metadata
datasets:
- imdb (Movie corpus for Domain Adaptive Pretraining)
- cornell_movie_dialogue
- MIT Movie (NER Dataset)
language:
- English
thumbnail: null
tags:
- roberta
- roberta-base
- token-classification
- NER
- named-entities
- BIO
- movies
- DAPT
license: cc-by-4.0
Movie Roberta + Movies NER Task
Objective: This is Roberta Base + Movie DAPT --> trained for the NER task using MIT Movie Dataset https://huggingface.co/thatdramebaazguy/movie-roberta-base was used as the MovieRoberta.
model_name = "thatdramebaazguy/movie-roberta-MITmovieroberta-base-MITmovie"
pipeline(model=model_name, tokenizer=model_name, revision="v1.0", task="ner")
Overview
Language model: roberta-base
Language: English
Downstream-task: NER
Training data: MIT Movie
Eval data: MIT Movie
Infrastructure: 2x Tesla v100
Code: See example
Hyperparameters
Num examples = 6253
Num Epochs = 5
Instantaneous batch size per device = 64
Total train batch size (w. parallel, distributed & accumulation) = 128
Performance
Eval on MIT Movie
- epoch = 5.0
- eval_accuracy = 0.9472
- eval_f1 = 0.8876
- eval_loss = 0.2211
- eval_mem_cpu_alloc_delta = 3MB
- eval_mem_cpu_peaked_delta = 2MB
- eval_mem_gpu_alloc_delta = 0MB
- eval_mem_gpu_peaked_delta = 38MB
- eval_precision = 0.887
- eval_recall = 0.8881
- eval_runtime = 0:00:03.73
- eval_samples = 1955
- eval_samples_per_second = 523.095
Github Repo: