README.md

metadata

language:
  - te
  - en
license: llama2
datasets:
  - Telugu-LLM-Labs/yahma_alpaca_cleaned_telugu_filtered_and_romanized
  - >-
    Telugu-LLM-Labs/teknium_GPTeacher_general_instruct_telugu_filtered_and_romanized
model-index:
  - name: Telugu-Llama2-7B-v0-Instruct
    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: 53.58
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Telugu-LLM-Labs/Telugu-Llama2-7B-v0-Instruct
          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: 78.33
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Telugu-LLM-Labs/Telugu-Llama2-7B-v0-Instruct
          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: 47.63
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Telugu-LLM-Labs/Telugu-Llama2-7B-v0-Instruct
          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: 43.26
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Telugu-LLM-Labs/Telugu-Llama2-7B-v0-Instruct
          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: 73.95
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Telugu-LLM-Labs/Telugu-Llama2-7B-v0-Instruct
          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: 20.39
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=Telugu-LLM-Labs/Telugu-Llama2-7B-v0-Instruct
          name: Open LLM Leaderboard

Telugu-Llama2-7B-v0-Instruct

This model is based on Telugu-Llama2-7B-v0-Base and hase been finetuned on instruction datasets:

1. yahma_alpaca_cleaned_telugu_filtered_and_romanized
2. teknium_GPTeacher_general_instruct_telugu_filtered_and_romanized

Input Text Format

### Instruction: {instruction}

### Input: {input}

## Response: {response}

Usage

With Romanized Telugu

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

model_name = "Telugu-LLM-Labs/Telugu-Llama2-7B-v0-Instruct"

tokenizer = AutoTokenizer.from_pretrained(model_name, padding_side="right")
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16).to(device)

instruction = "Krindi samaacharam prakaram google app eppudu release ayyindi?"
input ="Google News is a news aggregator service developed by Google. It presents a continuous flow of links to articles organized from thousands of publishers and magazines. Google News is available as an app on Android, iOS, and the Web. Google released a beta version in September 2002 and the official app in January 2006."

text = f"""Instruction: {instruction} \nInput: {input} \nResponse:"""

encodings = tokenizer(text, padding=True, return_tensors="pt")
encodings = encodings.to(device)

with torch.inference_mode():
    outputs = model.generate(encodings.input_ids, do_sample=False, max_new_tokens=500)

output = tokenizer.batch_decode(outputs.detach(), skip_special_tokens=True)

Sample Output:

1. September 2002 Google released a beta version of Google News.
2. January 2006 Google released the official version of Google News.

With Native Telugu

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

model_name = "Telugu-LLM-Labs/Telugu-Llama2-7B-v0-Instruct"

tokenizer = AutoTokenizer.from_pretrained(model_name, padding_side="right")
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16).to(device)

instruction = "కింది వచనాన్ని సంగ్రహించండి"
input="గూగుల్ వార్తలు అనేది గూగుల్ ద్వారా అభివృద్ధి చేయబడిన వార్తా అగ్రిగేటర్ సేవ. ఇది వేలకొద్దీ ప్రచురణకర్తలు మరియు మ్యాగజైన్‌ల నుండి నిర్వహించబడిన కథనాలకు నిరంతర లింక్‌లను అందిస్తుంది. గూగుల్ వార్తలు Android, iOS మరియు వెబ్‌లో యాప్‌గా అందుబాటులో ఉన్నాయి. గూగుల్ సెప్టెంబరు 2002లో బీటా వెర్షన్‌ను మరియు జనవరి 2006లో అధికారిక యాప్‌ను విడుదల చేసింది."

text = f"""Instruction: {instruction} \nInput: {input} \nResponse:"""

encodings = tokenizer(text, padding=True, return_tensors="pt")
encodings = encodings.to(device)

with torch.inference_mode():
    outputs = model.generate(encodings.input_ids, do_sample=False, max_new_tokens=500)

output = tokenizer.batch_decode(outputs.detach(), skip_special_tokens=True)

Sample Output:

1. గూగుల్ వార్తలు అనేది గూగుల్ ద్వారా అభివృద్ధి చేయబడిన వార్తా అగ్రిగేటర్ సేవ, వేలకొద్దీ ప్రచురణకర్తలు మరియు మ్యాగజైన్‌ల నుండి నిర్వహించబడిన కథనాలకు నిరంతర లింక్‌లను అందిస్తుంది.
2. గూగుల్ సెప్టెంబరు 2002లో బీటా వెర్షన్ మరియు జనవరి 2006లో అధికారిక యాప్ ను విడుదల చేసింది.

Developers:

The model is a collaborative effort by Ravi Theja and Ramsri Goutham. Feel free to DM either of us if you have any questions.

Note:

The model is quite sensitive to parameters and inputs and is not yet ready for production. It remains in the experimental phase, and we recommend using it accordingly.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	52.86
AI2 Reasoning Challenge (25-Shot)	53.58
HellaSwag (10-Shot)	78.33
MMLU (5-Shot)	47.63
TruthfulQA (0-shot)	43.26
Winogrande (5-shot)	73.95
GSM8k (5-shot)	20.39