Hindi-wiki-LLaMA

Hindi Wikipedia Article Generation Model This repository contains a language generation model trained on Hindi Wikipedia articles using the Hugging Face Transformers library. The model is based on the Llama-2 architecture and fine-tuned on a large dataset of Hindi text from Wikipedia.

Model Details

• Base Model: Llama-2
• Pretraining Dataset: Hindi Wikipedia Articles
• Tokenizer: Hugging Face Tokenizer
• Model Architecture: Causal Language Modeling

from peft import AutoPeftModelForCausalLM

base_model_name = "meta-llama/Llama-2-7b-hf"
tokenizer = AutoTokenizer.from_pretrained(base_model_name, trust_remote_code=True)
tokenizer.pad_token = tokenizer.eos_token
output_dir = "./final_checkpoint"
device_map = {"": 0}
model = AutoPeftModelForCausalLM.from_pretrained(output_dir, device_map=device_map, torch_dtype=torch.bfloat16)
device = torch.device("cuda")
text = ""
inputs = tokenizer(text, return_tensors="pt").to(device)
outputs = model.generate(input_ids=inputs["input_ids"].to("cuda"), attention_mask=inputs["attention_mask"], max_new_tokens=100, pad_token_id=tokenizer.eos_token_id)

print(tokenizer.decode(outputs[0][len(inputs["input_ids"][0]):], skip_special_tokens=True))

Model Performance:--

The model has been trained on a substantial amount of Hindi Wikipedia articles, which allows it to generate coherent and contextually relevant text.