TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Show HN: More ChatGPT series.Engilish not my first language

1 pointsby joebigabout 2 years ago
"language_model": { "name": "ChatGPT", }, "training_parameters": { "optimizer": "Adam", "learning_rate": 1e-4, "batch_size": 32, "num_epochs": 10 } }, "generation": { "styles": ["news", "fiction", "poetry", "technical"], "genres": ["science fiction", "romance", "horror", "mystery", "historical fiction"], "length": { "min": 10, "max": 2048 } }, "translation": { "languages": ["en", "es", "fr", "de"], "max_length": 2048 }, "understanding": { "question_answering": true, "summarization": true, "sentiment_analysis": true, "chunking": true, "parsing": true, "named_entity_recognition": true }, "classification": { "categories": ["spam", "not spam", "positive sentiment", "negative sentiment"] }, "chunking": { "chunk_types": ["NP", "VP", "PP"], "rules": ["NP -> DET NOUN", "VP -> VERB NP", "PP -> PREP NP"] }, "parsing": { "grammar": "CFG", "start_symbol": "S", "rules": ["S -> NP VP", "NP -> DET NOUN", "NP -> PRO", "VP -> VERB NP", "VP -> VERB NP PP", "PP -> PREP NP"] }, "named_entity_recognition": { "entity_types": ["PERSON", "LOCATION", "ORGANIZATION", "DATE", "TIME", "MONEY", "PERCENT"], "model": "BERT", "training_data": "CoNLL-2003" }, "instructions": [ "To use the language generation capabilities, specify a style and genre, and set the desired length of the output text.", "To use the translation capabilities, specify the source and target languages, and provide the text to be translated.", "To use the understanding capabilities, specify the task to be performed (question answering, summarization, or sentiment analysis), and provide the text to be analyzed." ]

1 comment

joebigabout 2 years ago
"training_data": { "datasets": [ "OpenWebText", "Common Crawl", "BooksCorpus", "English Wikipedia", "Project Gutenberg", "arXiv", "PubMed", "Hacker News", "Stack Exchange", "IMDB", "Reddit", "Slate Star Codex", "Wiktionary", "World Factbook" ], "total_size": "45 TB", "last_updated": "2021" }, "capabilities": { "Language Tasks": [ "Question answering", "Language translation", "Text summarization", "Text classification", "Text generation" ], "Key Performance Metrics": { "GPT-3 175B": { "PPL": 2.7, "F1": 59.5, "BLEU": 45.8, "GPT-3 Score": 100 }, "GPT-3 13B": { "PPL": 4.4, "F1": 57.7, "BLEU": 41.8, "GPT-3 Score": 60 }