rename model_weights_samples -> model_weights_preview_samples

- doing this before anyone else runs these migrations
- also added "squashed" documentation on the model_weights and
  associated tables
- also added (commented) good indices for model_weights

_sql/migrations/2023-12-15-044416_model_weights_preview_samples/down.sql

@@ -2,4 +2,4 @@
 -- noinspection SqlNoDataSourceInspectionForFile
 -- noinspection SqlResolveForFile
 
-DROP TABLE model_weights_samples;
+DROP TABLE model_weights_preview_samples;

_sql/migrations/2023-12-15-044416_model_weights_preview_samples/up.sql

@@ -2,7 +2,7 @@
 -- noinspection SqlNoDataSourceInspectionForFile
 -- noinspection SqlResolveForFile
 
-CREATE TABLE model_weights_samples (
+CREATE TABLE model_weights_preview_samples (
   -- Not used for anything except replication.
   id BIGINT(20) NOT NULL AUTO_INCREMENT,
 

_sql/migrations_squashed/model_weights/model_weights.sql

@@ -190,13 +190,12 @@ CREATE TABLE model_weights (
   PRIMARY KEY (id),
   UNIQUE KEY (token)
 
-  -- TODO:
-  -- KEY index_model_type (model_type),
-  -- KEY fk_maybe_vocoder_token (maybe_vocoder_token),
+  -- TODO (this set should be good to go):
+  -- KEY index_weights_type (weights_type),
+  -- KEY index_weights_category (weights_category),
   -- KEY fk_creator_user_token (creator_user_token),
   -- KEY index_creator_set_visibility (creator_set_visibility),
-  -- KEY index_private_bucket_hash (private_bucket_hash),
-  -- KEY index_is_public_listing_approved (is_public_listing_approved),
-  -- KEY index_is_locked_from_user_modification (is_locked_from_user_modification)
+  -- KEY fk_maybe_avatar_media_file_token (maybe_avatar_media_file_token),
+  -- KEY fk_maybe_cover_media_file_token (maybe_cover_media_file_token)
 
 ) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations_squashed/model_weights/model_weights_image_generation_details.sql

@@ -0,0 +1,50 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Every type of fine tuned model that users can upload.
+CREATE TABLE model_weights_image_generation_details (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Simultaneous primary and foreign key.
+  -- This should match a single record in the `model_weights` table.
+  -- This turns this table into a polymorphic extension table.
+  model_weights_token VARCHAR(32) NOT NULL,
+
+
+  -- ========== IMAGE GENERATION MODEL METADATA ==========
+
+  -- RVC (v2) specific - whether the model has an index file associated with it
+  -- These files improve the quality of the results.
+  vc_has_index_file BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- ========== VECTOR CLOCK ==========
+
+  -- Incremented with every update.
+  version INT NOT NULL DEFAULT 0,
+
+
+  -- ========== TIMESTAMPS ==========
+
+  created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
+
+  -- If this is removed by a mod.
+  -- It completely disappears from the system.
+  user_deleted_at TIMESTAMP NULL,
+  mod_deleted_at TIMESTAMP NULL,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (model_weights_token)
+  -- TODO:
+  -- KEY index_model_type (model_type),
+  -- KEY fk_maybe_vocoder_token (maybe_vocoder_token),
+  -- KEY fk_creator_user_token (creator_user_token),
+  -- KEY index_creator_set_visibility (creator_set_visibility),
+  -- KEY index_private_bucket_hash (private_bucket_hash),
+  -- KEY index_is_public_listing_approved (is_public_listing_approved),
+  -- KEY index_is_locked_from_user_modification (is_locked_from_user_modification)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations_squashed/model_weights/model_weights_preview_samples.sql

