_l1": 0.2442 }, "after_alignment": { "optimizer_type": "huygens_adam", "variant": "HuygensAdam (Circular Order Parameter R_p Splay Torque)", "steps": 200, "final_train_loss": 2.8188, "val_loss": 2.8317, "val_rvq_loss": 2.8317, "val_ppl": 16.975, "val_rvq_ppl": 16.975, "val_rvq_mean_acc": 0.2519, "rvq_token_acc": 0.2519, "val_rvq_cb0_acc": 0.2086, "rvq_cb0_acc": 0.2086, "rvq_codebook_accs": [ 0.2086, 0.2563, 0.2604, 0.2823 ], "text_conditioning_margin": 0.644, "mismatched_text_rvq_loss": 3.4757, "decoder_val_loss": 0.6859, "val_mse": 0.0524, "patch_polar_cosine_sim": 0.6737, "image_polar_cosine_sim": 0.6303, "spectral_magnitude_cosine": 0.7857, "spectral_snr_db": 2.31, "psnr_db": 19.14, "spectral_l1": 0.0515, "wall_time_s": 55.59, "elapsed_s": 55.59, "elapsed_sec": 55.59 }, "optimizer_ablations_8way": { "pair_adam": { "optimizer_type": "pair_adam", "variant": "PairAdam (Factored Toroidal Baseline)", "steps": 200, "final_train_loss": 2.8306, "val_loss": 2.8417, "val_rvq_loss": 2.8417, "val_ppl": 17.1447, "val_rvq_ppl": 17.1447, "val_rvq_mean_acc": 0.2504, "rvq_token_acc": 0.2504, "val_rvq_cb0_acc": 0.2048, "rvq_cb0_acc": 0.2048, "rvq_codebook_accs": [ 0.2048, 0.2577, 0.2575, 0.2816 ], "text_conditioning_margin": 0.6356, "mismatched_text_rvq_loss": 3.4773, "decoder_val_loss": 0.6869, "val_mse": 0.0524, "patch_polar_cosine_sim": 0.6731, "image_polar_cosine_sim": 0.63, "spectral_magnitude_cosine": 0.7847, "spectral_snr_db": 2.31, "psnr_db": 19.14, "spectral_l1": 0.0513, "wall_time_s": 196.0, "elapsed_s": 196.0, "elapsed_sec": 196.0 }, "hessian_pair_adam": { "optimizer_type": "hessian_pair_adam", "variant": "HessianPairAdam (Sophia-Torus Cauchy-Riemann Curvature)", "steps": 200, "final_train_loss": 2.826, "val_loss": 2.8382, "val_rvq_loss": 2.8382, "val_ppl": 17.085, "val_rvq_ppl": 17.085, "val_rvq_mean_acc": 0.2504, "rvq_token_acc": 0.2504, "val_rvq_cb0_acc": 0.2067, "rvq_cb0_acc": 0.2067, "rvq_codebook_accs": [ 0.2067, 0.2558, 0.2595, 0.2797 ], "text_conditioning_margin": 0.6341, "mismatched_text_rvq_loss": 3.4723, "decoder_val_loss": 0.6885, "val_mse": 0.0525, "patch_polar_cosine_sim": 0.6723, "image_polar_cosine_sim": 0.629, "spectral_magnitude_cosine": 0.7844, "spectral_snr_db": 2.3, "psnr_db": 19.13, "spectral_l1": 0.0513, "wall_time_s": 172.73, "elapsed_s": 172.73, "elapsed_sec": 172.73 }, "polar_natural_adam": { "optimizer_type": "polar_natural_adam", "variant": "PolarNaturalAdam (Riemannian Polar Metric + Cauchy-Riemann Hessian)", "steps": 200, "final_train_loss": 2.8244, "val_loss": 2.8368, "val_rvq_loss": 2.8368, "val_ppl": 17.0605, "val_rvq_ppl": 17.0605, "val_rvq_mean_acc": 0.2508, "rvq_token_acc": 0.2508, "val_rvq_cb0_acc": 0.2085, "rvq_cb0_acc": 0.2085, "rvq_codebook_accs": [ 0.2085, 0.256, 0.2588, 0.2798 ], "text_conditioning_margin": 0.6333, "mismatched_text_rvq_loss": 3.4701, "decoder_val_loss": 0.6896, "val_mse": 0.0526, "patch_polar_cosine_sim": 0.6717, "image_polar_cosine_sim": 0.6285, "spectral_magnitude_cosine": 0.7837, "spectral_snr_db": 2.29, "psnr_db": 19.13, "spectral_l1": 0.0514, "wall_time_s": 185.71, "elapsed_s": 185.71, "elapsed_sec": 185.71 }, "cyclide_adam": { "optimizer_type": "cyclide_adam", "variant": "CyclideAdam (Dupin Cyclide Conic Metric + Deviatoric Twist)", "steps": 200, "final_train_loss": 2.8337, "val_loss": 2.8454, "val_rvq_loss": 2.8454, "val_ppl": 17.2093, "val_rvq_ppl": 17.2093, "val_rvq_mean_acc": 0.2502, "rvq_token_acc": 0.2502, "val_rvq_cb0_acc": 0.2087, "rvq_cb0_acc": 0.2087, "rvq_codebook_accs": [ 0.2087, 