README.md

---
license: mit
base_model: karpathy/tinyllamas
tags:
- llama2
- gguf
- safetensors
- transformers
- tinyllamas
- validation
- test-suite
---

# TinyStories Llama2 1M (tinybpe1m) GGUF & HF Validation Suite

This repository provides ultra-lightweight Llama2 model files across various formats (both **GGUF** and **Hugging Face / Safetensors**), trained on the TinyStories dataset and optimized for testing and validation.

### Why this repository exists
When developing a custom LLM inference engine, debugging with a full-sized model is slow. This suite offers a true **1M parameter scale model** (~0.5MB to ~4MB depending on the quantization format), allowing developers to validate their loaders, serialization, quantization kernels, and inference logic step-by-step with maximum efficiency.

### Difference from `tiny1m`
This is a **BPE-based model variant**. Unlike the standard `tiny1m` model, this model is **NOT compatible with `llama2.c`**. 
The custom SentencePiece BPE tokenizer utilized here relies on the `byte_fallback` mechanism to handle unknown characters. Because `llama2.c`'s simplified native C loader/tokenizer cannot interpret or process `byte_fallback` routines, text generation will fail or corrupt in that environment. This suite is strictly designed and optimized for **`llama.cpp` (GGUF)** and **Hugging Face `transformers` (Python)** execution.

---

## 📂 Repository Structure & File Descriptions

### 1. GGUF Formats (Root Directory `./`)
A comprehensive validation suite converted for `llama.cpp` and compatible engines. The tokenizer vocabulary and special tokens are fully embedded within each GGUF binary. Every compiled quantization variant available in the root directory is explicitly covered below:

| Filename(s) / Wildcard Pattern | Type | Size | Purpose / Validation Target |
| :--- | :--- | :--- | :--- |
| **`tinybpe1m.F32.gguf`** | `F32` | ~4.0 MB | **Baseline Test.** Validates GGUF parsing, tensor layout, matrix multiplication, RoPE, and Attention logic without dequantization overhead. |
| **`tinybpe1m.F16.gguf`**<br>**`tinybpe1m.BF16.gguf`** | `F16`<br>`BF16` | ~2.0 MB | **Half-Precision Test.** Validates 16-bit floating point loading, type casting, and inference stability. |
| **`tinybpe1m.Q8_0.gguf`** | `Q8_0` | ~1.1 MB | **Quantization Level 1.** Validates block-based uniform scaling with 32 elements. |
| **`tinybpe1m.Q4_0.gguf`**<br>**`tinybpe1m.Q4_1.gguf`** | `Q4_0`<br>`Q4_1` | ~0.7 MB | **Quantization Level 2.** Validates classic 4-bit linear quantization and bit-unpacking logic. |
| **`tinybpe1m.Q2_K.gguf`** | `Q2_K` | ~0.5 MB | **Standard K-Quant (2-bit).** Validates 2-bit super-block quantization parsing. |
| **`tinybpe1m.Q3_K_*.gguf`**<br>↳ *`tinybpe1m.Q3_K_S.gguf`*<br>↳ *`tinybpe1m.Q3_K_M.gguf`*<br>↳ *`tinybpe1m.Q3_K_L.gguf`* | `Q3_K` | ~0.6 MB | **Standard K-Quant (3-bit).** Validates Small, Medium, and Large sub-variants of 3-bit multi-block structures. |
| **`tinybpe1m.Q4_K_*.gguf`**<br>↳ *`tinybpe1m.Q4_K_S.gguf`*<br>↳ *`tinybpe1m.Q4_K_M.gguf`* | `Q4_K` | ~0.7 MB | **Standard K-Quant (4-bit).** Validates Small and Medium sub-variants of modern 4-bit super-block structural parsing. |
| **`tinybpe1m.Q5_K_*.gguf`**<br>↳ *`tinybpe1m.Q5_K_S.gguf`*<br>↳ *`tinybpe1m.Q5_K_M.gguf`* | `Q5_K` | ~0.8 MB | **Standard K-Quant (5-bit).** Validates Small and Medium sub-variants of 5-bit mixed precision super-blocks. |
| **`tinybpe1m.Q6_K.gguf`** | `Q6_K` | ~0.9 MB | **Standard K-Quant (6-bit).** Validates 6-bit high-fidelity super-block quantization. |
| **`tinybpe1m.IQ3_*.gguf`**<br>↳ *`tinybpe1m.IQ3_XXS.gguf`*<br>↳ *`tinybpe1m.IQ3_S.gguf`* | `I-Quants` | ~0.5 MB | **Importance Quants (3-bit).** Non-linear 3-bit importance quantization targeting lookup table (codebook) decoding logic. |
| **`tinybpe1m.IQ4_*.gguf`**<br>↳ *`tinybpe1m.IQ4_NL.gguf`*<br>↳ *`tinybpe1m.IQ4_XS.gguf`* | `I-Quants` | ~0.6 MB | **Importance Quants (4-bit).** Non-linear 4-bit importance quantization variants (Non-Linear and Extra Small). |
| **`tinybpe1m.TQ1_0.gguf`**<br>**`tinybpe1m.TQ2_0.gguf`** | `Ternary` | ~0.4 MB | **Experimental.** Ternary (-1, 0, 1) state quantization for cutting-edge engine testing. |

