vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-05-08 18:30:50 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

updated lite

Browse source
This commit is contained in:
Concedo 2025-10-23 11:52:59 +08:00
parent 7446e03851
commit be3328887f
1 changed files with 124 additions and 31 deletions
Download patch file Download diff file Expand all files Collapse all files

155
embd_res/klite.embd
View file

@ -12,7 +12,7 @@ Current version indicated by LITEVER below.
-->
<script id="init-config">
const LITEVER = 289;
const LITEVER = 290;
const urlParams = new URLSearchParams(window.location.search);
var localflag = urlParams.get('local'); //this will be replaced automatically in embedded kcpp
const STORAGE_PREFIX = (localflag?"e_":"")+"kaihordewebui_";
@ -3424,11 +3424,12 @@ Current version indicated by LITEVER below.
const claude_submit_endpoint = "/complete";
const claude_submit_endpoint_v3 = "/messages";
const claude_models_endpoint = "/v1/models";
const claude_models_endpoint = "/models";
const default_openrouter_base = "https://openrouter.ai/api/v1";
const default_mistralai_base = "https://api.mistral.ai/v1";
const default_featherless_base = "https://api.featherless.ai/v1";
const default_nvidianim_base = "https://integrate.api.nvidia.com/v1";
const default_grok_base = "https://api.x.ai/v1";
const default_oai_base = "https://api.openai.com/v1";
const default_claude_base = "https://api.anthropic.com";
@ -3644,6 +3645,7 @@ Current version indicated by LITEVER below.
saved_openrouter_key: "",
saved_mistralai_key: "",
saved_featherless_key: "",
saved_nvidianim_key: "",
saved_pollinations_key: "", //this is a dummy key and isnt needed
saved_grok_key:"",
saved_claude_key: "", //do not ever share this in save files!
@ -4574,6 +4576,8 @@ Current version indicated by LITEVER below.
polyfills();
document.getElementById("lastreq1lp").innerHTML = document.getElementById("lastreq2lp").innerHTML = document.getElementById("lastreq3lp").innerHTML = "";
document.getElementById("lastreq1").innerHTML =
document.getElementById("lastreq2").innerHTML =
document.getElementById("lastreq3").innerHTML =
@ -6264,6 +6268,7 @@ Current version indicated by LITEVER below.
{
synchro_pending_stream = "";
streaming_was_thinking = false;
let logprobs_content_arr = [];
let reqOpt =
{method: 'POST',
headers: submit_headers,
@ -6376,6 +6381,12 @@ Current version indicated by LITEVER below.
synchro_pending_stream += event.data.choices[0].delta.reasoning;
}
//collect logprobs if detected
if(event.data.choices[0].logprobs && event.data.choices[0].logprobs.content && event.data.choices[0].logprobs.content.length>0)
{
logprobs_content_arr = logprobs_content_arr.concat(event.data.choices[0].logprobs.content);
}
//mistral prefill interception
if(oaiemulatecompletionscontent!="" && synchro_pending_stream!="" && synchro_pending_stream.startsWith(oaiemulatecompletionscontent))
{
@ -6436,6 +6447,14 @@ Current version indicated by LITEVER below.
}
synchro_pending_stream = "";
streaming_was_thinking = false;
if (logprobs_content_arr.length>0 && last_response_obj==null) {
//fake a last response obj
let fakedresponse = {
"artificial_response": true,
"results":[{"logprobs":{"content":logprobs_content_arr}}]
};
last_response_obj = fakedresponse;
}
poll_pending_response();
//handle gen failures
if(resp.status==503)
@ -6881,7 +6900,7 @@ Current version indicated by LITEVER below.
connect_custom_endpoint();
document.getElementById("lastreq3").innerHTML = document.getElementById("lastreq2").innerHTML = document.getElementById("lastreq1").innerHTML =
`<a href="#" class="color_grayurl" onclick="msgbox('Source code is available at https://github.com/LostRuins/lite.koboldai.net \\nPlease report any bugs you find there.','Information')">KoboldAI Lite</a> v${LITEVER} Embedded`;
document.getElementById("lastreq1lp").innerHTML = document.getElementById("lastreq2lp").innerHTML = document.getElementById("lastreq3lp").innerHTML = "";
read_url_params_data();
}
else
@ -8700,6 +8719,7 @@ Current version indicated by LITEVER below.
