added more options to the scale request
This commit is contained in:
parent
06902279a9
commit
ba925f388c
|
@ -710,7 +710,7 @@
|
|||
</div>
|
||||
</div>
|
||||
</div>
|
||||
<div class="range-block" data-source="openai,claude,openrouter,ai21">
|
||||
<div class="range-block" data-source="openai,claude,openrouter,ai21,scale">
|
||||
<div class="range-block-title" data-i18n="Top-p">
|
||||
Top P
|
||||
</div>
|
||||
|
@ -1578,7 +1578,7 @@
|
|||
</div>
|
||||
</div>
|
||||
</div>
|
||||
<div class="range-block m-t-1" data-source="openai,openrouter">
|
||||
<div class="range-block m-t-1" data-source="openai,openrouter,scale">
|
||||
<div class="range-block-title openai_restorable" data-i18n="Logit Bias">
|
||||
Logit Bias
|
||||
</div>
|
||||
|
|
|
@ -1084,23 +1084,34 @@ function saveModelList(data) {
|
|||
}
|
||||
}
|
||||
|
||||
async function sendAltScaleRequest(openai_msgs_tosend, signal) {
|
||||
async function sendAltScaleRequest(openai_msgs_tosend, logit_bias, signal) {
|
||||
const generate_url = '/generate_altscale';
|
||||
|
||||
let firstMsg = substituteParams(openai_msgs_tosend[0].content);
|
||||
let subsequentMsgs = openai_msgs_tosend.slice(1);
|
||||
let firstSysMsgs = []
|
||||
for(let msg of openai_msgs_tosend){
|
||||
if(msg.role === 'system') {
|
||||
firstSysMsgs.push(substituteParams(msg.content));
|
||||
} else {
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
const joinedMsgs = subsequentMsgs.reduce((acc, obj) => {
|
||||
let subsequentMsgs = openai_msgs_tosend.slice(firstSysMsgs.length);
|
||||
|
||||
const joinedSysMsgs = substituteParams(firstSysMsgs.join("\n"));
|
||||
const joinedSubsequentMsgs = subsequentMsgs.reduce((acc, obj) => {
|
||||
return acc + obj.role + ": " + obj.content + "\n";
|
||||
}, "");
|
||||
openai_msgs_tosend = substituteParams(joinedMsgs);
|
||||
console.log(openai_msgs_tosend)
|
||||
|
||||
openai_msgs_tosend = substituteParams(joinedSubsequentMsgs);
|
||||
|
||||
const generate_data = {
|
||||
sysprompt: firstMsg,
|
||||
sysprompt: joinedSysMsgs,
|
||||
prompt: openai_msgs_tosend,
|
||||
temp: parseFloat(oai_settings.temp_openai),
|
||||
top_p: parseFloat(oai_settings.top_p_openai),
|
||||
max_tokens: parseFloat(oai_settings.openai_max_tokens),
|
||||
logit_bias: logit_bias,
|
||||
}
|
||||
|
||||
const response = await fetch(generate_url, {
|
||||
|
@ -1109,6 +1120,7 @@ async function sendAltScaleRequest(openai_msgs_tosend, signal) {
|
|||
headers: getRequestHeaders(),
|
||||
signal: signal
|
||||
});
|
||||
|
||||
const data = await response.json();
|
||||
return data.output;
|
||||
}
|
||||
|
@ -1143,17 +1155,13 @@ async function sendOpenAIRequest(type, openai_msgs_tosend, signal) {
|
|||
openai_msgs_tosend = substituteParams(joinedMsgs);
|
||||
}
|
||||
|
||||
if (isScale && !!$('#scale-alt').prop('checked')) {
|
||||
return sendAltScaleRequest(openai_msgs_tosend, signal)
|
||||
}
|
||||
|
||||
// If we're using the window.ai extension, use that instead
|
||||
// Doesn't support logit bias yet
|
||||
if (oai_settings.chat_completion_source == chat_completion_sources.WINDOWAI) {
|
||||
return sendWindowAIRequest(openai_msgs_tosend, signal, stream);
|
||||
}
|
||||
|
||||
const logitBiasSources = [chat_completion_sources.OPENAI, chat_completion_sources.OPENROUTER];
|
||||
const logitBiasSources = [chat_completion_sources.OPENAI, chat_completion_sources.OPENROUTER, chat_completion_sources.SCALE];
|
||||
if (oai_settings.bias_preset_selected
|
||||
&& logitBiasSources.includes(oai_settings.chat_completion_source)
|
||||
&& Array.isArray(oai_settings.bias_presets[oai_settings.bias_preset_selected])
|
||||
|
@ -1162,6 +1170,10 @@ async function sendOpenAIRequest(type, openai_msgs_tosend, signal) {
|
|||
biasCache = logit_bias;
|
||||
}
|
||||
|
||||
if (isScale && oai_settings.use_alt_scale) {
|
||||
return sendAltScaleRequest(openai_msgs_tosend, logit_bias, signal)
|
||||
}
|
||||
|
||||
const model = getChatCompletionModel();
|
||||
const generate_data = {
|
||||
"messages": openai_msgs_tosend,
|
||||
|
|
|
@ -3329,11 +3329,11 @@ app.post("/generate_altscale", jsonParser, function (request, response_generate_
|
|||
modelType: 'OpenAi',
|
||||
maxTokens: request.body.max_tokens,
|
||||
temperature: request.body.temp,
|
||||
stop: null,
|
||||
stop: "user:",
|
||||
suffix: null,
|
||||
topP: null,
|
||||
topP: request.body.top_p,
|
||||
logprobs: null,
|
||||
logitBias: null
|
||||
logitBias: request.body.logit_bias
|
||||
},
|
||||
inputs: [
|
||||
{
|
||||
|
@ -3348,11 +3348,8 @@ app.post("/generate_altscale", jsonParser, function (request, response_generate_
|
|||
values: {
|
||||
'variant.taxonomy': ['undefined'],
|
||||
'prompt.variablesSourceDataId': ['undefined'],
|
||||
'modelParameters.stop': ['undefined'],
|
||||
'modelParameters.suffix': ['undefined'],
|
||||
'modelParameters.topP': ['undefined'],
|
||||
'modelParameters.logprobs': ['undefined'],
|
||||
'modelParameters.logitBias': ['undefined']
|
||||
}
|
||||
}
|
||||
})
|
||||
|
|
Loading…
Reference in New Issue