### 2. Hugging Face Native Format (`./hf/`)
This directory contains the standard files required to load the model using the PyTorch `transformers` library:
* **`hf/model.safetensors`**: The raw, unquantized model weights stored securely in Safetensors format.
* **`hf/config.json`**: The architectural configuration file defining hyperparameters (layers, heads, dimensions).
* **`hf/generation_config.json`**: Default parameters optimized for text generation.
* **`hf/tokenizer_config.json`**: Tokenizer behavior layout enabling automatic BOS token injection and padding setup.
* **`hf/special_tokens_map.json`**: Architectural mappings tying token strings to exact internal special token IDs.
* **`hf/tokenizer.model`**: The custom 512-vocab SentencePiece tokenizer model file.

---

## 🚀 Usage Examples

### A. Running GGUF via llama.cpp
To verify your local setup or test custom execution backends using the official native utilities:
```bash
./llama-cli -m tinybpe1m.Q4_K_M.gguf -p "Tom and Jerry are " -n 64 --temp 0.0

```

### B. Loading Hugging Face Formats via Python

You can import the Hugging Face variant directly into Python using the `transformers` library.

```python
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

repo_id = "shibatch/tinybpe1m"

# The library automatically loads from the hf/ subfolder
tokenizer = AutoTokenizer.from_pretrained(repo_id, subfolder="hf")
model = AutoModelForCausalLM.from_pretrained(repo_id, subfolder="hf")

prompt = "Tom and Jerry are "
inputs = tokenizer(prompt, return_tensors="pt")

with torch.no_grad():
    outputs = model.generate(
        **inputs, 
        max_new_tokens=64, 
        do_sample=False,
        pad_token_id=tokenizer.eos_token_id
    )

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

```

---

## 📝 Model Specifications

The network architecture features an unshared output layer (`lm_head`) to keep memory structures consistent with standard Llama 2 definitions. Thanks to the highly optimized 512 vocabulary size, the token embedding and output layers remain extremely lightweight.

* **Architecture:** Llama 2 (Scaled-down variant)
* **Dataset:** TinyStories
* **Total Parameters:** ~1M (Exactly 896,256 parameters)
* **Vocabulary Size:** 512 (Custom SentencePiece BPE Tokenizer with `byte_fallback` enabled)
* **Hidden Size (`hidden_size`):** 128
* **Number of Hidden Layers (`num_hidden_layers`):** 4
* **Number of Attention Heads (`num_heads`):** 2
* **Number of Key-Value Heads (`num_kv_heads`):** 2
* **Intermediate Size (`intermediate_size`):** 352
* **Max Position Embeddings (`max_position_embeddings`):** 256

## 📜 Acknowledgments & License

* **Original Implementation:** Inspired by Andrej Karpathy's `llama2.c` project.
* **Dataset:** TinyStories dataset.
* **License:** **MIT License**. You are free to use, modify, and distribute these assets for any purpose.