let tmp_or1 = localsettings.saved_openrouter_key;
let tmp_mai = localsettings.saved_mistralai_key;
let tmp_fai = localsettings.saved_featherless_key;
let tmp_nvai = localsettings.saved_nvidianim_key;
let tmp_pol = localsettings.saved_pollinations_key;
let tmp_grok = localsettings.saved_grok_key;
let tmp_claude1 = localsettings.saved_claude_key;
@ -8755,6 +8775,7 @@ Current version indicated by LITEVER below.
localsettings.saved_openrouter_key = tmp_or1;
localsettings.saved_mistralai_key = tmp_mai;
localsettings.saved_featherless_key = tmp_fai;
localsettings.saved_nvidianim_key = tmp_nvai;
localsettings.saved_pollinations_key = tmp_pol;
localsettings.saved_grok_key = tmp_grok;
localsettings.saved_claude_key = tmp_claude1;
@ -11172,6 +11193,8 @@ Current version indicated by LITEVER below.
return document.getElementById("custom_grok_model");
case "10":
return document.getElementById("custom_pollinations_model");
case "11":
return document.getElementById("custom_nvidianim_model");
default:
return document.getElementById("custom_oai_model"); //just return oai
}
@ -11179,7 +11202,7 @@ Current version indicated by LITEVER below.
function ep_should_always_use_chat_completions()
{
let epchoice = document.getElementById("customapidropdown").value;
return (epchoice==7 || epchoice==10);
return (epchoice==7 || epchoice==10 || epchoice==11);
}
function select_custom_oai_model()
@ -11222,23 +11245,27 @@ Current version indicated by LITEVER below.
{
let desired_claude_key = document.getElementById("custom_claude_key").value.trim();
let claude_ep = document.getElementById("custom_claude_endpoint").value.trim();
if(claude_ep.toLowerCase().includes("api.anthropic.com"))
{
//official API has broken cors settings
claude_ep = apply_proxy_url(claude_ep,true);
}
if(desired_claude_key=="")
{
msgbox("Claude requires an API key to fetch model list!");
return;
}
if (document.getElementById("claudeaddversion").checked)
{
if (claude_ep != "" && claude_ep.length > 4 && !claude_ep.slice(-4).toLowerCase().includes("/v")) {
claude_ep = claude_ep + "/v1";
}
}
let dropdown = document.getElementById("custom_claude_model");
fetch((claude_ep + claude_models_endpoint), {
method: 'GET',
headers: {
'Content-Type': 'application/json',
'anthropic-version':'2023-06-01',
'anthropic-dangerous-direct-browser-access':true,
'x-api-key': desired_claude_key,
},
referrerPolicy: 'no-referrer',
@ -11398,10 +11425,12 @@ Current version indicated by LITEVER below.
let dropdown = get_custom_ep_model_dropdown();
let fetchurl = desired_oai_ep + oai_models_endpoint;
let epchoice = document.getElementById("customapidropdown").value;
if(epchoice==2 && !is_local_url(desired_oai_ep) && document.getElementById("corsproxy_oai").checked)
if((epchoice==2 && !is_local_url(desired_oai_ep) && document.getElementById("corsproxy_oai").checked) || fetchurl.toLowerCase().includes("api.nvidia.com"))
{
fetchurl = cors_proxy + "?" + fetchurl; //temp cors proxy if selected
}
fetch(fetchurl, {
method: 'GET',
headers: oaiheaders,
@ -11501,6 +11530,7 @@ Current version indicated by LITEVER below.
document.getElementById("custom_openrouter_model").classList.add("hidden");
document.getElementById("custom_mistralai_model").classList.add("hidden");
document.getElementById("custom_featherless_model").classList.add("hidden");
document.getElementById("custom_nvidianim_model").classList.add("hidden");
document.getElementById("custom_grok_model").classList.add("hidden");
document.getElementById("custom_pollinations_model").classList.add("hidden");
document.getElementById("hordeloadmodelcontainer").classList.add("hidden");
@ -11520,7 +11550,7 @@ Current version indicated by LITEVER below.
document.getElementById("customkoboldkey").value = localsettings.saved_kai_key;
}
}
else if(epchoice==2 || epchoice==3 || epchoice==7 || epchoice==8 || epchoice==9 || epchoice==10)
else if(epchoice==2 || epchoice==3 || epchoice==7 || epchoice==8 || epchoice==9 || epchoice==10 || epchoice==11)
{
document.getElementById("oaicustom").classList.remove("hidden");
document.getElementById("custom_oai_endpoint").classList.add("hidden");
@ -11528,6 +11558,7 @@ Current version indicated by LITEVER below.
