import gradio as gr
import pandas as pd
from css_html_js import custom_css

TITLE = """<h1 align="center" id="space-title">🇲🇾 Malaysian RAG Embedding Leaderboard</h1>"""

INTRODUCTION_TEXT = """
📐 The 🇲🇾 Malaysian RAG Embedding Leaderboard aims to track, rank and evaluate Top-k retrieval using embedding models. All notebooks at https://github.com/mesolitica/embedding-benchmarks, feel free to submit your own score at https://huggingface.co/spaces/mesolitica/Malaysian-Embedding-Leaderboard/discussions with link to the notebook.

## Dataset

📈 We evaluate models based on 2 datasets,

1. Research paper keyword `melayu` using Crossref, https://huggingface.co/datasets/mesolitica/malaysian-ultrachat/resolve/main/ultrachat-crossref-melayu-malay.jsonl
2. Epenerbitan, https://huggingface.co/datasets/mesolitica/malaysian-ultrachat/resolve/main/ultrachat-epenerbitan-malay.jsonl
"""

close_source = [
    {
        'model': 'OpenAI ADA-002',
        'Crossref Melayu top-1': 0.3155939351340496,
        'Crossref Melayu top-3': 0.5120996083944171,
        'Crossref Melayu top-5': 0.5878100210864544,
        'Crossref Melayu top-10': 0.6721558389396526,
        'lom.agc.gov.my top-1': 0.19168533731640527,
        'lom.agc.gov.my top-3': 0.2827981080408265,
        'lom.agc.gov.my top-5': 0.322504356484939,
        'lom.agc.gov.my top-10': 0.36855862584017923,
    }
]

open_source = [
    {
        'model': '[mesolitica/llama2-embedding-600m-8k](https://huggingface.co/mesolitica/llama2-embedding-600m-8k)',
        'Crossref Melayu top-1': 0.09549151521237072,
        'Crossref Melayu top-3': 0.1834521538307059,
        'Crossref Melayu top-5': 0.23375840947886334,
        'Crossref Melayu top-10': 0.3098704689225826,
        'lom.agc.gov.my top-1': 0.05215334826985312,
        'lom.agc.gov.my top-3': 0.09932785660941,
        'lom.agc.gov.my top-5': 0.12969878018421707,
        'lom.agc.gov.my top-10': 0.1797361214836943,
    },
    {
        'model': '[mesolitica/llama2-embedding-1b-8k](https://huggingface.co/mesolitica/llama2-embedding-1b-8k)',
        'Crossref Melayu top-1': 0.06777788934631991,
        'Crossref Melayu top-3': 0.142584596847073,
        'Crossref Melayu top-5': 0.18817150316296816,
        'Crossref Melayu top-10': 0.25715433276433375,
        'lom.agc.gov.my top-1': 0.06870799103808813,
        'lom.agc.gov.my top-3': 0.1343042071197411,
        'lom.agc.gov.my top-5': 0.1717699775952203,
        'lom.agc.gov.my top-10': 0.23089370176748816,
    },
    {
        'model': '[mesolitica/mistral-embedding-191m-8k-contrastive](https://huggingface.co/mesolitica/mistral-embedding-191m-8k-contrastive)',
        'Crossref Melayu top-1': 0.08001654088700506,
        'Crossref Melayu top-3': 0.17378269409697095,
        'Crossref Melayu top-5': 0.232192701333609,
        'Crossref Melayu top-10': 0.32482166856197664,
        'lom.agc.gov.my top-1': 0.041075429424943986,
        'lom.agc.gov.my top-3': 0.09148618371919343,
        'lom.agc.gov.my top-5': 0.12758277321384118,
        'lom.agc.gov.my top-10': 0.18707991038088126,
    },
    {
        'model': '[mesolitica/mistral-embedding-349m-8k-contrastive](https://huggingface.co/mesolitica/mistral-embedding-349m-8k-contrastive)',
        'Crossref Melayu top-1': 0.09045797580895276,
        'Crossref Melayu top-3': 0.18742892587615012,
        'Crossref Melayu top-5': 0.2444949860436266,
        'Crossref Melayu top-10': 0.3398118474103174,
        'lom.agc.gov.my top-1': 0.039581777445855115,
        'lom.agc.gov.my top-3': 0.08849887976101568,
        'lom.agc.gov.my top-5': 0.12335075927308937,
        'lom.agc.gov.my top-10': 0.18558625840179238,
    },
    {
        'model': '[mesolitica/llama2-embedding-600m-8k-contrastive](https://huggingface.co/mesolitica/llama2-embedding-600m-8k)',
        'Crossref Melayu top-1': 0.11433888142251629,
        'Crossref Melayu top-3': 0.2554533236844826,
        'Crossref Melayu top-5': 0.33185154553912954,
        'Crossref Melayu top-10': 0.44639718804920914,
        'lom.agc.gov.my top-1': 0.05738113019666418,
        'lom.agc.gov.my top-3': 0.12571570823998007,
        'lom.agc.gov.my top-5': 0.16455065969629076,
        'lom.agc.gov.my top-10': 0.24446104057754542,
    },
    {
        'model': '[mesolitica/llama2-embedding-1b-8k-contrastive](https://huggingface.co/mesolitica/llama2-embedding-1b-8k)',
        'Crossref Melayu top-1': 0.17843481856714566,
        'Crossref Melayu top-3': 0.3528377959268066,
        'Crossref Melayu top-5': 0.45208311795720046,
        'Crossref Melayu top-10': 0.5800682311588959,
        'lom.agc.gov.my top-1': 0.09733632063729152,
        'lom.agc.gov.my top-3': 0.1919342793129201,
        'lom.agc.gov.my top-5': 0.2534229524520787,
        'lom.agc.gov.my top-10': 0.34602937515558874,
    },
    {
        'model': '[mesolitica/llama2-embedding-2b-8k-contrastive](https://huggingface.co/mesolitica/llama2-embedding-2b-8k)',
        'Crossref Melayu top-1': 0.1522795409903856,
        'Crossref Melayu top-3': 0.3051793652434612,
        'Crossref Melayu top-5': 0.3899514111444226,
        'Crossref Melayu top-10': 0.503049726041559,
        'lom.agc.gov.my top-1': 0.09123724172267862,
        'lom.agc.gov.my top-3': 0.1779935275080906,
        'lom.agc.gov.my top-5': 0.22479462285287527,
        'lom.agc.gov.my top-10': 0.30931043066965397,
    },
]

data = pd.DataFrame(close_source + open_source)

demo = gr.Blocks(css=custom_css)
with demo:
    gr.HTML(TITLE)
    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
    gr.DataFrame(data, datatype = 'markdown')

demo.launch()