0.2542, 0.2569, 0.2809 ], "text_conditioning_margin": 0.6292, "mismatched_text_rvq_loss": 3.4746, "decoder_val_loss": 0.6924, "val_mse": 0.0527, "patch_polar_cosine_sim": 0.6701, "image_polar_cosine_sim": 0.6274, "spectral_magnitude_cosine": 0.7831, "spectral_snr_db": 2.28, "psnr_db": 19.12, "spectral_l1": 0.0515, "wall_time_s": 85.62, "elapsed_s": 85.62, "elapsed_sec": 85.62 }, "symplectic_adam": { "optimizer_type": "symplectic_adam", "variant": "SymplecticAdam (Gromov-Liouville Non-Squeezing Floor + Casimir Cavity)", "steps": 200, "final_train_loss": 2.8965, "val_loss": 2.9089, "val_rvq_loss": 2.9089, "val_ppl": 18.337, "val_rvq_ppl": 18.337, "val_rvq_mean_acc": 0.2401, "rvq_token_acc": 0.2401, "val_rvq_cb0_acc": 0.1965, "rvq_cb0_acc": 0.1965, "rvq_codebook_accs": [ 0.1965, 0.245, 0.2459, 0.2731 ], "text_conditioning_margin": 0.6056, "mismatched_text_rvq_loss": 3.5145, "decoder_val_loss": 0.7264, "val_mse": 0.0541, "patch_polar_cosine_sim": 0.6508, "image_polar_cosine_sim": 0.613, "spectral_magnitude_cosine": 0.7727, "spectral_snr_db": 2.14, "psnr_db": 18.97, "spectral_l1": 0.0513, "wall_time_s": 65.04, "elapsed_s": 65.04, "elapsed_sec": 65.04 }, "gyric_adam": { "optimizer_type": "gyric_adam", "variant": "GyricAdam (Poinsot-Lande Gyroscopic Projection + Thomas Precession)", "steps": 200, "final_train_loss": 2.852, "val_loss": 2.8651, "val_rvq_loss": 2.8651, "val_ppl": 17.5505, "val_rvq_ppl": 17.5505, "val_rvq_mean_acc": 0.2464, "rvq_token_acc": 0.2464, "val_rvq_cb0_acc": 0.2004, "rvq_cb0_acc": 0.2004, "rvq_codebook_accs": [ 0.2004, 0.2519, 0.256, 0.2774 ], "text_conditioning_margin": 0.6298, "mismatched_text_rvq_loss": 3.4949, "decoder_val_loss": 0.7089, "val_mse": 0.0534, "patch_polar_cosine_sim": 0.6608, "image_polar_cosine_sim": 0.6204, "spectral_magnitude_cosine": 0.7786, "spectral_snr_db": 2.21, "psnr_db": 19.04, "spectral_l1": 0.0516, "wall_time_s": 47.46, "elapsed_s": 47.46, "elapsed_sec": 47.46 }, "huygens_adam": { "optimizer_type": "huygens_adam", "variant": "HuygensAdam (Circular Order Parameter R_p Splay Torque)", "steps": 200, "final_train_loss": 2.8188, "val_loss": 2.8317, "val_rvq_loss": 2.8317, "val_ppl": 16.975, "val_rvq_ppl": 16.975, "val_rvq_mean_acc": 0.2519, "rvq_token_acc": 0.2519, "val_rvq_cb0_acc": 0.2086, "rvq_cb0_acc": 0.2086, "rvq_codebook_accs": [ 0.2086, 0.2563, 0.2604, 0.2823 ], "text_conditioning_margin": 0.644, "mismatched_text_rvq_loss": 3.4757, "decoder_val_loss": 0.6859, "val_mse": 0.0524, "patch_polar_cosine_sim": 0.6737, "image_polar_cosine_sim": 0.6303, "spectral_magnitude_cosine": 0.7857, "spectral_snr_db": 2.31, "psnr_db": 19.14, "spectral_l1": 0.0515, "wall_time_s": 55.59, "elapsed_s": 55.59, "elapsed_sec": 55.59 }, "huygens_symplectic_adam": { "optimizer_type": "huygens_symplectic_adam", "variant": "HuygensSymplecticAdam (Two-Stage Huygens -> Symplectic Curriculum)", "steps": 200, "final_train_loss": 2.8962, "val_loss": 2.9087, "val_rvq_loss": 2.9087, "val_ppl": 18.3322, "val_rvq_ppl": 18.3322, "val_rvq_mean_acc": 0.2401, "rvq_token_acc": 0.2401, "val_rvq_cb0_acc": 0.1965, "rvq_cb0_acc": 0.1965, "rvq_codebook_accs": [ 0.1965, 0.2447, 0.2459, 0.2733 ], "text_conditioning_margin": 0.6061, "mismatched_text_rvq_loss": 3.5147, "decoder_val_loss": 0.7263, "val_mse": 0.0541, "patch_polar_cosine_sim": 0.6508, "image_polar_cosine_sim": 0.6131, "spectral_magnitude_cosine": 0.7727, "spectral_snr_db": 2.14, "psnr_db": 18.97, "spectral_l1": 0.0513, "wall_time_s": 55.41, "elapsed_s": 55.41, "elapsed_sec": 55.41 } } }