document.getElementById("openrouterdesc").classList.add("hidden");
document.getElementById("mistralaidesc").classList.add("hidden");
document.getElementById("featherlessdesc").classList.add("hidden");
document.getElementById("nvidianimdesc").classList.add("hidden");
document.getElementById("grokdesc").classList.add("hidden");
document.getElementById("oaidesc").classList.add("hidden");
document.getElementById("pollinationsdesc").classList.add("hidden");
@ -11586,6 +11617,14 @@ Current version indicated by LITEVER below.
document.getElementById("custom_oai_key").classList.add("hidden");
try_fetch_oai_models_auto();
}
else if(epchoice==11)
{
document.getElementById("custom_nvidianim_model").classList.remove("hidden");
document.getElementById("nvidianimdesc").classList.remove("hidden");
document.getElementById("custom_oai_key").value = (localsettings.saved_nvidianim_key==dummy_api_key?"":localsettings.saved_nvidianim_key);
document.getElementById("custom_oai_endpoint").value = default_nvidianim_base;
try_fetch_oai_models_auto();
}
oai_model_change(ep_should_always_use_chat_completions() || force_autotoggle_chatcompl);
toggleoaichatcompl();
}
@ -12050,7 +12089,7 @@ Current version indicated by LITEVER below.
});
}
}
else if(epchoice==2 || epchoice==3 || epchoice==7 || epchoice==8 || epchoice==9 || epchoice==10) //connect to OAI / OpenRouter / MistralAI / Featherless / Grok Endpoint
else if(epchoice==2 || epchoice==3 || epchoice==7 || epchoice==8 || epchoice==9 || epchoice==10 || epchoice==11) //connect to OAI / OpenRouter / MistralAI / Featherless / Grok Endpoint / NvidiaNim
{
let desired_oai_key = document.getElementById("custom_oai_key").value.trim();
let desired_oai_ep = document.getElementById("custom_oai_endpoint").value.trim();
@ -12089,7 +12128,7 @@ Current version indicated by LITEVER below.
localsettings.saved_oai_key = custom_oai_key;
localsettings.saved_oai_addr = custom_oai_endpoint;
if(!is_local_url(custom_oai_endpoint) && document.getElementById("corsproxy_oai").checked)
if((!is_local_url(custom_oai_endpoint) && document.getElementById("corsproxy_oai").checked) || custom_oai_endpoint.toLowerCase().includes("api.nvidia.com"))
{
uses_cors_proxy = true;
}
@ -12110,6 +12149,11 @@ Current version indicated by LITEVER below.
{
localsettings.saved_pollinations_key = dummy_api_key; //placeholder key, not actually used
}
else if(epchoice==11)
{
localsettings.saved_nvidianim_key = custom_oai_key;
uses_cors_proxy = true;
}
else
{
localsettings.saved_openrouter_key = custom_oai_key;
@ -12183,11 +12227,6 @@ Current version indicated by LITEVER below.
{
dismiss_endpoint_container();
if(desired_claude_ep.toLowerCase().includes("api.anthropic.com"))
{
document.getElementById("connectstatusproxied").classList.remove("hidden");
}
//good to go
custom_claude_endpoint = desired_claude_ep;
custom_claude_key = desired_claude_key;
@ -15588,6 +15627,34 @@ Current version indicated by LITEVER below.
}
else if(is_using_kcpp_with_streaming())
{
//if logprobs used, request last
if(localsettings.request_logprobs && last_response_obj==null)
{
fetch(apply_proxy_url(custom_kobold_endpoint + koboldcpp_logprobs_endpoint), {
method: 'POST',
headers: get_kobold_header(),
body: JSON.stringify({
"genkey": lastcheckgenkey
}),
})
.then((response) => response.json())
.then((data) => {
//makes sure a delayed response doesnt arrive late and mess up
if (data && data.logprobs != null && last_response_obj==null) {
//fake a last response obj
let fakedresponse = {
"artificial_response": true,
"results":[{"logprobs":data.logprobs}]
};
last_response_obj = fakedresponse;
update_logprobs_link();
}
})
.catch((error) => {
console.error('Error:', error);
});
}
//we can use abort functions
fetch(apply_proxy_url(custom_kobold_endpoint + koboldcpp_abort_endpoint), {
method: 'POST', // or 'PUT'
@ -18205,11 +18272,6 @@ Current version indicated by LITEVER below.
|| custom_claude_model.toLowerCase().includes("claude-opus-4");
let actualep = (custom_claude_endpoint + (claudev3mode?claude_submit_endpoint_v3:claude_submit_endpoint));
let targetep = actualep;
if(custom_claude_endpoint.toLowerCase().includes("api.anthropic.com"))
{
//official API has broken cors settings
targetep = apply_proxy_url(actualep,true);
}
let claude_payload = null;
if(claudev3mode)
{
@ -18306,6 +18368,7 @@ Current version indicated by LITEVER below.
