Issues

• threshold_ema_dead_code is in units of an EMA of per-batch counts, so a value copied from large-batch configs resamples most of the codebook every step at small batches

  lucidrains/vector-quantize-pytorch#257 · Filter by author raunak2007 opened on Sep 24, 2026
  2

• Learning codebooks also updates the encoder ?

  lucidrains/vector-quantize-pytorch#236 · Filter by author gdoras opened on Dec 12, 2025
  3

• Multi-Scale Residual Vector Quantization

  enhancement
  lucidrains/vector-quantize-pytorch#225 · Filter by author mahdip72 opened on Oct 4, 2025
  12

• Which technique works?

  lucidrains/vector-quantize-pytorch#216 · Filter by author realwenlongwang opened on May 3, 2025
  2

• SimVQ trains Failed!

  lucidrains/vector-quantize-pytorch#212 · Filter by author lovekittynine opened on Apr 18, 2025

• about BSQ Loss

  lucidrains/vector-quantize-pytorch#199 · Filter by author VJJJJJJ1 opened on Mar 6, 2025
  1

• FSQ Oddness

  lucidrains/vector-quantize-pytorch#194 · Filter by author zaptrem opened on Feb 6, 2025
  5

• LFQ: why can't we have projections?

  lucidrains/vector-quantize-pytorch#192 · Filter by author denadai2 opened on Feb 4, 2025

• Building intuition about the latent quantizer

  lucidrains/vector-quantize-pytorch#190 · Filter by author mikesol opened on Jan 28, 2025

• Residual simVQ loss

  lucidrains/vector-quantize-pytorch#184 · Filter by author mzc2113391 opened on Dec 26, 2024

• About multi-head vq

  lucidrains/vector-quantize-pytorch#183 · Filter by author FacePoluke opened on Dec 23, 2024

• Extracting learnt embeddings for analysis

  lucidrains/vector-quantize-pytorch#182 · Filter by author rohanbanerjee opened on Dec 20, 2024

• DDP: RuntimeError: shape mismatch: value tensor of shape [535, 256] cannot be broadcast to indexing result of shape [537, 256]

  lucidrains/vector-quantize-pytorch#169 · Filter by author forwiat opened on Nov 6, 2024
  3

• output dimensions for vector quantize

  lucidrains/vector-quantize-pytorch#168 · Filter by author dorsa-zeinali opened on Oct 28, 2024
  2

• What is the difference between Grouped VQ and Multi-codebook VQ when used for Autoregressive LLM training?

  lucidrains/vector-quantize-pytorch#157 · Filter by author JohnHerry opened on Aug 30, 2024

• Difference between FSQ and LFQ

  lucidrains/vector-quantize-pytorch#152 · Filter by author JasonShen-SH opened on Aug 6, 2024
  2

• LatentQuantize exploding loss

  lucidrains/vector-quantize-pytorch#151 · Filter by author jbmaxwell opened on Aug 1, 2024
  2

• Seeking clarifications regarding learnable codebook

  lucidrains/vector-quantize-pytorch#147 · Filter by author anupsingh15 opened on Jul 17, 2024

• Bug in FSQ, with return_indices to False and image-type feature input

  lucidrains/vector-quantize-pytorch#139 · Filter by author MisterBourbaki opened on Jun 20, 2024

• How to generate image using this repository

  lucidrains/vector-quantize-pytorch#134 · Filter by author neilwen987 opened on Jun 13, 2024

• Why do I get almost the same codes after the 1st batch?

  lucidrains/vector-quantize-pytorch#131 · Filter by author tanyz0208 opened on May 23, 2024
  1

• Some clarification required

  lucidrains/vector-quantize-pytorch#130 · Filter by author inspirit opened on May 19, 2024
  3

• Will the 16 or 32 codebook dimension hamper the representational capacity of autoencoders?

  lucidrains/vector-quantize-pytorch#127 · Filter by author yilinliu77 opened on May 13, 2024
  2

• codebook merge

  lucidrains/vector-quantize-pytorch#126 · Filter by author zhouilu opened on May 8, 2024
  1

• Examples Codebook Utilization does not generalize

  lucidrains/vector-quantize-pytorch#109 · Filter by author AmitMY opened on Feb 25, 2024