Create README.md

README.md

@@ -0,0 +1,106 @@
+# Model Overview
+
+This model is a fine-tuned version of the Helsinki-NLP OPUS-MT model for multiple language pairs. It has been fine-tuned on the Tatoeba dataset for the following language pairs:
+
+English to Marathi (en-mr)
+
+Esperanto to Dutch (eo-nl)
+
+Spanish to Portuguese (es-pt)
+
+French to Russian (fr-ru)
+
+Spanish to Galician (es-gl)
+
+The model supports sequence-to-sequence translation and has been optimized for performance using FP16 quantization.
+
+# Model Details
+```
+Base Model: Helsinki-NLP/opus-mt-en-roa
+
+Training Dataset: Tatoeba dataset
+
+Fine-tuned Language Pairs: en-mr, eo-nl, es-pt, fr-ru, es-gl
+
+Evaluation Metric: BLEU Score (using sacreBLEU)
+
+Training Framework: Hugging Face Transformers
+
+Training Configuration
+
+Optimizer: AdamW
+
+Learning Rate: 2e-5
+
+Batch Size: 16 (per device)
+
+Weight Decay: 0.01
+
+Epochs: 3
+
+Precision: FP32 (initial training), converted to FP16 for inference
+```
+
+Quantization and FP16 Conversion
+
+To improve inference efficiency, models were converted to FP16:
+
+import torch
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
+
+# List of fine-tuned models
+models = [
+    "fine_tuned_models/en-mr/final/", 
+    "fine_tuned_models/es-pt/final/", 
+    "fine_tuned_models/eo-nl/final/", 
+    "fine_tuned_models/en-mr/final/"
+]
+
+output_fp16_dir = "fine_tuned_models_fp16"
+
+# Convert each model to FP16
+for model_path in models:
+    print(f"Quantizing {model_path} to FP16...")
+    
+    # Load model and tokenizer
+    model = AutoModelForSeq2SeqLM.from_pretrained(model_path, torch_dtype=torch.float16)
+    tokenizer = AutoTokenizer.from_pretrained(model_path)
+
+    # Define save path
+    save_path = model_path.replace("fine_tuned_models", output_fp16_dir)
+    
+    # Save quantized model
+    model.save_pretrained(save_path)
+    tokenizer.save_pretrained(save_path)
+
+    print(f"Saved quantized model to: {save_path}\n")
+
+# Inference Example
+```
+python
+from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
+import torch
+
+model = AutoModelForSeq2SeqLM.from_pretrained("fine_tuned_models_fp16/en-mr/final/", torch_dtype=torch.float16).to("cuda")
+tokenizer = AutoTokenizer.from_pretrained("fine_tuned_models_fp16/en-mr/final/")
+
+inputs = tokenizer("Hello, how are you?", return_tensors="pt").to("cuda")
+outputs = model.generate(**inputs)
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))
+```
+# Usage
+
+The models can be used for translation tasks in various NLP applications, including chatbots, document translation, and real-time communication.
+
+# Limitations
+
+May not generalize well for domain-specific text.
+
+FP16 quantization may lead to minor loss in precision.
+
+Translation accuracy depends on the dataset quality.
+
+# Citation
+
+If you use this model, please cite the original OPUS-MT paper and acknowledge the fine-tuning process conducted using the Tatoeba dataset.
+