'Content-Type': 'application/json',
'x-api-key': custom_claude_key,
'Authorization': 'Bearer '+custom_claude_key,
'anthropic-dangerous-direct-browser-access': true,
};
if(claudev3mode)
{
@ -19856,13 +19919,23 @@ Current version indicated by LITEVER below.
{
shownotify();
}
let kcpp_has_logprobs = (last_response_obj!=null && last_response_obj.results && last_response_obj.results.length > 0 && last_response_obj.results[0].logprobs!=null);
let oai_has_logprobs = (last_response_obj!=null && last_response_obj.choices && last_response_obj.choices.length > 0 && last_response_obj.choices[0].logprobs!=null);
let lastresp = ` <a href="#" class="color_blueurl" onclick="show_last_logprobs()">(View Logprobs)</a>`;
let lastreq = `<a href="#" onclick="show_last_req()">Last request</a> served by <a href="#" onclick="get_and_show_workers()">${genworker}</a> using <span class="color_darkgreen">${genmdl}</span>${(genkudos>0?` for ${genkudos} kudos`:``)} in ${get_time_taken()} seconds.${(last_response_obj!=null && (kcpp_has_logprobs || oai_has_logprobs)?lastresp:"")}`;
let genworkertxt = (genworker=="Custom Endpoint")?`<a href="#" onclick="get_and_show_workers()">${genworker}</a>`:`${genworker}`;
let lastreq = `<a href="#" onclick="show_last_req()">Last request</a> served by ${genworkertxt} using <span class="color_darkgreen">${genmdl}</span>${(genkudos>0?` for ${genkudos} kudos`:``)} in ${get_time_taken()} seconds.`;
document.getElementById("lastreq1").innerHTML = lastreq;
document.getElementById("lastreq2").innerHTML = lastreq;
document.getElementById("lastreq3").innerHTML = lastreq;
document.getElementById("lastreq1lp").innerHTML = document.getElementById("lastreq2lp").innerHTML = document.getElementById("lastreq3lp").innerHTML = "";
update_logprobs_link();
}
function update_logprobs_link()
{
let lastresp = ` <a href="#" class="color_blueurl" onclick="show_last_logprobs()">(View Logprobs)</a>`;
let kcpp_has_logprobs = (last_response_obj!=null && last_response_obj.results && last_response_obj.results.length > 0 && last_response_obj.results[0].logprobs!=null);
let oai_has_logprobs = (last_response_obj!=null && last_response_obj.choices && last_response_obj.choices.length > 0 && last_response_obj.choices[0].logprobs!=null);
let lastresptxt = `${(last_response_obj!=null && (kcpp_has_logprobs || oai_has_logprobs)?lastresp:"")}`;
document.getElementById("lastreq1lp").innerHTML = document.getElementById("lastreq2lp").innerHTML = document.getElementById("lastreq3lp").innerHTML = lastresptxt;
}
function poll_interrogation_db()
@ -23107,6 +23180,9 @@ Current version indicated by LITEVER below.
}else if(custom_oai_endpoint.toLowerCase().includes("text.pollinations.ai"))
{
localsettings.prev_custom_endpoint_type = 10;
}else if(custom_oai_endpoint.toLowerCase().includes("api.nvidia.com"))
{
localsettings.prev_custom_endpoint_type = 11;
}
}
else if(custom_claude_key!="")
@ -25471,7 +25547,7 @@ Current version indicated by LITEVER below.
</div>
</div>
</div>
<div class="lastreq color_gray" id="lastreq1"></div>
<div class="lastreq color_gray"><span id="lastreq1"></span><span id="lastreq1lp"></span></div>
</div>
<div id="enhancedchatinterface" class="chat_mesgs hidden">
@ -25506,7 +25582,7 @@ Current version indicated by LITEVER below.
</div>
</div>
<div class="lastreq color_gray" id="lastreq2"></div>
<div class="lastreq color_gray"><span id="lastreq2"></span><span id="lastreq2lp"></span></div>
</div>
</div>
@ -25543,7 +25619,7 @@ Current version indicated by LITEVER below.
