ak-lm-small / README.md
endurasolution's picture
Update README.md
5730777 verified
|
Raw
History Blame Contribute Delete
961 Bytes

ak-lm-small

Tiny pretrained Malayalam language model (75M parameters)

AK-LM-SMALL (Akshara Language Model Small) is a 75M small Malayalam GPT style Model ,Build from Scratch

Developed by EnduraSolution.

Overview

  • GPT-style decoder-only Transformer
  • 75 Million parameters
  • Custom Malayalam SentencePiece tokenizer
  • Vocabulary: 24,000
  • Context length: 256

Training Data

  • IndicCorp Malayalam
  • Malayalam Wikipedia
  • Kerala-focused corpus
  • Small English Wikipedia subset

Intended Uses

  • Malayalam text generation
  • Text completion
  • Research
  • Fine-tuning

Limitations

This is a pretrained base model.

It is not instruction tuned.

Example

from transformers import GPT2LMHeadModel
import sentencepiece as spm

sp = spm.SentencePieceProcessor(model_file='ml_sp.model')
model = GPT2LMHeadModel.from_pretrained('endurasolution/ak-lm-small')

License

Apache-2.0


Made with ❤️ by EnduraSolution