gemma-2b-it-tamil-v0.1-alpha/README.md

language, license, datasets, license_name, license_link, base_model, model-index

language	license	datasets	license_name	license_link	base_model	model-index
en ta	other	vicgalle/alpaca-gpt4 abhinand/tamil-alpaca	gemma-terms-of-use	https://ai.google.dev/gemma/terms	abhinand/gemma-2b-tamil	name results gemma-2b-it-tamil-v0.1-alpha task dataset metrics source type name text-generation Text Generation name type config split args AI2 Reasoning Challenge (25-Shot) ai2_arc ARC-Challenge test num_few_shot 25 type value name acc_norm 50.09 normalized accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=abhinand/gemma-2b-it-tamil-v0.1-alpha Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type split args HellaSwag (10-Shot) hellaswag validation num_few_shot 10 type value name acc_norm 71.41 normalized accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=abhinand/gemma-2b-it-tamil-v0.1-alpha Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args MMLU (5-Shot) cais/mmlu all test num_few_shot 5 type value name acc 39.94 accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=abhinand/gemma-2b-it-tamil-v0.1-alpha Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args TruthfulQA (0-shot) truthful_qa multiple_choice validation num_few_shot 0 type value mc2 42.63 url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=abhinand/gemma-2b-it-tamil-v0.1-alpha Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args Winogrande (5-shot) winogrande winogrande_xl validation num_few_shot 5 type value name acc 64.96 accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=abhinand/gemma-2b-it-tamil-v0.1-alpha Open LLM Leaderboard task dataset metrics source type name text-generation Text Generation name type config split args GSM8k (5-shot) gsm8k main test num_few_shot 5 type value name acc 16.6 accuracy url name https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=abhinand/gemma-2b-it-tamil-v0.1-alpha Open LLM Leaderboard

Gemma 2B Tamil v0.1 Alpha [Experimental Release]

This is a Tamil instruction finetuned version of Google's Gemma 2B model. This is an experiment to see if Gemma can be adapted for Tamil without expanding vocabulary. While the responses may be rusty at times, it shows a lot of promise for a 2B parameter model.

Procedure:

1. The Gemma base model was continually pretrained on all available Tamil Wikipedia data for 3 epochs.
2. The updated model was then finetuned on a mix of English and Tamil alpaca datasets for 5 epochs.

Note: This project is currently under development (FOR TAMIL). The initial pretraining phase may not have been extensive enough, which suggests that the model's performance could improve by extending the pretraining on a larger dataset, such as CulturaX.

🏆 Benchmarks

This model outperforms Google's Gemma 2B base and instruct models on all benchmarks in Nous evaluation suite. It also surprisingly outperforms mlabonne/Gemmalpaca-2B (the best performing 2B model in benchmarks as of Feb 25, 2024) despite being a model aimed at language adaptation.

Model	Average	AGIEval	GPT4All	TruthfulQA	Bigbench
gemma-2b-it-tamil-v0.1-alpha📄	39.41	23.38	58.94	43.18	32.14
mlabonne/Gemmalpaca-2B 📄	38.39	24.48	51.22	47.02	30.85
google/gemma-2b-it 📄	36.1	23.76	43.6	47.64	29.41
google/gemma-2b 📄	34.26	22.7	43.35	39.96	31.03

Model description

• Model type: A 2B parameter GPT-like model finetuned on 100,000 samples consisting of an equal proportion of English and Tamil samples.
• Language(s): Bilingual. English and Tamil.
• License: Google Gemma Terms of Use
• Finetuned from model: abhinand/gemma-2b-tamil
• Training Precision: bfloat16
• Training Hardware: 4x Nvidia RTX 3090 GPUs
• Training Cost: $20

Support my work

If you appreciate this work and would like to support its continued development, consider buying me a coffee. Your support is invaluable and greatly appreciated.

Prompting Format [Alpaca]

Prompt Template Without Input

{system_prompt}

### Instruction:
{instruction or query}

### Response:
{response}

Prompt Template With Input

{system_prompt}

### Instruction:
{instruction or query}

### Input:
{input}

### Response:
{response}

Usage Note

It's important to note that the models have not undergone detoxification. Therefore, while they possess impressive linguistic capabilities, there is a possibility for them to generate content that could be deemed harmful or offensive. We urge users to exercise discretion and supervise the model's outputs closely, especially in public or sensitive applications.

Meet the Developers

Get to know the creators behind this innovative model and follow their contributions to the field:

• Abhinand Balachandran

We hope this model serves as a valuable tool in your NLP toolkit and look forward to seeing the advancements it will enable in the understanding and generation of the Tamil language.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	47.60
AI2 Reasoning Challenge (25-Shot)	50.09
HellaSwag (10-Shot)	71.41
MMLU (5-Shot)	39.94
TruthfulQA (0-shot)	42.63
Winogrande (5-shot)	64.96
GSM8k (5-shot)	16.60