Add context sizes for new groq models

15769a7643d3c4e063d60749bd2654ba9dc6bbab

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

Parents: 9b631ed0489c80aeaf11b675b6fb2b151e796f60
1 files changed, +35 -27Ignore whitespace
public/scripts/openai.js+35 -27
@@ -4115,6 +4115,39 @@ function getMaxContextWindowAI(value) {
4115 }4115 }
4116}4116}
41174117
4118/**
4119 * Get the maximum context size for the Groq model
4120 * @param {string} model Model identifier
4121 * @param {boolean} isUnlocked Whether context limits are unlocked
4122 * @returns {number} Maximum context size in tokens
4123 */
4124function getGroqMaxContext(model, isUnlocked) {
4125 if (isUnlocked) {
4126 return unlocked_max;
4127 }
4128
4129 const contextMap = {
4130 'gemma2-9b-it': max_8k,
4131 'llama-3.3-70b-versatile': max_128k,
4132 'llama-3.1-8b-instant': max_128k,
4133 'llama3-70b-8192': max_8k,
4134 'llama3-8b-8192': max_8k,
4135 'mixtral-8x7b-32768': max_32k,
4136 'deepseek-r1-distill-llama-70b': max_128k,
4137 'llama-3.3-70b-specdec': max_8k,
4138 'llama-3.2-1b-preview': max_128k,
4139 'llama-3.2-3b-preview': max_128k,
4140 'llama-3.2-11b-vision-preview': max_128k,
4141 'llama-3.2-90b-vision-preview': max_128k,
4142 'qwen-2.5-32b': max_128k,
4143 'deepseek-r1-distill-qwen-32b': max_128k,
4144 'deepseek-r1-distill-llama-70b-specdec': max_128k,
4145 };
4146
4147 // Return context size if model found, otherwise default to 128k
4148 return Object.entries(contextMap).find(([key]) => model.includes(key))?.[1] || max_128k;
4149}
4150
4118async function onModelChange() {4151async function onModelChange() {
4119 biasCache = undefined;4152 biasCache = undefined;
4120 let value = String($(this).val() || '');4153 let value = String($(this).val() || '');
@@ -4416,33 +4449,8 @@ async function onModelChange() {
4416 }4449 }
44174450
4418 if (oai_settings.chat_completion_source == chat_completion_sources.GROQ) {4451 if (oai_settings.chat_completion_source == chat_completion_sources.GROQ) {
4419 if (oai_settings.max_context_unlocked) {4452 const maxContext = getGroqMaxContext(oai_settings.groq_model, oai_settings.max_context_unlocked);
4420 $('#openai_max_context').attr('max', unlocked_max);4453 $('#openai_max_context').attr('max', maxContext);
4421 } else if (oai_settings.groq_model.includes('gemma2-9b-it')) {
4422 $('#openai_max_context').attr('max', max_8k);
4423 } else if (oai_settings.groq_model.includes('llama-3.3-70b-versatile')) {
4424 $('#openai_max_context').attr('max', max_128k);
4425 } else if (oai_settings.groq_model.includes('llama-3.1-8b-instant')) {
4426 $('#openai_max_context').attr('max', max_128k);
4427 } else if (oai_settings.groq_model.includes('llama3-70b-8192')) {
4428 $('#openai_max_context').attr('max', max_8k);
4429 } else if (oai_settings.groq_model.includes('llama3-8b-8192')) {
4430 $('#openai_max_context').attr('max', max_8k);
4431 } else if (oai_settings.groq_model.includes('mixtral-8x7b-32768')) {
4432 $('#openai_max_context').attr('max', max_32k);
4433 } else if (oai_settings.groq_model.includes('deepseek-r1-distill-llama-70b')) {
4434 $('#openai_max_context').attr('max', max_128k);
4435 } else if (oai_settings.groq_model.includes('llama-3.3-70b-specdec')) {
4436 $('#openai_max_context').attr('max', max_8k);
4437 } else if (oai_settings.groq_model.includes('llama-3.2-1b-preview')) {
4438 $('#openai_max_context').attr('max', max_128k);
4439 } else if (oai_settings.groq_model.includes('llama-3.2-3b-preview')) {
4440 $('#openai_max_context').attr('max', max_128k);
4441 } else if (oai_settings.groq_model.includes('llama-3.2-11b-vision-preview')) {
4442 $('#openai_max_context').attr('max', max_128k);
4443 } else if (oai_settings.groq_model.includes('llama-3.2-90b-vision-preview')) {
4444 $('#openai_max_context').attr('max', max_128k);
4445 }
4446 oai_settings.openai_max_context = Math.min(Number($('#openai_max_context').attr('max')), oai_settings.openai_max_context);4454 oai_settings.openai_max_context = Math.min(Number($('#openai_max_context').attr('max')), oai_settings.openai_max_context);
4447 $('#openai_max_context').val(oai_settings.openai_max_context).trigger('input');4455 $('#openai_max_context').val(oai_settings.openai_max_context).trigger('input');
4448 oai_settings.temp_openai = Math.min(oai_max_temp, oai_settings.temp_openai);4456 oai_settings.temp_openai = Math.min(oai_max_temp, oai_settings.temp_openai);