manifest.json

{
  "browser_note": "Same input and output names as the full M1 SciBERT ONNX export; can query semantic_m1 paper vector files.",
  "built_at": 1777171028,
  "dimension": 768,
  "interactive_level": "exports/huggingface/paper_universe_interactive_v1/interactive/papers_all.json",
  "model_kind": "m1_lite_distilled_embedding",
  "normalization": "l2",
  "onnx": "onnx/model.onnx",
  "pooling": "mean_pool_attention_mask_project_l2_normalized",
  "quantized_onnx": "onnx/model.int8.onnx",
  "scale": 127.0,
  "sequence_length": 128,
  "source_dir": "/arxiv/huggingface/paper_text_1m_dedup_v1",
  "student_id": "google/bert_uncased_L-4_H-256_A-4",
  "target_embeddings": "exports/huggingface/paper_universe_interactive_v1/semantic_m1/papers_all.emb.i8",
  "teacher_base_id": "allenai/scibert_scivocab_uncased",
  "teacher_id": "PeytonT/1m-paper-embedding-model",
  "tokenizer": "tokenizer",
  "train_stats": {
    "eval_cosine": 0.7350965235382318,
    "eval_mse": 0.0006898527972225565,
    "final_cosine": 0.739337682723999,
    "final_loss": 0.2609059512615204,
    "global_steps": 7813,
    "missing_texts": 0,
    "rows": 1000000,
    "train_seconds": 473.90148282051086
  }
}