Merge pull request #3112 from kallewoof/202411-backend-maxctx feature: allow auto-use of max context size given by backend

39c3924b3faa9ed92b2dd35f67c133e54b81db2d

Cohee <18619528+Cohee1207@users.noreply.github.com>

Signed
4 files changed, +50 -16Ignore whitespace
public/index.html+10 -4
@@ -2566,15 +2566,21 @@
25662566 <input id="koboldcpp_api_url_text" class="text_pole wide100p" value="" autocomplete="off" data-server-history="koboldcpp">
25672567 </div>
25682568 </div>
2569+ <div class="flex-container flexFlowColumn marginTopBot5">
2570+ <label data-tg-type="ooba" class="checkbox_label" for="bypass_status_check_textgenerationwebui">
2571+ <input type="checkbox" id="bypass_status_check_textgenerationwebui" />
2572+ <span data-i18n="Bypass status check">Bypass status check</span>
2573+ </label>
2574+ <label data-tg-type="koboldcpp, llamacpp" class="checkbox_label" for="context_size_derived">
2575+ <input type="checkbox" id="context_size_derived" />
2576+ <span data-i18n="Derive context size from backend">Derive context size from backend</span>
2577+ </label>
2578+ </div>
25692579 <div class="flex-container">
25702580 <div id="api_button_textgenerationwebui" class="api_button menu_button menu_button_icon" type="submit" data-i18n="Connect" data-server-connect="ooba_blocking,vllm,aphrodite,tabby,koboldcpp,ollama,llamacpp,huggingface">Connect</div>
25712581 <div data-tg-type="openrouter" class="menu_button menu_button_icon openrouter_authorize" title="Get your OpenRouter API token using OAuth flow. You will be redirected to openrouter.ai" data-i18n="Authorize;[title]Get your OpenRouter API token using OAuth flow. You will be redirected to openrouter.ai">Authorize</div>
25722582 <div class="api_loading menu_button menu_button_icon" data-i18n="Cancel">Cancel</div>
25732583 </div>
2574- <label data-tg-type="ooba" class="checkbox_label margin-bot-10px" for="bypass_status_check_textgenerationwebui">
2575- <input type="checkbox" id="bypass_status_check_textgenerationwebui" />
2576- <span data-i18n="Bypass status check">Bypass status check</span>
2577- </label>
25782584 </form>
25792585 <div class="online_status">
25802586 <div class="online_status_indicator"></div>
public/script.js+17 -1
@@ -1238,8 +1238,9 @@ async function getStatusTextgen() {
12381238
12391239 const wantsInstructDerivation = (power_user.instruct.enabled && power_user.instruct.derived);
12401240 const wantsContextDerivation = power_user.context_derived;
1241+ const wantsContextSize = power_user.context_size_derived;
12411242 const supportsChatTemplate = [textgen_types.KOBOLDCPP, textgen_types.LLAMACPP].includes(textgen_settings.type);
12421243 if (supportsChatTemplate && (wantsInstructDerivation || wantsContextDerivation || wantsContextSize)) {
12431244 const response = await fetch('/api/backends/text-completions/props', {
12441245 method: 'POST',
12451246 headers: getRequestHeaders(),
@@ -1253,6 +1254,17 @@ async function getStatusTextgen() {
12531254 const data = await response.json();
12541255 if (data) {
12551256 const { chat_template, chat_template_hash } = data;
1257+ if (wantsContextSize && 'default_generation_settings' in data) {
1258+ const backend_max_context = data['default_generation_settings']['n_ctx'];
1259+ const old_value = max_context;
1260+ if (max_context !== backend_max_context) {
1261+ setGenerationParamsFromPreset({ max_length: backend_max_context });
1262+ }
1263+ if (old_value !== max_context) {
1264+ console.log(`Auto-switched max context from ${old_value} to ${max_context}`);
1265+ toastr.info(`${old_value} ⇒ ${max_context}`, 'Context Size Changed');
1266+ }
1267+ }
12561268 console.log(`We have chat template ${chat_template.split('\n')[0]}...`);
12571269 const templates = await deriveTemplatesFromChatTemplate(chat_template, chat_template_hash);
12581270 if (templates) {
@@ -6822,6 +6834,10 @@ export async function saveSettings(type) {
68226834 });
68236835}
68246836
6837+/**
6838+ * Sets the generation parameters from a preset object.
6839+ * @param {{ genamt?: number, max_length?: number }} preset Preset object
6840+ */
68256841export function setGenerationParamsFromPreset(preset) {
68266842 const needsUnlock = (preset.max_length ?? max_context) > MAX_CONTEXT_DEFAULT || (preset.genamt ?? amount_gen) > MAX_RESPONSE_DEFAULT;
68276843 $('#max_context_unlocked').prop('checked', needsUnlock).trigger('change');
public/scripts/power-user.js+12 -0
