SillyTavern/public/scripts/extensions/vectors/settings.html

<div class="vectors_settings">
    <div class="inline-drawer">
        <div class="inline-drawer-toggle inline-drawer-header">
            <b>Vector Storage</b>
            <div class="inline-drawer-icon fa-solid fa-circle-chevron-down down"></div>
        </div>
        <div class="inline-drawer-content">
            <div class="flex-container flexFlowColumn">
                <label for="vectors_source">
                    Vectorization Source
                </label>
                <select id="vectors_source" class="text_pole">
                    <option value="cohere">Cohere</option>
                    <option value="extras">Extras</option>
                    <option value="palm">Google MakerSuite</option>
                    <option value="llamacpp">llama.cpp</option>
                    <option value="transformers">Local (Transformers)</option>
                    <option value="mistral">MistralAI</option>
                    <option value="nomicai">NomicAI</option>
                    <option value="ollama">Ollama</option>
                    <option value="openai">OpenAI</option>
                    <option value="togetherai">TogetherAI</option>
                    <option value="vllm">vLLM</option>
                </select>
            </div>
            <div class="flex-container flexFlowColumn" id="ollama_vectorsModel">
                <label for="vectors_ollama_model">
                    Vectorization Model
                </label>
                <input id="vectors_ollama_model" class="text_pole" type="text" placeholder="Model tag, e.g. llama3" />
                <label for="vectors_ollama_keep" class="checkbox_label" title="When checked, the model will not be unloaded after use.">
                    <input id="vectors_ollama_keep" type="checkbox" />
                    <span>Keep model in memory</span>
                </label>
                <i>
                    Hint: Download models and set the URL in the API connection settings.
                </i>
            </div>
            <div class="flex-container flexFlowColumn" id="llamacpp_vectorsModel">
                <span>
                    The server MUST be started with the <code>--embedding</code> flag to use this feature!
                </span>
                <i>
                    Hint: Set the URL in the API connection settings.
                </i>
            </div>
            <div class="flex-container flexFlowColumn" id="openai_vectorsModel">
                <label for="vectors_openai_model">
                    Vectorization Model
                </label>
                <select id="vectors_openai_model" class="text_pole">
                    <option value="text-embedding-ada-002">text-embedding-ada-002</option>
                    <option value="text-embedding-3-small">text-embedding-3-small</option>
                    <option value="text-embedding-3-large">text-embedding-3-large</option>
                </select>
            </div>
            <div class="flex-container flexFlowColumn" id="cohere_vectorsModel">
                <label for="vectors_cohere_model">
                    Vectorization Model
                </label>
                <select id="vectors_cohere_model" class="text_pole">
                    <option value="embed-english-v3.0">embed-english-v3.0</option>
                    <option value="embed-multilingual-v3.0">embed-multilingual-v3.0</option>
                    <option value="embed-english-light-v3.0">embed-english-light-v3.0</option>
                    <option value="embed-multilingual-light-v3.0">embed-multilingual-light-v3.0</option>
                    <option value="embed-english-v2.0">embed-english-v2.0</option>
                    <option value="embed-english-light-v2.0">embed-english-light-v2.0</option>
                    <option value="embed-multilingual-v2.0">embed-multilingual-v2.0</option>
                </select>
            </div>
            <div class="flex-container flexFlowColumn" id="together_vectorsModel">
                <label for="vectors_togetherai_model">
                    Vectorization Model
                </label>
                <select id="vectors_togetherai_model" class="text_pole">
                    <option value="togethercomputer/m2-bert-80M-32k-retrieval">M2-BERT-Retrieval-32k</option>
                    <option value="togethercomputer/m2-bert-80M-8k-retrieval">M2-BERT-Retrieval-8k</option>
                    <option value="togethercomputer/m2-bert-80M-2k-retrieval">M2-BERT-Retrieval-2K</option>
                    <option value="WhereIsAI/UAE-Large-V1">UAE-Large-V1</option>
                    <option value="BAAI/bge-large-en-v1.5">BAAI-Bge-Large-1p5</option>
                    <option value="BAAI/bge-base-en-v1.5">BAAI-Bge-Base-1p5</option>
                    <option value="sentence-transformers/msmarco-bert-base-dot-v5">Sentence-BERT</option>
                    <option value="bert-base-uncased">Bert Base Uncased</option>
                </select>
            </div>
            <div class="flex-container flexFlowColumn" id="vllm_vectorsModel">
                <label for="vectors_vllm_model">
                    Vectorization Model
                </label>
                <input id="vectors_vllm_model" class="text_pole" type="text" placeholder="Model name, e.g. intfloat/e5-mistral-7b-instruct" />
                <i>
                    Hint: Set the URL in the API connection settings.
                </i>
            </div>