@@ -0,0 +1,47 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+CREATE TABLE model_weights_preview_samples (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- ========== WEIGHTS CLASSIFICATION ==========
+
+  -- Foreign key to model weights.
+  -- This is in a composite index with media_file_token to make sure a media file only applies
+  -- to a given model once.
+  model_weight_token VARCHAR(32) NOT NULL,
+
+  -- Foreign key to media files
+  -- This is in a composite index with model_weight_token to make sure a media file only applies
+  -- to a given model once.
+  media_file_token VARCHAR(32) NOT NULL,
+
+  -- Information about who set the sample
+  --  * automation
+  --  * author
+  --  * community
+  set_by VARCHAR(16) NOT NULL,
+
+  -- Incremented with every update.
+  version INT NOT NULL DEFAULT 0,
+
+  -- ========== TIMESTAMPS ==========
+
+  created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
+
+  -- If this is removed by a mod.
+  -- It completely disappears from the system.
+  user_deleted_at TIMESTAMP NULL,
+  mod_deleted_at TIMESTAMP NULL,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (model_weight_token, media_file_token),
+
+  KEY index_model_weight_token (model_weight_token),
+  KEY index_media_file_token (media_file_token)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations_squashed/model_weights/model_weights_tts_details.sql

@@ -0,0 +1,73 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Extra metadata on tts models
+CREATE TABLE model_weights_tts_details (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Simultaneous primary and foreign key.
+  -- This should match a single record in the `model_weights` table.
+  -- This turns this table into a polymorphic extension table.
+  model_weights_token VARCHAR(32) NOT NULL,
+
+  -- ========== TTS MODEL METADATA ==========
+
+  -- The full IETF BCP47 language tag (eg. en, en-US, es-419, ja-JP, pt, etc.)
+  -- (Not that it matters apart from categorization, since voice conversion is universal.)
+  ietf_language_tag VARCHAR(64) NOT NULL DEFAULT 'en',
+
+  -- The IETF BCP47 language tag's primary language subtag (eg. "es-419" becomes "es")
+  -- (Not that it matters apart from categorization, since voice conversion is universal.)
+  ietf_primary_language_subtag VARCHAR(12) NOT NULL DEFAULT 'en',
+
+  -- This is for Tacotron2, specifically.
+  -- The name of the type of pipeline to use, eg "legacy_fakeyou", "legacy_vocodes", "english_v1", etc.
+  -- New pipelines may be added at any point in the future and the source of truth for them will mostly
+  -- live in the `storyteller-ml` repository.
+  -- Older models and brand new models will have this field set to null, and we'll infer which value to
+  -- lazily backfill based on the creation date and language setting.
+  -- tts_text_pipeline_type VARCHAR(64) DEFAULT NULL,
+
+  -- If true, multiply the mel outputs before being vocoded by a globally default constant.
+  use_default_mel_multiply_factor BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- If not null, multiply the mel outputs before being vocoded by this value.
+  -- This is used instead of the default if `tts_use_default_mel_multiply_factor` is
+  -- set (ie. `tts_use_default_mel_multiply_factor` is ignored and this custom value is used instead)
+  maybe_custom_mel_multiply_factor DOUBLE DEFAULT NULL,
+
+  -- If set, use a different pretrained vocoder.
+  -- If not set, use the website default. (Currently HifiGan-SS)
+  maybe_default_pretrained_vocoder VARCHAR(64) DEFAULT NULL,
+
+  -- If set, we'll use a custom vocoder (from table vocoder_models) instead of a default vocoder.
+  -- If not set, we'll use `maybe_default_pretrained_vocoder` (or the website default).
+  maybe_custom_vocoder_token VARCHAR(32) DEFAULT NULL,
+
+
+  -- ========== VECTOR CLOCK ==========
+
+  -- Incremented with every update.
+  version INT NOT NULL DEFAULT 0,
+
+
+  -- ========== TIMESTAMPS ==========
+
+  created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (model_weights_token)
+  -- TODO:
+  -- KEY index_model_type (model_type),
+  -- KEY fk_maybe_vocoder_token (maybe_vocoder_token),
+  -- KEY fk_creator_user_token (creator_user_token),
+  -- KEY index_creator_set_visibility (creator_set_visibility),
+  -- KEY index_private_bucket_hash (private_bucket_hash),
+  -- KEY index_is_public_listing_approved (is_public_listing_approved),
+  -- KEY index_is_locked_from_user_modification (is_locked_from_user_modification)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations_squashed/model_weights/model_weights_vocoder_details.sql

