README.md

metadata

language:
  - en
  - kn
license: mit
tags:
  - bilingual
  - kannada
  - english
metrics:
  - accuracy
pipeline_tag: text-generation
model-index:
  - name: Ambari-7B-base-v0.1-sharded
    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: 47.95
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fierysurf/Ambari-7B-base-v0.1-sharded
          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: 74.62
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fierysurf/Ambari-7B-base-v0.1-sharded
          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: 40.39
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fierysurf/Ambari-7B-base-v0.1-sharded
          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: 38.91
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fierysurf/Ambari-7B-base-v0.1-sharded
          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: 72.06
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fierysurf/Ambari-7B-base-v0.1-sharded
          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: 1.59
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=fierysurf/Ambari-7B-base-v0.1-sharded
          name: Open LLM Leaderboard

(This repo contains the sharded version of the original Ambari-7B model)

Ambari-7B-Base-v0.1 (sharded)

Overview

Ambari-7B-Base-v0.1 is the first bilingual English/Kannada model in the Ambari series, developed and released by Cognitivelab.in. Based on the Llama2 model by Meta, this 7B parameter model is the outcome of the pretraining stage, involving training on approximately 500 million new Kannada tokens.

Usage

To use the Ambari-7B-Base-v0.1 model, you can follow the example code below:

# Usage
import torch
from transformers import LlamaTokenizer, LlamaForCausalLM

model = LlamaForCausalLM.from_pretrained('Cognitive-Lab/Ambari-7B-Base-v0.1')
tokenizer = LlamaTokenizer.from_pretrained('Cognitive-Lab/Ambari-7B-Base-v0.1')

prompt = "ಕನ್ನಡದ ಇತಿಹಾಸವನ್ನು ವಿವರವಾಗಿ ತಿಳಿಸಿ"
inputs = tokenizer(prompt, return_tensors="pt")

# Generate
generate_ids = model.generate(inputs.input_ids, max_length=30)
decoded_output = tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]

print(decoded_output)

Important: The provided model serves as a foundation and is not designed for independent use. We strongly advise conducting finetuning tailored to your particular task(s) of interest before deploying it in a production environment. Feel free to customize the code according to your specific use case, ensuring that the model undergoes finetuning for optimal performance in your desired application.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	45.92
AI2 Reasoning Challenge (25-Shot)	47.95
HellaSwag (10-Shot)	74.62
MMLU (5-Shot)	40.39
TruthfulQA (0-shot)	38.91
Winogrande (5-shot)	72.06
GSM8k (5-shot)	1.59