            <small id="vectors_modelWarning">
                <i class="fa-solid fa-exclamation-triangle"></i>
                <span data-i18n="Vectors Model Warning">
                    It is recommended to purge vectors when changing the model mid-chat. Otherwise, it will lead to sub-par results.
                </span>
            </small>

            <div class="flex-container flexFlowColumn" id="nomicai_apiKey">
                <label for="api_key_nomicai">
                    <span>NomicAI API Key</span>
                </label>
                <div class="flex-container">
                    <input id="api_key_nomicai" name="api_key_nomicai" class="text_pole flex1 wide100p" maxlength="500" size="35" type="text" autocomplete="off">
                    <div title="Clear your API key" class="menu_button fa-solid fa-circle-xmark clear-api-key" data-key="api_key_nomicai">
                    </div>
                </div>
                <div data-for="api_key_nomicai" class="neutral_warning" data-i18n="For privacy reasons, your API key will be hidden after you reload the page.">
                    For privacy reasons, your API key will be hidden after you reload the page.
                </div>
            </div>

            <div class="flex-container marginTopBot5">
                <div class="flex-container flex1 flexFlowColumn" title="How many last messages will be matched for relevance.">
                    <label for="vectors_query">
                        <span>Query messages</span>
                    </label>
                    <input type="number" id="vectors_query" class="text_pole widthUnset" min="1" max="99" />
                </div>
                <div class="flex-container flex1 flexFlowColumn" title="Cut-off score for relevance. Helps to filter out irrelevant data.">
                    <label for="vectors_query">
                        <span>Score threshold</span>
                    </label>
                    <input type="number" id="vectors_score_threshold" class="text_pole widthUnset" min="0" max="1" step="0.05" />
                </div>
            </div>

            <div class="flex-container">
                <label class="checkbox_label expander" for="vectors_include_wi" title="Query results can activate World Info entries.">
                    <input id="vectors_include_wi" type="checkbox" class="checkbox">
                    Include in World Info Scanning
                </label>
            </div>

            <hr>

            <h4>
                World Info settings
            </h4>

            <label class="checkbox_label" for="vectors_enabled_world_info" title="Enable activation of World Info entries based on vector similarity.">
                <input id="vectors_enabled_world_info" type="checkbox" class="checkbox">
                Enabled for World Info
            </label>

            <div id="vectors_world_info_settings" class="marginTopBot5">
                <div class="flex-container">
                    <label for="vectors_enabled_for_all" class="checkbox_label">
                        <input id="vectors_enabled_for_all" type="checkbox" />
                        <span>Enabled for all entries</span>
                    </label>
                    <ul class="margin0">
                        <li>
                            <small>Checked: all entries except ❌ status can be activated.</small>
                        </li>
                        <li>
                            <small>Unchecked: only entries with 🔗 status can be activated.</small>
                        </li>
                    </ul>
                </div>
                <div class="flex-container">
                    <div class="flex1">
                        <!--  Vacant for future use -->
                    </div>
                    <div class="flex1" title="Maximum number of entries to be activated">
                        <label for="vectors_max_entries" >
                            <small>Max Entries</small>
                        </label>
                        <input id="vectors_max_entries" type="number" class="text_pole widthUnset" min="1" max="9999" />
                    </div>
                    <div class="flex1">
                        <!--  Vacant for future use -->
                    </div>
                </div>
            </div>

            <h4>
                File vectorization settings
            </h4>

            <label class="checkbox_label" for="vectors_enabled_files">
                <input id="vectors_enabled_files" type="checkbox" class="checkbox">
                Enabled for files
            </label>

