README.md

---
license: apache-2.0
language:
- en
- ru
pipeline_tag: text-generation
tags:
- llama-3
- custom-tokenizer
- bilingual
- code
- math
---

# LLM D6 (~0.8B)

**LLM D6** is the 6th model in the series, built on a custom LLaMA-3 architecture. It features **0.8B parameters** and was pre-trained from scratch on **300 Billion tokens** of high-quality, diverse data spanning English, Russian, programming code, and mathematical reasoning.

This repository (`firdavsus/LLM_multimodal`) contains both the **Pre-trained** base weights and the **Fine-tuned** instruct/chat weights, along with training logs and evaluation benchmarks.

---

## Benchmarks
![benchmark not available](benchmarks.png)

## Fine Tuning Grapth
![benchmark not available](training_curves_finetune.png)

## 💡 Key Highlights

* **Bilingual & Domain-Focused:** Pre-trained on a balanced mix of English, Russian, Code, and Mathematics.
* **Custom Tokenizer:** Trained from scratch on the dataset with a **52,000 vocabulary size**, yielding higher token compression efficiency for Russian and technical syntax.
* **High Token Efficiency:** Achieves competitive performance at 300B tokens compared to models trained on 9T–18T tokens.
* **Dual Weights Included:** Structured subdirectories for both `pre-train` and `fine-tune` stages.

---

## 📐 Model Architecture & Specifications

| Parameter | Value |
| :--- | :--- |
| **Architecture** | LLaMA-3 style (`LlamaForCausalLM`) |
| **Total Parameters** | ~800M (0.8B) |
| **Hidden Dimension ($d_{\text{model}}$)** | 1536 |
| **Number of Layers** | 24 |
| **Head Dimension** | 128 |
| **Attention Heads** | 12 |
| **Vocabulary Size** | 52,000 (Custom Trained) |
| **Context Pre-training Tokens** | 300 Billion Tokens |

---

## 📁 Repository Structure

```text
firdavsus/LLM_multimodal/
├── pre-train/                 # Pre-trained base weights & tokenizer
├── fine-tune/                 # Supervised Fine-Tuned (SFT) weights & configs
├── benchmarks.png             # Evaluation comparison curves & charts
├── training_curves_finetune.png # SFT Loss, LR, and Gradient stability plots
└── README.md                  # Model Card

```

---

## 📊 Evaluation & Benchmarks

The model was evaluated using `lm-evaluation-harness` across zero-shot and few-shot tasks, comparing **Custom 0.8B (300B tokens)** against **Qwen 2.5 0.5B (18.0T tokens)** and **Llama 3.2 1B (9.0T tokens)**.

### Final Benchmark Performance Summary

| Benchmark | Task Metric | **Custom 0.8B** *(300B Tokens)* | **Qwen 2.5 0.5B** *(18,000B Tokens)* | **Llama 3.2 1B** *(9,000B Tokens)* |
| --- | --- | --- | --- | --- |
| **PIQA** | Physical Commonsense | **69.3%** | 69.8% | 74.5% |
| **HellaSwag** | General Reasoning | **47.8%** | 52.2% | 64.1% |
| **XWinograd RU** | Russian Language Understanding | **58.7%** | 56.8% | 66.0% |
| **ARC-Challenge** | Grade-School Science | **30.6%** | 32.2% | 37.2% |
| **MMLU** | Multi-task Accuracy | **25.5%** | 47.4% | 37.8% |

> **Note:** Despite being trained on significantly fewer tokens (0.3T vs 18T/9T), **LLM D6** demonstrates strong bilingual capability, outperforming Qwen 2.5 0.5B on Russian Winograd evaluation (`58.7%` vs `56.8%`) and maintaining competitive common-sense physical reasoning (`69.3%` on PIQA).

---

## 📉 Fine-Tuning Dynamics

Fine-tuning was conducted over **3,600 steps** using a cosine learning rate decay schedule with a peak learning rate of $2.0 \times 10^{-5}$ and initial linear warmup.

* **Loss Convergence:** Both training and evaluation loss decayed smoothly down to **~1.16**, showing consistent learning without over-fitting.
* **Gradient Stability:** Gradient norm remained tightly bounded between **0.5 and 1.5** throughout the execution.

---

## 🚀 How to Use

### Loading the Fine-Tuned Model

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

model_id = "firdavsus/LLM_multimodal"

# Load Custom Tokenizer and Model Weights from fine-tune folder
tokenizer = AutoTokenizer.from_pretrained(model_id, subfolder="fine-tune")
model = AutoModelForCausalLM.from_pretrained(
    model_id, 
    subfolder="fine-tune",
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

prompt = "Solve the following problem step by step: What is the derivative of f(x) = x^3 + 2x?"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(
    **inputs, 
    max_new_tokens=256, 
    temperature=0.7, 
    top_p=0.9
)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

```

### Loading the Pre-Trained Base Weights

```python
tokenizer = AutoTokenizer.from_pretrained("firdavsus/LLM_multimodal", subfolder="pre-train")
model = AutoModelForCausalLM.from_pretrained("firdavsus/LLM_multimodal", subfolder="pre-train")

```

---

## ⚠️ Limitations & Intended Use

* **Primary Intended Use:** General text generation, bilingual English/Russian translation, code assistance, and basic mathematical reasoning.
* **Limitations:** Due to its lightweight 0.8B parameter size, complex multi-step reasoning (e.g., MMLU domain knowledge) may show limitations compared to much larger models. Users should implement system prompts or validation guardrails for mission-critical deployments.

```

```