NLP

30 problems

• Teacher forcing — next-token targetsMedium
• Map a model to its transformer familyEasy
• Pretraining objective per familyEasy
• Bag-of-words encodingEasy
• BERTScore (Greedy Token-Cosine F1)Medium
• BLEU score (unigram)Medium
• BM25 ScoringMedium
• BPE: apply mergesMedium
• Train BPE merges from a corpusHard
• Byte-level UTF-8 tokenizerMedium
• The autoregressive generation loopMedium
• How big is the nucleus?Easy
• Levenshtein Edit DistanceEasy
• Longest Common SubsequenceMedium
• Sentence Embedding (Masked Mean-Pool)Easy
• METEOR scoreMedium
• Perplexity from log-probsEasy
• PII Detection & RedactionMedium
• Planner Output ParserEasy
• Pointwise mutual information (PMI)Medium
• QA Token F1 / Exact MatchMedium
• Recursive Chunking with OverlapMedium
• ROUGE-1 scoreEasy
• Build a vocabularyEasy
• Encode a sentence (with UNK)Easy
• One-hot encode a token sequenceEasy
• Sequence padding & maskingEasy
• TF-IDFMedium
• Unigram probability from a corpusEasy
• WordPiece Tokenization (Greedy Longest-Match)Medium