Embeddings

25 problems

• Fuse token + positional embeddingsEasy
• The input layer — token lookup + positionalMedium
• Autoencoder forward passEasy
• BERTScore (Greedy Token-Cosine F1)Medium
• Binary Quantization + Hamming RetrievalMedium
• Token embedding lookupEasy
• CLIP InfoNCE LossMedium
• Contrastive lossMedium
• CLIP — zero-shot classificationMedium
• The block_size hard capEasy
• How big is GPT-2's embedding table?Medium
• Embeddings.forward — token + position (batched)Medium
• One-hot tokens are all equidistantEasy
• Why one-hot inputs blow up memoryEasy
• Learned positional lookup P[:T]Easy
• Weight tying — the LM head reuses EMedium
• Hard Negative Mining for RetrieverMedium
• Jaccard similarityEasy
• Sentence Embedding (Masked Mean-Pool)Easy
• Matryoshka Embedding TruncationMedium
• Pairwise cosine-similarity matrixEasy
• Random-Hyperplane LSH (Cosine)Medium
• Triplet lossMedium
• Two-Tower Top-k RetrievalEasy
• ViT Patch EmbeddingMedium