sajalkuikel/eng-to-hindi-seq2seq

English -> Hindi Seq2Seq Translation (LSTM Encoder-Decoder)

A word-level LSTM encoder-decoder trained on a Hindi-English parallel corpus.

Files

• eng_to_hindi_seq2seq.keras - full training model (encoder + decoder combined)
• encoder_model.keras / decoder_model.keras - inference-ready encoder/decoder
• eng_tokenizer.pkl / hin_tokenizer.pkl - Keras Tokenizer objects (pickled)
• model_config.pkl - dict with max_length_eng, max_length_hin, latent_dim

Usage

import pickle, numpy as np, tensorflow as tf
from huggingface_hub import hf_hub_download

repo_id = "sajalkuikel/eng-to-hindi-seq2seq"
encoder_model = tf.keras.models.load_model(hf_hub_download(repo_id, "encoder_model.keras"))
decoder_model = tf.keras.models.load_model(hf_hub_download(repo_id, "decoder_model.keras"))
eng_tokenizer = pickle.load(open(hf_hub_download(repo_id, "eng_tokenizer.pkl"), "rb"))
hin_tokenizer = pickle.load(open(hf_hub_download(repo_id, "hin_tokenizer.pkl"), "rb"))
config = pickle.load(open(hf_hub_download(repo_id, "model_config.pkl"), "rb"))

See the linked Space for a live demo with full inference code.