From 306cf51da499f52dd638cf684818b77eaee49968 Mon Sep 17 00:00:00 2001
From: Cohee <18619528+Cohee1207@users.noreply.github.com>
Date: Wed, 27 Sep 2023 22:09:09 +0300
Subject: [PATCH] #1180 Add custom token bans. Return grammar strings
---
public/index.html | 36 ++++++++++++++++-----
public/scripts/textgen-settings.js | 51 +++++++++++++++++++++++++++---
2 files changed, 75 insertions(+), 12 deletions(-)
diff --git a/public/index.html b/public/index.html
index 1b7c0f0c8..04134a3e2 100644
--- a/public/index.html
+++ b/public/index.html
@@ -1148,10 +1148,10 @@
Banned Tokens
- Sequences you don't want to appear in the output. One per line.
+ Sequences you don't want to appear in the output. One per line. Text or [token ids].
-
+
@@ -1493,6 +1493,26 @@
+
+
+ Banned Tokens
+ (LLaMA models)
+
+
+ Sequences you don't want to appear in the output. One per line. Text or [token ids].
+
+
+
+
+
+
+
+
+ Most tokens have a leading space.
+
+
+
CFG Scale
@@ -1513,12 +1533,12 @@
Negative Prompt
-
+
+
+ Used if CFG Scale is unset globally, per chat or character
+
-
- Used if CFG Scale is unset globally, per chat or character
-
Beam search
@@ -1622,10 +1642,10 @@
-
+
Grammar
-
+
Type in the desired custom grammar (GBNF).
diff --git a/public/scripts/textgen-settings.js b/public/scripts/textgen-settings.js
index 1ea81f5e7..98fbcd704 100644
--- a/public/scripts/textgen-settings.js
+++ b/public/scripts/textgen-settings.js
@@ -9,6 +9,8 @@ import {
import {
power_user,
} from "./power-user.js";
+import { getTextTokens, tokenizers } from "./tokenizers.js";
+import { onlyUnique } from "./utils.js";
export {
textgenerationwebui_settings,
@@ -50,7 +52,8 @@ const textgenerationwebui_settings = {
mirostat_eta: 0.1,
guidance_scale: 1,
negative_prompt: '',
- grammar_file: '',
+ grammar_string: '',
+ banned_tokens: '',
};
export let textgenerationwebui_presets = [];
@@ -86,7 +89,8 @@ const setting_names = [
"mirostat_eta",
"guidance_scale",
"negative_prompt",
- //"grammar_file",
+ "grammar_string",
+ "banned_tokens",
];
function selectPreset(name) {
@@ -126,6 +130,44 @@ function convertPresets(presets) {
return Array.isArray(presets) ? presets.map(JSON.parse) : [];
}
+/**
+ * @returns {string} String with comma-separated banned token IDs
+ */
+function getCustomTokenBans() {
+ if (!textgenerationwebui_settings.banned_tokens) {
+ return '';
+ }
+
+ const sequences = textgenerationwebui_settings.banned_tokens.split('\n');
+ const result = [];
+
+ for (const line of sequences) {
+ // Raw token ids, JSON serialized
+ if (line.startsWith('[') && line.endsWith(']')) {
+ try {
+ const tokens = JSON.parse(line);
+
+ if (Array.isArray(tokens) && tokens.every(t => Number.isInteger(t))) {
+ result.push(...tokens);
+ } else {
+ throw new Error('Not an array of integers');
+ }
+ } catch (err) {
+ console.log(`Failed to parse bad word token list: ${line}`, err);
+ }
+ } else {
+ try {
+ const tokens = getTextTokens(tokenizers.LLAMA, line);
+ result.push(...tokens);
+ } catch {
+ console.log(`Could not tokenize raw text: ${line}`);
+ }
+ }
+ }
+
+ return result.filter(onlyUnique).map(x => String(x)).join(',');
+}
+
function loadTextGenSettings(data, settings) {
textgenerationwebui_presets = convertPresets(data.textgenerationwebui_presets);
textgenerationwebui_preset_names = data.textgenerationwebui_preset_names ?? [];
@@ -149,7 +191,7 @@ function loadTextGenSettings(data, settings) {
}
$(document).ready(function () {
- $('#settings_preset_textgenerationwebui').on('change', function() {
+ $('#settings_preset_textgenerationwebui').on('change', function () {
const presetName = $(this).val();
selectPreset(presetName);
});
@@ -268,6 +310,7 @@ export function getTextGenGenerationData(finalPrompt, this_amount_gen, isImperso
'mirostat_mode': textgenerationwebui_settings.mirostat_mode,
'mirostat_tau': textgenerationwebui_settings.mirostat_tau,
'mirostat_eta': textgenerationwebui_settings.mirostat_eta,
- //'grammar_file': textgenerationwebui_settings.grammar_file,
+ 'grammar_string': textgenerationwebui_settings.grammar_string,
+ 'custom_token_bans': getCustomTokenBans(),
};
}