add full text context for chat feature

do-me · Mar 4, 2024 · d97cf5d · d97cf5d
1 parent 34bb5e8
commit d97cf5d
Show file tree

Hide file tree

Showing 2 changed files with 18 additions and 13 deletions.
diff --git a/index.html b/index.html
@@ -707,26 +707,23 @@ <h4>Dimensionality Reduction (New🔥) </h4>
                     <div>
                         <hr />
                         <h4>Chat (Retrieval Augmented Generation, RAG)</h4>
-                        <p>Enter a question to be answered. The model is automatically prompted with the top search
-                            results in the form: <br><b>"Based on the following input, answer the question: [your
-                                question] [top search results]"</b>.<br> If you encounter errors, the input is probably
+                        <p>Enter a question to be answered and use the placeholders <b>SEARCH_RESULTS</b> or <b>FULL_TEXT</b> for context.<br> If you encounter errors, the input is probably
                             too long (either too many or too
                             long results or too long prompt). Also, make
-                            sure to check the right prompting style! Xenova/LaMini-Flan-T5-783M is by far the best
+                            sure to check the right prompting style! Xenova/Qwen1.5-1.8B-Chat is by far the best
                             quantized
-                            model currently available and delivers good results while the others produce nonsense in
-                            most cases. At some point <a href="https://github.com/xenova/transformers.js/pull/379"
+                            model currently available and delivers good results. At some point <a href="https://github.com/xenova/transformers.js/pull/379"
                                 target="_blank">Falcon & Mistral/Zephyr models</a> will probably become available
                             here.<br><b>Attention</b>: Loads very large models with
                             more than 1.5Gb (!) of resources.</p>
                         <div class="form-floating input-group mb-2">
                             <input id="chat_query" class="form-control"
-                                value="What do these paragraphs have in common?">
+                                value="Based on the following context, answer the question: What do these paragraphs have in common? Context: SEARCH_RESULTS">
                             <label for="query-text">Chat Query</label>
                         </div>
 
                         <div class="d-flex flex-row">
-                            <button id="get_chat" class="btn btn-md btn-primary mb-2 nav-button" disabled>💬 Chat</button>
+                            <button id="get_chat" class="btn btn-md btn-primary mb-2 nav-button">💬 Chat</button>
                             <div class="row">
                                 <div class="col-md-8">
                                     <div class="form-floating mb-2">

diff --git a/src/js/index.js b/src/js/index.js
@@ -506,12 +506,20 @@ async function summarizeTopResults() {
 }
 
 async function chatTopResults() {
-    document.getElementById("chat_text").innerHTML = "";
-    var chatQuery = "Based on the following input, answer the question:" + document.getElementById("chat_query").value;
-    var max_new_tokens = document.getElementById("chat_max_new_tokens").value;
+    document.getElementById("chat_text").innerHTML = ""; // progress bar
 
-    var topResultsString = chatQuery + " Context:\nParagraph: " + Array.from(document.querySelectorAll('#results-list .card-title')).map(title => title.textContent).join('\nParagraph: ');
-    const currentChat = await chatText(topResultsString, max_new_tokens);
+    const chatQuery = document.getElementById("chat_query").value;
+    const search_results = "\nParagraph: " + Array.from(document.querySelectorAll('#results-list .card-title')).map(title => title.textContent).join('\nParagraph: ');
+    const full_text = editor.getValue()
+
+    const finalQuery = chatQuery.replace("SEARCH_RESULTS", `"""${search_results}"""`).replace("FULL_TEXT",`"""${full_text}"""`) // add variables in string
+    console.log(finalQuery)
+    if (finalQuery.length > 10000){
+        alert("Attention: Context might be too large (> 10.000 chars) and require too much RAM to be processed. Try working with fewer results or shorter chunks.")
+    }
+
+    var max_new_tokens = document.getElementById("chat_max_new_tokens").value;
+    const currentChat = await chatText(finalQuery, max_new_tokens);
 }
 
 function resetMetadata() {