from-scratch
asr_rnnt_eou_from_scratch
NeMo ASR-EOU 训练脚本解读与论文出处梳理
目标文件：examples/asr/asr_eou/speech_to_text_rnnt_eou_train.py链接：https://github.com/NVIDIA-NeMo/NeMo/blob/main/examples/asr/asr_eou/speech_to_text_rnnt_eou_train.py
这份脚本本身是一个 训练入口脚本（Hydra + PyTorch Lightning），核心功能是：按配置创建 EncDecRNNTBPEEOUModel，并支持从已有 .nemo 初始化、添加/训练 adapter，以及在“词表扩展（新增 <EOU>/<EOB>）”时做权重迁移。
下面按“它用到的技术点 → 在代码/配置里怎么体现 → 原始论文出处”总结。
1) ASR-EOU：把“端点/话轮信息”并入 ASR（<EOU>, <EOB>）
它做什么：
除了输出转写文本外，还让模型在时间轴上预测：
EOU：End Of Utterance（一句话结束）… See the full description on the dataset page: https://huggingface.co/datasets/echodict/asr_rnnt_eou_from_scratch.transformer-from-scratch-tutorial
Implementing Transformer from Scratch: A Step-by-Step Guide
This repository provides a detailed guide and implementation of the Transformer architecture from the "Attention Is All You Need" paper. The implementation focuses on understanding each component through clear code, comprehensive testing, and visual aids.
For implementions of more recent architectural innovations from DeepSeek, see the Related Implementations section.
Quick Start
View the complete… See the full description on the dataset page: https://huggingface.co/datasets/bird-of-paradise/transformer-from-scratch-tutorial.cjk-tokenizer-from-scratch
CJK Tokenizer From Scratch
Offline research dump for building Chinese / Japanese tokenizers from scratch.
Download tokenizer.zip (≈464 MB), then:
unzip tokenizer.zip
Contents:
papers/ — 41 free arXiv / ACL PDFs (BPE, WordPiece, Unigram, SentencePiece, Chinese SubChar, Japanese MeCab/Sudachi bake-off, …)
github/ — 23 shallow-cloned repos (minbpe, sentencepiece, tokenizers, jieba, mecab, Sudachi, …)
docs/ — Hugging Face NLP course chapter 6 (train BPE / WordPiece / Unigram)… See the full description on the dataset page: https://huggingface.co/datasets/aicrazy124/cjk-tokenizer-from-scratch.turkish-sft-from-scratch-120k
Turkish SFT From Scratch 120K
Sıfırdan üretilmiş, kategori kontrollü Türkçe SFT dataset'i. Eski/temizlenmiş datasetlerden satır kopyalanmadı.
Kapsam
12 kategori x 10,000 örnek = 120,000 örnek:
instruction-following
qa
summarization
cot
multi-turn-dialogue
rewriting
text-classification
error-correction
formal-writing
translation
code-explanation
creative-writing
Doğrulamalar
Canonical messages formatı: system/user/assistant.
Exact duplicate hash kontrolü.… See the full description on the dataset page: https://huggingface.co/datasets/kilicai/turkish-sft-from-scratch-120k.turkish-sft-from-scratch-150k-extended
Turkish SFT From Scratch 150K Extended
kilicai/turkish-sft-from-scratch-120k üzerine 30K akıl yürütme, görev takibi ve analiz verisi eklenmiş genişletilmiş sürüm.
Audit
{
"rows": 150000,
"base_rows": 120000,
"extension_rows": 30000,
"duplicates_removed_on_merge": 0,
"categories": {
"formal-writing": 10000,
"rewriting": 10000,
"text-classification": 10000,
"instruction-following": 10000,
"translation": 10000,
"cot": 10000… See the full description on the dataset page: https://huggingface.co/datasets/kilicai/turkish-sft-from-scratch-150k-extended.diffsion_from_scratch
Dataset Card for "diffsion_from_scratch"
More Information needed