projector_training_config.json

{
  "pairs": "/data/datasets/monet-100k/train_pairs.jsonl",
  "val_pairs": "/data/datasets/monet-100k/validation_pairs.jsonl",
  "model": "/data/models/LFM-2.5-Encoder-230M",
  "vision_model": "google/siglip2-base-patch16-256",
  "output": "outputs/monet-100k/projector-siglip2-clean",
  "num_image_tokens": 32,
  "batch_size": 8,
  "steps": 1000,
  "max_text_length": 128,
  "lr": 0.0001,
  "temperature": 0.07,
  "dtype": "float32",
  "seed": 42,
  "eval_every": 250,
  "eval_pairs": 256,
  "image_only": true,
  "device": "cuda:0",
  "trust_remote_code": false,
  "force": true
}