nicolay-r/book-persona-retriever

Issues

• 📕 Further advances towards LLM

  #61 opened a month ago
  0

• `README.md` -- Move dependencies into Limitation Section

  #60 opened 4 months ago
  0

• Remove `pg19` API

  #59 opened 5 months ago
  0

• Remove legacy parts and leave them into separated project branches.

  #58 opened 5 months ago
  0

• :wrench: Refurbish supplementary `ParlAI` task description repository

  #57 opened 5 months ago
  0

• :fire: Release 1.0

  #56 opened 5 months ago
  0

• Rename `my_api` with `ldc_api`

  #55 opened 5 months ago
  0

• `plot` -- X axis scaling :sparkles:

  #54 opened 6 months ago
  0

• `Readme.md` aligning with the final paper

  #53 opened 5 months ago
  0

• Provide `spectrums` and `tsne` output into `README.md`

  #52 opened 6 months ago
  0

• `implicit` default `gpu` support may lead to issues in final matrix serialization to `npz`

  #51 opened 7 months ago
  0

• Speakers diagram has `{` entries mentioning

  #50 opened 6 months ago
  0

• provide google-colab notebook

  #49 opened 7 months ago
  1

• Project accessibility enancement :sparkles:

  #48 opened 6 months ago
  0

• :warning: Refactor Dataset Documentation

  #47 opened 7 months ago
  0

• List of Personalization Schemes :exclamation:

  #46 opened 7 months ago
  1

• Remove models

  #45 opened a year ago
  0

• API polishing

  #44 opened a year ago
  0

• :exclamation: `iter_paragraphs_with_n_speakers` -- implementation relies on that all the speakers are separated words

  #43 opened a year ago
  0

• :exclamation: `iter_paragraphs_with_n_speakers` -- there were no support of multiple mentions of the same speaker.

  #42 opened a year ago
  0

• Core generalization and refactoring

  #41 opened 7 months ago
  1

• Documentation -- original books count and resources

  #40 opened a year ago
  0

• GPT-3 (Inference Only)

  #38 opened a year ago
  0

• GPT-2 experiments

  #37 opened a year ago
  0

• ParlAI dataset construction script relies both on SQLite and textual version of the same datset

  #39 opened 6 months ago
  1

• Dialogue lines were treated with meta-information

  #36 opened a year ago
  0

• Filtering response utterances that at least mention $k$ speakers

  #35 opened a year ago
  1

• Documentation -- filtering dialogues for Q/R pairs

  #34 opened a year ago
  0

• Move the actual algorithm of lexicon construction into a sepearated `core` file

  #33 opened a year ago
  0

• :memo: Documentation -- source of spectrums for annotation

  #32 opened a year ago
  0

• :memo: Documentation -- Lexicon construction for utterances annotation (Prefix Analysis)

  #31 opened a year ago
  0

• The amount of training samples reprent ~24% of the same amount in ALOHA studies

  #30 opened a year ago
  0

• Why we have a few amount of dialogs per speaker than expected 100? (v. 3.1)

  #29 opened a year ago
  0

• `Uniform` modeling across all utterances from all books of the same set

  #28 opened a year ago
  0

• Limitation on HLA utilized for selection; auhtors consider `40` highest HLA per speaker, among which we perform random selection

  #27 opened a year ago
  1

• Authors use non-intersected speaker sets!

  #26 opened a year ago
  0

• Speaker set is expected to be smaller

  #25 opened a year ago
  0

• Speed up `cos` calculation

  #24 opened a year ago
  0

• `iter_dataset_dialogs` -- provide `dialog_id` from this method

  #23 opened 5 months ago
  1

• Dataset V4.0

  #22 opened a year ago
  0

• In some rare cases speaker might be missed in list of the clusters

  #21 opened a year ago
  0

• Wrap `select_from_table(...)` into more simple API

  #20 opened 7 months ago
  0

• Dataset V3.1

  #19 opened a year ago
  0

• Use Masked `_` character names instead of their IDS.

  #18 opened a year ago
  0

• Aloha Users Factorization

  #17 opened a year ago
  0

• Speakers within the book should be also better separated with the chosen traits

  #16 opened a year ago
  1

• Training dataset with spectrums could be enhanced with better selection of candidates

  #15 opened a year ago
  0

• In some rare cases spectrums for speakers might be missed in the result ParlAI dataset

  #14 opened 7 months ago
  1

• `speaker_ids` -- generating Sentiment datasets results in the equal set of attrubutes for all speakers

  #13 opened a year ago
  0

• Dataset V3

  #12 opened a year ago
  0