{
  "model_type": "grinpress-rectified-flow-unet",
  "total_params": 33860000,
  "unet": {
    "in_channels": 3,
    "out_channels": 3,
    "base_channels": 64,
    "channel_mults": [1, 2, 3, 4],
    "num_res_blocks": 2,
    "attn_levels": [2, 3],
    "attn_heads": 8,
    "mid_block": ["ResBlock", "AttnBlock", "ResBlock"],
    "time_embed_dim": 512,
    "norm": "GroupNorm(32)",
    "context_dim": 256,
    "params": 32230000
  },
  "text_encoder": {
    "tokenizer": "ascii_bytes",
    "vocab_size": 128,
    "pad_id": 0,
    "max_len": 64,
    "dim": 256,
    "layers": 2,
    "heads": 4,
    "ffn_mult": 4,
    "positional": "sinusoidal_1d",
    "cfg_dropout": 0.1,
    "null": "learned_sequence",
    "params": 1630000
  },
  "flow": {
    "type": "rectified_flow",
    "interpolation": "x_t = (1 - t) * x0 + t * noise",
    "target": "v = noise - x0",
    "t_sampling": "logit_normal",
    "time_input_scale": 1000
  },
  "training": {
    "resolution": 128,
    "dataset_size": 50000,
    "steps": 60000,
    "batch_size": 64,
    "optimizer": "AdamW",
    "lr": 0.0002,
    "betas": [0.9, 0.99],
    "weight_decay": 0.01,
    "lr_schedule": "linear warmup 1000 steps + cosine to 5% of lr",
    "grad_clip": 1.0,
    "ema_decay": 0.9999,
    "precision": "bf16"
  },
  "sampling": {
    "solver": "euler",
    "default_steps": 50,
    "default_cfg_scale": 5.0
  },
  "weights_file": "ema_final_bf16.pt",
  "image_range": [-1, 1]
}