# Model Repository Structure

This document describes the files in this Hugging Face model repository.

```
VibeThinker-3B/
├── README.md                  # Model card (this fork's documentation)
├── ORIGINAL_README.md         # Original model card from WeiboAI (preserved)
├── ATTRIBUTION.md             # Credits and license information
├── USAGE.md                   # Usage and inference guide
├── MODEL_STRUCTURE.md         # This file
├── .gitattributes             # Git LFS configuration
│
├── config.json                # Model configuration (Qwen2.5-Coder-3B based)
├── generation_config.json     # Default generation settings
│
├── model-00001-of-00002.safetensors  # Model weights (shard 1, ~4.9 GB)
├── model-00002-of-00002.safetensors  # Model weights (shard 2, ~1.2 GB)
├── model.safetensors.index.json      # Weight index mapping
│
├── tokenizer.json             # Full tokenizer (BBPE)
├── tokenizer_config.json      # Tokenizer configuration
├── vocab.json                 # Vocabulary file
├── merges.txt                 # BPE merges
├── added_tokens.json          # Added/special tokens
├── special_tokens_map.json    # Special token mappings
│
├── trainer_state.json         # Training state metadata
│
└── pictures/                  # Figures from the original model card
    ├── Abstrct.png
    ├── Acc_and_Scale.png
    ├── Architecture.png
    ├── LeetCode.png
    ├── VibeThiinker-3B.png
    └── VibeThinker-3B+CLR.png
```

## File Sizes

| File | Size | Description |
|------|------|-------------|
| model-00001-of-00002.safetensors | ~4.9 GB | First weight shard |
| model-00002-of-00002.safetensors | ~1.2 GB | Second weight shard |
| tokenizer.json | ~11 MB | Full tokenizer |
| vocab.json | ~2.8 MB | Vocabulary |
| merges.txt | ~1.7 MB | BPE merge rules |
| All other files | < 1 MB each | Config, metadata, images |

All `.safetensors` files are tracked with Git LFS.