Plane 0 (0–65535): All single Unicode code points (monograms) are mapped 1:1 to token codes, directly matching standard Unicode BMP.
Private and unused code ranges (Plane 0 high + supplementary, e.g., 0xE000–0xF8FF and 65536–131071):
- All multi-character tokens (bigrams, trigrams, SOTA model token strings) are placed exclusively in these ranges.
This design achieves total, lossless Unicode text coverage, with all multi-symbol tokens isolated above the core Unicode range.
Tokenizer uses the Mistral-Nemo model's full vocabulary
Vocabulary size: 131,072 tokens,
Embedding dimension: 1024.

The associated normalized_embeddings_weights.pt file contains a [vocab_size x embed_dim] matrix of precomputed, L2-normalized, frozen embeddings.
No semantic information is encoded; embeddings remain fixed throughout LM pretraining. No training or adaptation; suitable for plug-and-play use in research on embedding-free semantic emergence and modular LMs.

How to Get Started with the Tokenizer

python

from transformers import AutoTokenizer
from huggingface_hub import hf_hub_download
import torch
tokenizer = AutoTokenizer.from_pretrained('Bochkov/bvv241-nemo')
emb_path = hf_hub_download(
    repo_id="Bochkov/bvv241-nemo",
    filename="normalized_embeddings_weights.pt"
)
embeddings = torch.load(emb_path)

🧑‍🔬 Citation & Concept

If you use this model or the underlying concepts in your research, please cite our work:

@article{
      bochkov2025emergent,
      title={Emergent Semantics Beyond Token Embeddings: Transformer {LM}s with Frozen Visual Unicode Representations},
      author={Andrey Bochkov},
      journal={Transactions on Machine Learning Research},
      issn={2835-8856},
      year={2025},
      url={https://openreview.net/forum?id=Odh8IynO1o},
      note={}
}

@misc{bochkov2025growingtransformersmodularcomposition,
      title={Growing Transformers: Modular Composition and Layer-wise Expansion on a Frozen Substrate}, 
      author={A. Bochkov},
      year={2025},
      eprint={2507.07129},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2507.07129}, 
}

This work demonstrates that transformer blocks, not token embeddings, carry the semantic burden in LLMs — a step toward modular, fusable, multilingual LMs.