---
license: apache-2.0
tags:
- causal-lm
- russian
- chinese
- mixture-of-experts
- frozen-embeddings
- research
- demonstration
- low-resource
pipeline_tag: text-generation
library_name: transformers
---

# BVV-MoE: Mixture-of-Experts LLM with Frozen Shared Embeddings (Russian + Chinese, Demo-Scale)

This repository contains the model and associated resources from the papers

[📚 Paper (Emergent Semantics Beyond Token Embeddings: Transformer LMs with Frozen Visual Unicode Representations)](https://huggingface.co/papers/2507.04886) - 

[📚 Paper (Growing Transformers: Modular Composition and Layer-wise Expansion on a Frozen Substrate)](https://huggingface.co/papers/2507.07129) - 

[💻 Code](https://github.com/AVBochkov/Embeddings)

**Model size**: ~0.9B parameters  
**Languages**: Russian, Chinese, some English  

---

## Model Summary

**best_bvv_moe** is a demonstration-scale Mixture-of-Experts (MoE) decoder-only causal language model combining two independently trained models (Russian and Chinese) with strictly frozen, shared **visual/Unicode-based token embeddings**.  
- Each "expert" was pre-trained on a small subordinate corpus (English-Russian, English-Chinese) with ~9B total tokens, mixing 10% SFT-like samples, using the same, fully frozen embedding matrix for all languages.
- After separate training, the two models were seamlessly merged at the transformer block level using a "mean logits" MoE fusion approach – thanks to the shared frozen token embeddings, no retraining/alignment of embeddings was needed.
- This model is a **conceptual/research artifact**, designed to illustrate that frozen, non-semantic embeddings enable combining multilingual LMs into a working MoE model *without catastrophic loss* of performance.

---

## Key Features

- **Frozen, Unicode/visual token embeddings**: All tokens (for all supported languages) share the same frozen embedding matrix, based on Unicode and visual forms, not statistical co-occurrence.
- **Direct Mixture-of-Experts merge**: Two language models (Russian-, Chinese-oriented) are combined *without retraining* via simple logits averaging, made possible by the strictly-shared embeddings.
- **Demo-scale**: Trained on a modest dataset (9B tokens), with a small SFT fraction (~10%), intended to illustrate the principle, not to maximize absolute scores.
- **Comparison available**: Separately released standard (unfrozen embeddings) models for direct comparison of convergence and generalization.
- **Extremely "clean" codebase**: No reliance on exotic pipeline tricks; clear transformer architecture, easy to review and experiment with.

---

## Use Case / Intended Purpose

This model is **not** an end-user chatbot solution.  
Its purpose is:
- To **demonstrate** new possibilities in LM architecture:  
    - Multilingual/multimodal MoE with frozen, shared embeddings  
    - Modular, "plug-and-play" scaling and mixing of LMs  
    - Comparison between frozen and unfrozen/learnable embeddings in real convergence  
- As a **reference implementation** for research communities investigating model unification, low-resource language mixing, or studying where "meaning" emerges inside LLM architectures.

---
## Evaluation
MMLU (across tasks, test set mean ± std):

MMLU: 23.44% ± 0.28%

ARC-e: 23.74% ± 1.02%

ARC-c: 25.28% ± 2.07%

C-SENSE: 19.69% ± 1.13%

SQUAD: 19.73% ± 1.45%

BLEU:

en-ru: 6.52% ± 0.62%
ru-en: 6.22% ± 0.38%
en-zh: 2.93% ± 0.34%
zh-en: 4.95% ± 0.59%

## 🧑‍🔬 Citation & Concept

If you use or build upon this demo, please cite:

```
@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.


## Example Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
model = AutoModelForCausalLM.from_pretrained('Bochkov/best_bvv_moe', trust_remote_code=True).to('cuda')
tokenizer = AutoTokenizer.from_pretrained('Bochkov/best_bvv_moe')
inputs = tokenizer("Hello, мир! ", return_tensors="pt").to('cuda')
outputs = model.generate(
    **inputs, 
    max_new_tokens=100, 
    temperature=0.8, 
    top_k=50, 
    top_p=0.95, 
    do_sample=True
)
print(tokenizer.decode(outputs[0]))
```