File size: 5,086 Bytes
06eaf0f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
---
language:
- en
- hi
license: llama2
tags:
- multilingual
- instruction-tuning
- llama2
datasets:
- ai4bharat/indic-instruct-data-v0.1
---

# Airavata

This model is a 7B [OpenHathi](https://huggingface.co/sarvamai/OpenHathi-7B-Hi-v0.1-Base) model finetuned on [IndicInstruct dataset](https://huggingface.co/datasets/ai4bharat/indic-instruct-data-v0.1) 
which is a collection of instruction datasets (Anudesh, wikiHow, Flan v2, Dolly, Anthropic-HHH, OpenAssistant v1, and LymSys-Chat). 
Please check the corresponding huggingface dataset card for more details.

This was trained as part of the technical report [Airavata: Introducing Hindi Instruction-tuned LLM](https://arxiv.org/abs/2401.15006).
The codebase used to train and evaluate this model can be found at [https://github.com/AI4Bharat/IndicInstruct](https://github.com/AI4Bharat/IndicInstruct).

## Usage

Clone [https://github.com/AI4Bharat/IndicInstruct](https://github.com/AI4Bharat/IndicInstruct) and install the required dependencies. Then download or clone this model to the same machine.

## Input Format

The model is trained to use the chat format similar to [open-instruct code repository](https://github.com/allenai/open-instruct) (note the newlines):
```
<|user|>
Your message here!
<|assistant|>
```

For best results, format all inputs in this manner. **Make sure to include a newline after `<|assistant|>`, this can affect generation quality quite a bit.**

## Hyperparameters

We fine-tune OpenHathi base model on the aforementioned IndicInstruct dataset with LoRA. The hyperparameters for the LoRA fine-tuning are listed below:
- LoRA Rank: 16
- LoRA alpha: 32
- LoRA Dropout: 0.05
- LoRA Target Modules: ["q_proj", "v_proj", "k_proj", "down_proj", "gate_proj", "up_proj"]
- Epochs: 4
- Learning rate: 5e-4
- Batch Size: 128
- Floating Point Precision: bfloat16

We recommend the readers to check out [our official blog post](https://ai4bharat.github.io/airavata) for more details on the model training, ablations and evaluation results.

## Example

```python3
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

device = "cuda" if torch.cuda.is_available() else "cpu"


def create_prompt_with_chat_format(messages, bos="<s>", eos="</s>", add_bos=True):
    formatted_text = ""
    for message in messages:
        if message["role"] == "system":
            formatted_text += "<|system|>\n" + message["content"] + "\n"
        elif message["role"] == "user":
            formatted_text += "<|user|>\n" + message["content"] + "\n"
        elif message["role"] == "assistant":
            formatted_text += "<|assistant|>\n" + message["content"].strip() + eos + "\n"
        else:
            raise ValueError(
                "Tulu chat template only supports 'system', 'user' and 'assistant' roles. Invalid role: {}.".format(
                    message["role"]
                )
            )
    formatted_text += "<|assistant|>\n"
    formatted_text = bos + formatted_text if add_bos else formatted_text
    return formatted_text


def inference(input_prompts, model, tokenizer):
    input_prompts = [
        create_prompt_with_chat_format([{"role": "user", "content": input_prompt}], add_bos=False)
        for input_prompt in input_prompts
    ]

    encodings = tokenizer(input_prompts, padding=True, return_tensors="pt")
    encodings = encodings.to(device)

    with torch.inference_mode():
        outputs = model.generate(encodings.input_ids, do_sample=False, max_new_tokens=250)

    output_texts = tokenizer.batch_decode(outputs.detach(), skip_special_tokens=True)

    input_prompts = [
        tokenizer.decode(tokenizer.encode(input_prompt), skip_special_tokens=True) for input_prompt in input_prompts
    ]
    output_texts = [output_text[len(input_prompt) :] for input_prompt, output_text in zip(input_prompts, output_texts)]
    return output_texts


model_name = "ai4bharat/Airavata"

tokenizer = AutoTokenizer.from_pretrained(model_name, padding_side="left")
tokenizer.pad_token = tokenizer.eos_token
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16).to(device)

input_prompts = [
    "मैं अपने समय प्रबंधन कौशल को कैसे सुधार सकता हूँ? मुझे पांच बिंदु बताएं।",
    "मैं अपने समय प्रबंधन कौशल को कैसे सुधार सकता हूँ? मुझे पांच बिंदु बताएं और उनका वर्णन करें।",
]
outputs = inference(input_prompts, model, tokenizer)
print(outputs)
```

## Citation

```bibtex
@article{gala2024airavata,
  title   = {Airavata: Introducing Hindi Instruction-tuned LLM},
  author  = {Jay Gala and Thanmay Jayakumar and Jaavid Aktar Husain and Aswanth Kumar M and Mohammed Safi Ur Rahman Khan and Diptesh Kanojia and Ratish Puduppully and Mitesh M. Khapra and Raj Dabre and Rudra Murthy and Anoop Kunchukuttan},
  year    = {2024},
  journal = {arXiv preprint arXiv: 2401.15006}
}
```