README.md

metadata

language:
  - en
  - hi
license: llama2
library_name: transformers
tags:
  - hindi
  - 'english '
  - Bilingual
datasets:
  - sarvamai/samvaad-hi-v1
pipeline_tag: text-generation
model-index:
  - name: Gaja-v2.00
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: AI2 Reasoning Challenge (25-Shot)
          type: ai2_arc
          config: ARC-Challenge
          split: test
          args:
            num_few_shot: 25
        metrics:
          - type: acc_norm
            value: 51.79
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=damerajee/Gaja-v2.00
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: HellaSwag (10-Shot)
          type: hellaswag
          split: validation
          args:
            num_few_shot: 10
        metrics:
          - type: acc_norm
            value: 75.79
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=damerajee/Gaja-v2.00
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU (5-Shot)
          type: cais/mmlu
          config: all
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 40.69
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=damerajee/Gaja-v2.00
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: TruthfulQA (0-shot)
          type: truthful_qa
          config: multiple_choice
          split: validation
          args:
            num_few_shot: 0
        metrics:
          - type: mc2
            value: 41.5
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=damerajee/Gaja-v2.00
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: Winogrande (5-shot)
          type: winogrande
          config: winogrande_xl
          split: validation
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 71.9
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=damerajee/Gaja-v2.00
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GSM8k (5-shot)
          type: gsm8k
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 0.23
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=damerajee/Gaja-v2.00
          name: Open LLM Leaderboard

Model

🐘 Gaja

Gaja is a Hindi/Hinglish chat model, initially trained on SarvamAI's OpenHathi model and further fine-tuned for conversational interactions.

Additional Information

• It outperforms Airavata, AI4Bharat's chat version, on Huggingface OpenLLM benchmark suite.
• It was fine-tuned on only 5k samples

Inference

hey guys thanks to Bhabha AI, you guys can finally try my model

Additional Information

• The code for this can be found in The github code - Github

💬 Prompt template

<|im_start|>user
{}<|im_end|> 
<|im_start|>assistant
{}<|im_end|> 

😎 Features:

• Language Support: Gaja is designed to understand and generate responses in both Hindi and Hinglish, catering to a diverse range of users.
• Base Model: Built upon SarvamAI's OpenHathi model, Gaja inherits its foundational capabilities while being optimized for conversational tasks.
• Fine-tuning: Gaja has undergone fine-tuning specifically for chat-based interactions, enhancing its ability to engage in meaningful conversations with users.
• Experimental Platform: With its flexibility and adaptability, Gaja serves as a valuable platform for conducting experiments and exploring innovative approaches to chatbot development.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	46.98
AI2 Reasoning Challenge (25-Shot)	51.79
HellaSwag (10-Shot)	75.79
MMLU (5-Shot)	40.69
TruthfulQA (0-shot)	41.50
Winogrande (5-shot)	71.90
GSM8k (5-shot)	0.23