Create README.md

README.md

@@ -0,0 +1,84 @@
+---
+license: mit
+datasets:
+- FINGU-AI/llm_evaluation_full_dataset_unique_both_directions_7_to_10_scores
+language:
+- en
+- ko
+- ja
+- id
+- zh
+- my
+- bn
+- km
+- mn
+- ne
+- ru
+- uz
+- tl
+- pt
+- si
+- vi
+---
+# QWEN2.5-32B-2600s-FP8: Advanced Multilingual Translation Model
+
+## Overview
+**FINGU-AI/QWEN2.5-32B-2600s-FP8** is a fine-tuned version of Qwen 2.5 32B, specifically optimized for multilingual translation across **16 different languages**. This model has been extensively fine-tuned to enhance its translation capabilities, making it competitive with high-tier models like 72B in terms of translation accuracy and fluency.
+
+## Fine-Tuning Process
+### Data Collection
+To improve the model's understanding and translation capabilities, we curated and synthesized a large dataset consisting of:
+- High-quality multilingual conversational datasets.
+- Real-world dialogues spanning general, business, and technical domains.
+- Translated datasets covering diverse linguistic structures and idiomatic expressions.
+
+### Multilingual Enhancement
+To advance its translation capabilities, we leveraged:
+- **Translation Expansion**: The collected dataset was translated into **16 different languages** to ensure robust multilingual performance.
+- **Benchmarking Against High-Tier Models**: We utilized state-of-the-art translation models, including **Gemini** and other top-ranking translation models with high BLEU and COMET scores, to refine our translation quality.
+- **Reinforcement Learning with Human Feedback (RLHF)**: Translation outputs were evaluated and iteratively improved based on feedback from native speakers and linguistic experts.
+
+### Training and Optimization
+- **Base Model**: Qwen 2.5 32B FP8
+- **Fine-Tuning Framework**: LoRA + QLoRA for efficient training
+- **Batch Size**: Optimized for multi-GPU environments
+- **Precision**: FP8 for efficient computation without sacrificing performance
+- **Training Iterations**: Over 2600 steps on **multi-H100 GPUs**
+
+## Key Improvements
+- **Enhanced Multilingual Translation**: The model now achieves translation fluency comparable to 72B models across multiple language pairs.
+- **Diverse Conversational Understanding**: Improved ability to process and generate accurate translations for various contexts, including business, casual, and formal speech.
+- **Optimized for Low-Latency Inference**: Fine-tuned with efficiency in mind, making it suitable for real-time translation applications.
+
+## Performance Evaluation
+The model was evaluated using:
+- **BLEU, COMET, and chrF scores**: To measure translation quality across multiple languages.
+- **Human Evaluation**: Involving bilingual speakers and linguistic professionals to validate accuracy and fluency.
+- **Comparisons with SOTA Models**: Benchmarked against high-performance models like GPT-4, Gemini, and LLaMA-3 to ensure top-tier translation quality.
+
+## Usage
+This model is suitable for:
+- High-quality machine translation across multiple languages
+- Conversational AI with multilingual capabilities
+- Cross-lingual content generation and customer support
+- NLP applications requiring robust and accurate translation
+
+## Limitations
+- While translation quality is highly competitive, niche dialects or highly technical documents may require additional fine-tuning.
+- Performance may vary slightly depending on the deployment environment and inference settings.
+
+## Citation
+If you use this model, please cite:
+```
+@misc{FINGU-AI-QWEN2.5-32B-2600s-FP8,
+  author = {FINGU-AI},
+  title = {FINGU-AI/QWEN2.5-32B-2600s-FP8: Advanced Multilingual Translation Model},
+  year = {2025},
+  publisher = {Hugging Face},
+  url = {https://huggingface.co/FINGU-AI/QWEN2.5-32B-2600s-FP8}
+}
+```
+
+---
+### License
+This model follows the licensing terms of the original Qwen 2.5 32B model. Ensure compliance with regional translation regulations before deploying in production environments.