@@ -246,6 +246,7 @@ let power_user = {
246246 },
247247
248248 context_derived: false,
249+ context_size_derived: false,
249250
250251 sysprompt: {
251252 enabled: true,
@@ -1482,6 +1483,7 @@ async function loadPowerUserSettings(settings, data) {
14821483 $('#example_messages_behavior').val(getExampleMessagesBehavior());
14831484 $(`#example_messages_behavior option[value="${getExampleMessagesBehavior()}"]`).prop('selected', true);
14841485 $('#context_derived').parent().find('i').toggleClass('toggleEnabled', !!power_user.context_derived);
1486+ $('#context_size_derived').prop('checked', !!power_user.context_size_derived);
14851487
14861488 $('#console_log_prompts').prop('checked', power_user.console_log_prompts);
14871489 $('#request_token_probabilities').prop('checked', power_user.request_token_probabilities);
@@ -3079,6 +3081,16 @@ $(document).ready(() => {
30793081 $('#context_derived').parent().find('i').toggleClass('toggleEnabled', !!power_user.context_derived);
30803082 });
30813083
3084+ $('#context_size_derived').on('input', function () {
3085+ const value = !!$(this).prop('checked');
3086+ power_user.context_size_derived = value;
3087+ saveSettingsDebounced();
3088+ });
3089+
3090+ $('#context_size_derived').on('change', function () {
3091+ $('#context_size_derived').prop('checked', !!power_user.context_size_derived);
3092+ });
3093+
30823094 $('#always-force-name2-checkbox').change(function () {
30833095 power_user.always_force_name2 = !!$(this).prop('checked');
30843096 saveSettingsDebounced();
src/endpoints/backends/text-completions.js+11 -11
@@ -152,7 +152,7 @@ router.post('/status', jsonParser, async function (request, response) {
152152
153153 if (!modelsReply.ok) {
154154 console.log('Models endpoint is offline.');
155155 return response.statussendStatus(400);
156156 }
157157
158158 /** @type {any} */
@@ -173,7 +173,7 @@ router.post('/status', jsonParser, async function (request, response) {
173173
174174 if (!Array.isArray(data.data)) {
175175 console.log('Models response is not an array.');
176176 return response.statussendStatus(400);
177177 }
178178
179179 const modelIds = data.data.map(x => x.id);
@@ -224,7 +224,7 @@ router.post('/status', jsonParser, async function (request, response) {
224224 return response.send({ result, data: data.data });
225225 } catch (error) {
226226 console.error(error);
227227 return response.statussendStatus(500);
228228 }
229229});
230230
@@ -244,7 +244,7 @@ router.post('/props', jsonParser, async function (request, response) {
244244 const propsReply = await fetch(propsUrl, args);
245245
246246 if (!propsReply.ok) {
247247 return response.statussendStatus(400);
248248 }
249249
250250 /** @type {any} */
@@ -258,7 +258,7 @@ router.post('/props', jsonParser, async function (request, response) {
258258 return response.send(props);
259259 } catch (error) {
260260 console.error(error);
261261 return response.statussendStatus(500);
262262 }
263263});
264264
@@ -450,7 +450,7 @@ ollama.post('/download', jsonParser, async function (request, response) {
450450 return response.send({ ok: true });
451451 } catch (error) {
452452 console.error(error);
453453 return response.statussendStatus(500);
454454 }
455455});
456456
@@ -493,7 +493,7 @@ ollama.post('/caption-image', jsonParser, async function (request, response) {
493493 return response.send({ caption });
494494 } catch (error) {
495495 console.error(error);
496496 return response.statussendStatus(500);
497497 }
498498});
499499
@@ -540,7 +540,7 @@ llamacpp.post('/caption-image', jsonParser, async function (request, response) {
540540
541541 } catch (error) {
542542 console.error(error);
543543 return response.statussendStatus(500);
544544 }
545545});
546546
@@ -569,7 +569,7 @@ llamacpp.post('/props', jsonParser, async function (request, response) {
569569
570570 } catch (error) {
571571 console.error(error);
572572 return response.statussendStatus(500);
573573 }
574574});
575575
@@ -619,7 +619,7 @@ llamacpp.post('/slots', jsonParser, async function (request, response) {
619619
620620 } catch (error) {
621621 console.error(error);
622622 return response.statussendStatus(500);
623623 }
624624});
625625
@@ -665,7 +665,7 @@ tabby.post('/download', jsonParser, async function (request, response) {
665665 return response.send({ ok: true });
666666 } catch (error) {
667667 console.error(error);
668668 return response.statussendStatus(500);
669669 }
670670});
671671