            <div id="vectors_files_settings" class="marginTopBot5">
                <label class="checkbox_label" for="vectors_translate_files" title="This can help with retrieval accuracy if using embedding models that are trained on English data. Uses the selected API from Chat Translation extension settings.">
                    <input id="vectors_translate_files" type="checkbox" class="checkbox">
                    <span data-i18n="Translate files into English before processing">
                        Translate files into English before processing
                    </span>
                    <i class="fa-solid fa-flask" title="Experimental feature"></i>
                </label>
                <div class="flex justifyCenter" title="These settings apply to files attached directly to messages.">
                    <span>Message attachments</span>
                </div>
                <div class="flex-container">
                    <div class="flex1" title="Only files past this size will be vectorized.">
                        <label for="vectors_size_threshold">
                            <small>Size threshold (KB)</small>
                        </label>
                        <input id="vectors_size_threshold" type="number" class="text_pole" min="1" max="99999" />
                    </div>
                    <div class="flex1" title="Chunk size for file splitting.">
                        <label for="vectors_chunk_size">
                            <small>Chunk size (chars)</small>
                        </label>
                        <input id="vectors_chunk_size" type="number" class="text_pole" min="1" max="99999" />
                    </div>
                    <div class="flex1" title="The overlap between adjacent chunks in % from chunk size. The overlap text is trimmed to sentence boundaries. 0 = disabled.">
                        <label for="vectors_overlap_percent">
                            <small>Chunk overlap (%)</small>
                        </label>
                        <input id="vectors_overlap_percent" type="number" class="text_pole" min="0" max="99" step="1" />
                    </div>
                    <div class="flex1" title="How many chunks to retrieve when querying.">
                        <label for="vectors_chunk_count">
                            <small>Retrieve chunks</small>
                        </label>
                        <input id="vectors_chunk_count" type="number" class="text_pole" min="1" max="99999" />
                    </div>
                </div>
                <div class="flex justifyCenter" title="These settings apply to files stored in the Data Bank.">
                    <span>Data Bank files</span>
                </div>
                <div class="flex-container">
                    <div class="flex1" title="Only files past this size will be vectorized.">
                        <label for="vectors_size_threshold_db">
                            <small>Size threshold (KB)</small>
                        </label>
                        <input id="vectors_size_threshold_db" type="number" class="text_pole" min="1" max="99999" />
                    </div>
                    <div class="flex1" title="Chunk size for file splitting.">
                        <label for="vectors_chunk_size_db">
                            <small>Chunk size (chars)</small>
                        </label>
                        <input id="vectors_chunk_size_db" type="number" class="text_pole" min="1" max="99999" />
                    </div>
                    <div class="flex1" title="The overlap between adjacent chunks in % from chunk size. The overlap text is trimmed to sentence boundaries. 0 = disabled.">
                        <label for="vectors_overlap_percent_db">
                            <small>Chunk overlap (%)</small>
                        </label>
                        <input id="vectors_overlap_percent_db" type="number" class="text_pole" min="0" max="99" step="1" />
                    </div>
                    <div class="flex1" title="How many chunks to retrieve when querying.">
                        <label for="vectors_chunk_count_db">
                            <small>Retrieve chunks</small>
                        </label>
                        <input id="vectors_chunk_count_db" type="number" class="text_pole" min="1" max="99999" />
                    </div>
                </div>
                <div class="flex-container flexFlowColumn">
                    <label for="vectors_file_template_db">
                        <span>Injection Template</span>
                    </label>
                    <textarea id="vectors_file_template_db" class="margin0 text_pole textarea_compact" rows="3" placeholder="Use &lcub;&lcub;text&rcub;&rcub; macro to specify the position of retrieved text."></textarea>
                    <label for="vectors_file_position_db">Injection Position</label>
                    <div class="radio_group">
                        <label>
                            <input type="radio" name="vectors_file_position_db" value="2" />
                            <span>Before Main Prompt / Story String</span>
                        </label>
                        <!--Keep these as 0 and 1 to interface with the setExtensionPrompt function-->
                        <label>
                            <input type="radio" name="vectors_file_position_db" value="0" />
                            <span>After Main Prompt / Story String</span>
                        </label>
                        <label for="vectors_file_depth_db" title="How many messages before the current end of the chat." data-i18n="[title]How many messages before the current end of the chat.">
                            <input type="radio" name="vectors_file_position_db" value="1" />
                            <span>In-chat @ Depth</span>
                            <input id="vectors_file_depth_db" class="text_pole widthUnset" type="number" min="0" max="999" />
                            <span>as</span>
                            <select id="vectors_file_depth_role_db"  class="text_pole widthNatural">
                                <option value="0">System</option>
                                <option value="1">User</option>
                                <option value="2">Assistant</option>
                            </select>
                        </label>
                    </div>
                </div>
                <div class="flex-container">
                    <div id="vectors_files_vectorize_all" class="menu_button menu_button_icon" title="Vectorize all files in the Data Bank and current chat.">
                        Vectorize All
                    </div>
                    <div id="vectors_files_purge" class="menu_button menu_button_icon" title="Purge all file vectors in the Data Bank and current chat.">
                        Purge Vectors
                    </div>
                </div>
            </div>

