Openhathi 7B English To Hinglish

🚀 OpenHathi-7B English to Hinglish Translation Model

This model is designed for English to Hinglish translation, leveraging the power of a merged LoRA trained on specific datasets. It offers a practical solution for language translation tasks, with potential for further improvement.

🚀 Quick Start

You can start using this model by following the steps in the sample code below.

✨ Features

• Enhanced Translation: A merge of LoRA trained on an English to Hinglish translation dataset by NATERAW on llama2 - 7b and OPENHATHI - 7B - BASE. The inclusion of OpenHathi, which has more Hindi data in its pretraining compared to llama2, significantly improves the translation quality.
• Prompt Template: Utilize the prompt template provided by nateraw for seamless translation.

💻 Usage Examples

Basic Usage

from transformers import LlamaForCausalLM, AutoTokenizer
import torch

device = "cuda:0"
tokenizer = AutoTokenizer.from_pretrained('akashgoel-id/OpenHathi-7B-English-to-Hinglish')
model = LlamaForCausalLM.from_pretrained('akashgoel-id/OpenHathi-7B-English-to-Hinglish', torch_dtype=torch.bfloat16).to(device)

PROMPT_TEMPLATE = (
    f"Translate from english to hinglish:\n{{en}}\n---\nTranslation:\n"
)
while True:
    userInput = input("Enter input: ")
    prompt = PROMPT_TEMPLATE.format(en=userInput)
    inputs = tokenizer(prompt, return_tensors="pt").to(device)
    generate_ids = model.generate(inputs.input_ids, max_length=500)
    print(tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0])

📚 Documentation

Prompting

You can use the following prompt template provided by nateraw: "Translate from english to hinglish:\n{{en}}\n---\nTranslation:\n"

Limitations

The model has some limitations, especially when dealing with idioms:

Next steps

1. Reduce Censorship: The model seems to be highly censored due to its use of llama2. The next step is to remove some of the censorship by fine - tuning on more uncensored data, similar to what WizardLM has done for llama2.
2. Finetune on Idioms: Fine - tune the model on idioms to improve its performance in idiomatic translations.