README.md

metadata

language:
  - hi
license: llama2
model-index:
  - name: OpenHathi-7B-Hi-v0.1-Base
    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: 49.49
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=sarvamai/OpenHathi-7B-Hi-v0.1-Base
          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: 74.34
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=sarvamai/OpenHathi-7B-Hi-v0.1-Base
          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: 41.38
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=sarvamai/OpenHathi-7B-Hi-v0.1-Base
          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: 37.46
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=sarvamai/OpenHathi-7B-Hi-v0.1-Base
          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: 71.27
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=sarvamai/OpenHathi-7B-Hi-v0.1-Base
          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: 5.91
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=sarvamai/OpenHathi-7B-Hi-v0.1-Base
          name: Open LLM Leaderboard

This repository is the first model in the OpenHathi series of models that will be released by Sarvam AI. This is a 7B parameter, based on Llama2, trained on Hindi, English, and Hinglish. More details about the model, its training procedure, and evaluations can be found here.

Note: this is a base model and not meant to be used as is. We recommend first finetuning it on task(s) you are interested in.

# Usage
import torch
from transformers import LlamaTokenizer, LlamaForCausalLM

tokenizer = LlamaTokenizer.from_pretrained('sarvamai/OpenHathi-7B-Hi-v0.1-Base')
model = LlamaForCausalLM.from_pretrained('sarvamai/OpenHathi-7B-Hi-v0.1-Base', torch_dtype=torch.bfloat16)

prompt = "मैं एक अच्छा हाथी हूँ"
inputs = tokenizer(prompt, return_tensors="pt")

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

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	46.64
AI2 Reasoning Challenge (25-Shot)	49.49
HellaSwag (10-Shot)	74.34
MMLU (5-Shot)	41.38
TruthfulQA (0-shot)	37.46
Winogrande (5-shot)	71.27
GSM8k (5-shot)	5.91