            <hr>

            <h4>
                Chat vectorization settings
            </h4>
            <label class="checkbox_label" for="vectors_enabled_chats">
                <input id="vectors_enabled_chats" type="checkbox" class="checkbox">
                Enabled for chat messages
            </label>

            <hr>

            <div id="vectors_chats_settings">
                <div id="vectors_advanced_settings">
                    <label for="vectors_template">
                        Injection Template
                    </label>
                    <textarea id="vectors_template" class="text_pole textarea_compact" rows="3" placeholder="Use &lcub;&lcub;text&rcub;&rcub; macro to specify the position of retrieved text."></textarea>
                    <label for="vectors_position">Injection Position</label>
                    <div class="radio_group">
                        <label>
                            <input type="radio" name="vectors_position" value="2" />
                            Before Main Prompt / Story String
                        </label>
                        <!--Keep these as 0 and 1 to interface with the setExtensionPrompt function-->
                        <label>
                            <input type="radio" name="vectors_position" value="0" />
                            After Main Prompt / Story String
                        </label>
                        <label for="vectors_depth" title="How many messages before the current end of the chat." data-i18n="[title]How many messages before the current end of the chat.">
                            <input type="radio" name="vectors_position" value="1" />
                            In-chat @ Depth <input id="vectors_depth" class="text_pole widthUnset" type="number" min="0" max="999" />
                        </label>
                    </div>
                    <div class="flex-container">
                        <div class="flex1" title="Can increase the retrieval quality for the cost of processing. 0 = disabled.">
                            <label for="vectors_message_chunk_size">
                                <small>Chunk size (chars)</small>
                            </label>
                            <input id="vectors_message_chunk_size" type="number" class="text_pole widthUnset" min="0" max="9999" />
                        </div>
                        <div class="flex1" title="Prevents last N messages from being placed out of order.">
                            <label for="vectors_protect">
                                <small>Retain#</small>
                            </label>
                            <input type="number" id="vectors_protect" class="text_pole widthUnset" min="1" max="9999" />
                        </div>
                        <div class="flex1" title="How many past messages to insert as memories.">
                            <label for="vectors_insert">
                                <small>Insert#</small>
                            </label>
                            <input type="number" id="vectors_insert" class="text_pole widthUnset" min="1" max="9999" />
                        </div>
                    </div>
                    <hr class="m-b-1">
                    <div class="flex-container flexFlowColumn">
                        <div class="flex-container alignitemscenter justifyCenter">
                            <i class="fa-solid fa-flask" title="Summarization for vectors is an experimental feature that may improve vectors or may worsen them. Use at your own discretion."></i>
                            <span>Vector Summarization</span>
                        </div>
                        <label class="checkbox_label expander" for="vectors_summarize" title="Summarize chat messages before generating embeddings.">
                            <input id="vectors_summarize" type="checkbox" class="checkbox">
                            Summarize chat messages for vector generation
                        </label>
                        <i class="failure">Warning: This will slow down vector generation drastically, as all messages have to be summarized first.</i>

                        <label class="checkbox_label expander" for="vectors_summarize_user" title="Summarize sent chat messages before generating embeddings.">
                            <input id="vectors_summarize_user" type="checkbox" class="checkbox">
                            Summarize chat messages when sending
                        </label>
                        <i class="failure">Warning: This might cause your sent messages to take a bit to process and slow down response time.</i>

                        <label for="vectors_summary_source">Summarize with:</label>
                        <select id="vectors_summary_source" class="text_pole">
                            <option value="main">Main API</option>
                            <option value="extras">Extras API</option>
                        </select>

                        <label for="vectors_summary_prompt">Summary Prompt:</label>
                        <small>Only used when Main API is selected.</small>
                        <textarea id="vectors_summary_prompt" class="text_pole textarea_compact" rows="6" placeholder="This prompt will be sent to AI to request the summary generation."></textarea>
                    </div>
                </div>
                <small>
                    Old messages are vectorized gradually as you chat.
                    To process all previous messages, click the button below.
                </small>
                <div class="flex-container">
                    <div id="vectors_vectorize_all" class="menu_button menu_button_icon">
                        Vectorize All
                    </div>
                    <div id="vectors_purge" class="menu_button menu_button_icon">
                        Purge Vectors
                    </div>
                    <div id="vectors_view_stats" class="menu_button menu_button_icon">
                        View Stats
                    </div>
                </div>
                <div id="vectorize_progress" style="display: none;">
                    <small>
                        Processed <span id="vectorize_progress_percent">0</span>% of messages.
                        ETA: <span id="vectorize_progress_eta">...</span> seconds.
                    </small>
                </div>
            </div>
        </div>
    </div>
</div>