---
language:
  - hi
license: mit
library_name: transformers
tags:
  - hindi
  - story-generation
  - causal-lm
  - llama-style
  - transformer
  - from-scratch
  - text-generation
datasets:
  - SmallScale/Simple-Stories-Hindi
pipeline_tag: text-generation
model-index:
  - name: Simple-Stories-Hindi-10M
    results: []
---

# 📖 Simple-Stories-Hindi-10M (11.45M Parameters)

A **11.45M parameter** decoder-only Transformer language model trained **from scratch** on **2.11 million Hindi simple stories**. The model generates coherent, creative, and grammatically sound Hindi stories given a short text prompt.

---

## 📊 Evaluation & Training Metrics

| Metric / Property | Value |
|---|---|
| **Best Validation Loss** | **`1.8157`** (Cross-Entropy Loss) |
| **Total Training Steps** | **202,000 steps** |
| **Total Parameters** | **11,453,120 (11.45M)** |
| **Non-Embedding Parameters** | **10,173,120 (10.17M)** |
| **Training Dataset** | [SmallScale/Simple-Stories-Hindi](https://huggingface.co/datasets/SmallScale/Simple-Stories-Hindi) (~2.11M stories) |
| **Model Size on Disk** | ~45.8 MB (`model.safetensors`) |

---

## 🏗️ Model Architecture Details

| Parameter | Value | Notes |
|---|---|---|
| **Architecture** | LLaMA-style Decoder | RoPE + SwiGLU + RMSNorm |
| **Hidden Size (`d_model`)** | 320 | Vector dimension |
| **FFN Intermediate Size** | 896 | 8/3 × `d_model` rounded to multiple of 64 |
| **Layers (`n_layers`)** | 7 | Transformer blocks |
| **Attention Heads (`n_heads`)** | 5 | Multi-Head Self Attention |
| **Head Dimension** | 64 | `d_model` / `n_heads` |
| **Context Length (`max_seq_len`)** | 512 tokens | Sequence window |
| **Vocabulary Size** | 4,000 | SentencePiece Unigram (Devanagari optimized) |
| **Weight Tying** | Enabled | Token embeddings & output projection share weights |
| **Precision** | float32 | Weights stored in native FP32 safetensors |

---

## 🚀 Quick Start & Usage

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

# Load tokenizer and model directly from Hugging Face
tokenizer = AutoTokenizer.from_pretrained("SmallScale/Simple-Stories-Hindi-10M", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("SmallScale/Simple-Stories-Hindi-10M", trust_remote_code=True)

if torch.cuda.is_available():
    model = model.to("cuda")

# Prompt input
prompt = "एक समय की बात है"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

# Generate story
outputs = model.generate(
    **inputs,
    max_new_tokens=200,
    do_sample=True,
    top_k=40,
    top_p=0.95,
    temperature=0.8
)

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

---

## 📝 Sample Generated Stories

**Prompt:** `एक समय की बात है`
> *एक समय की बात है, और मैं छाया से देखता हूं। मेरे दो लोग, जीन और सैमुअल हैं, जो एक भव्य यात्रा पर जा रहे हैं। वे एक ही स्थान पर रहते हैं, लेकिन वे दोनों अपनी-अपनी कहानियाँ चाहते हैं...*

---

## 🔗 Related Resources

- **GGUF (FP16) Model Repo:** [SmallScale/Simple-Stories-Hindi-10M-GGUF](https://huggingface.co/SmallScale/Simple-Stories-Hindi-10M-GGUF)
- **20M Model Repo:** [SmallScale/Simple-Stories-Hindi-20M](https://huggingface.co/SmallScale/Simple-Stories-Hindi-20M)
- **Dataset:** [SmallScale/Simple-Stories-Hindi](https://huggingface.co/datasets/SmallScale/Simple-Stories-Hindi)

---

## 📄 License

MIT License