@@ -0,0 +1,50 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Every type of fine tuned model that users can upload.
+CREATE TABLE model_weights_vocoder_details (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Simultaneous primary and foreign key.
+  -- This should match a single record in the `model_weights` table.
+  -- This turns this table into a polymorphic extension table.
+  model_weights_token VARCHAR(32) NOT NULL,
+
+
+  -- ========== VOCODER MODEL METADATA ==========
+
+  -- Whether the vocoder is recommended for use
+  -- vocoder_is_staff_recommended BOOLEAN NOT NULL DEFAULT FALSE,
+
+
+  -- ========== VECTOR CLOCK ==========
+
+  -- Incremented with every update.
+  version INT NOT NULL DEFAULT 0,
+
+
+  -- ========== TIMESTAMPS ==========
+
+  created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
+
+  -- If this is removed by a mod.
+  -- It completely disappears from the system.
+  user_deleted_at TIMESTAMP NULL,
+  mod_deleted_at TIMESTAMP NULL,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (model_weights_token)
+  -- TODO:
+  -- KEY index_model_type (model_type),
+  -- KEY fk_maybe_vocoder_token (maybe_vocoder_token),
+  -- KEY fk_creator_user_token (creator_user_token),
+  -- KEY index_creator_set_visibility (creator_set_visibility),
+  -- KEY index_private_bucket_hash (private_bucket_hash),
+  -- KEY index_is_public_listing_approved (is_public_listing_approved),
+  -- KEY index_is_locked_from_user_modification (is_locked_from_user_modification)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;

_sql/migrations_squashed/model_weights/model_weights_voice_conversion_details.sql

@@ -0,0 +1,62 @@
+-- noinspection SqlDialectInspectionForFile
+-- noinspection SqlNoDataSourceInspectionForFile
+-- noinspection SqlResolveForFile
+
+-- Every type of fine tuned model that users can upload.
+CREATE TABLE model_weights_voice_conversion_details (
+  -- Not used for anything except replication.
+  id BIGINT(20) NOT NULL AUTO_INCREMENT,
+
+  -- Simultaneous primary and foreign key.
+  -- This should match a single record in the `model_weights` table.
+  -- This turns this table into a polymorphic extension table.
+  model_weights_token VARCHAR(32) NOT NULL,
+
+
+  -- ========== VOICE CONVERSION MODEL METADATA ==========
+
+  -- The full IETF BCP47 language tag (eg. en, en-US, es-419, ja-JP, pt, etc.)
+  -- (Not that it matters apart from categorization, since voice conversion is universal.)
+  ietf_language_tag VARCHAR(64) NOT NULL DEFAULT 'en',
+
+  -- The IETF BCP47 language tag's primary language subtag (eg. "es-419" becomes "es")
+  -- (Not that it matters apart from categorization, since voice conversion is universal.)
+  ietf_primary_language_subtag VARCHAR(12) NOT NULL DEFAULT 'en',
+
+  -- RVC (v2) specific - whether the model has an index file associated with it
+  -- These files improve the quality of the results.
+  has_index_file BOOLEAN NOT NULL DEFAULT FALSE,
+
+  -- Pitch algorithm to use by default
+  default_pitch_algorithm VARCHAR(64) DEFAULT NULL,
+
+
+  -- ========== VECTOR CLOCK ==========
+
+  -- Incremented with every update.
+  version INT NOT NULL DEFAULT 0,
+
+
+  -- ========== TIMESTAMPS ==========
+
+  created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
+  updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
+
+  -- If this is removed by a mod.
+  -- It completely disappears from the system.
+  user_deleted_at TIMESTAMP NULL,
+  mod_deleted_at TIMESTAMP NULL,
+
+  -- INDICES --
+  PRIMARY KEY (id),
+  UNIQUE KEY (model_weights_token)
+  -- TODO:
+  -- KEY index_model_type (model_type),
+  -- KEY fk_maybe_vocoder_token (maybe_vocoder_token),
+  -- KEY fk_creator_user_token (creator_user_token),
+  -- KEY index_creator_set_visibility (creator_set_visibility),
+  -- KEY index_private_bucket_hash (private_bucket_hash),
+  -- KEY index_is_public_listing_approved (is_public_listing_approved),
+  -- KEY index_is_locked_from_user_modification (is_locked_from_user_modification)
+
+) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;