<button title="Submit" onclick="submit_generation_button(true)" id="corpo_chat_send_btn" class="corpo_chat_send_btn mainnav" type="button"></button>
<button title="Abort" onclick="abort_generation()" id="corpo_chat_send_btn_abort" class="hidden corpo_chat_send_btn_abort mainnav" type="button"></button>
</div>
<div id="lastreq3" class="corpolastreq color_gray"></div>
<div class="corpolastreq color_gray"><span id="lastreq3"></span><span id="lastreq3lp"></span></div>
</div>
</div>
</div>
@ -27295,6 +27371,7 @@ Current version indicated by LITEVER below.
<option value="8">Featherless API</option>
<option value="9">Grok API</option>
<option value="10">PollinationsAI API</option>
<option value="11">Nvidia NIM API</option>
</select>
</div>
@ -27385,6 +27462,11 @@ Current version indicated by LITEVER below.
KoboldAI Lite takes no responsibility for your usage of this feature. This service is ad driven, ads may appear in the output.<br>Only Temperature, Top-P, Top-K and Repetition Penalty samplers are used.<br><br>
<span class="color_green" style="font-weight: bold;">No Key Required.</span><br><br>
</span>
<span id="nvidianimdesc" class="hidden">
Entering your Nvidia NIM API key will allow you to use KoboldAI Lite with their API.<br><br>
KoboldAI Lite takes no responsibility for your usage of this feature. Due to CORS restrictions, your connection WILL be proxied.<br>Only Temperature, Top-P and Repetition Penalty samplers are used.<br><br>
<span class="color_green" style="font-weight: bold;">Please input Nvidia NIM Key.</span><br><br>
</span>
<input class="form-control" type="text" id="custom_oai_endpoint" placeholder="OpenAI API URL" value="" onblur="">
<input class="form-control" type="password" id="custom_oai_key" placeholder="API Key (Required)" value="" onfocus="focus_api_keys()" onblur="blur_api_keys()"><br>
@ -27486,6 +27568,17 @@ Current version indicated by LITEVER below.
<option value="bidara">bidara</option>
<option style="display:none;" class="custom_model_option" value="custom">[Custom]</option>
</select>
<select title="Nvidia NIM AI Model Selection" style="padding:4px;display:inline;width:calc(100% - 220px)" class="form-control hidden" id="custom_nvidianim_model" onchange="oai_model_change(true)">
<option value="bytedance/seed-oss-36b-instruct">bytedance/seed-oss-36b-instruct</option>
<option value="deepseek-ai/deepseek-v3.1-terminus">deepseek-ai/deepseek-v3.1-terminus</option>
<option value="google/gemma-3-27b-it">google/gemma-3-27b-it</option>
<option value="meta/llama-3.3-70b-instruct">meta/llama-3.3-70b-instruct</option>
<option value="mistralai/mistral-large">mistralai/mistral-large</option>
<option value="moonshotai/kimi-k2-instruct" selected>moonshotai/kimi-k2-instruct</option>
<option value="nvidia/llama-3.3-nemotron-super-49b-v1">nvidia/llama-3.3-nemotron-super-49b-v1</option>
<option value="qwen/qwen3-235b-a22b">qwen/qwen3-235b-a22b</option>
<option style="display:none;" class="custom_model_option" value="custom">[Custom]</option>
</select>
<button type="button" class="btn btn-primary" style="display:inline;width:105px;" id="oaifetchlist" onclick="oai_fetch_models()">Fetch List</button>
<button type="button" class="btn btn-primary" style="display:inline;width:105px;" id="oaiusecustom" onclick="select_custom_oai_model()">Use Custom</button>
<div class="hidden" id="oaiemulatecompletionsbox">
@ -27548,7 +27641,7 @@ Current version indicated by LITEVER below.
</div>
<div id="claudecustom" class="menutext hidden">
Entering your Claude API key will allow you to use KoboldAI Lite with their API.<br><br>
KoboldAI Lite takes no responsibility for your usage of this feature. <br><span class="color_red">At this time, the official Claude API has CORS restrictions and must be accessed with a CORS proxy. Your connection WILL be proxied.</span><br>Only Temperature, Top-P and Top-K samplers are used.<br><br>
KoboldAI Lite takes no responsibility for your usage of this feature. <br>Only Temperature, Top-P and Top-K samplers are used.<br><br>
<span class="color_green" style="font-weight: bold;">Please input Claude API URL and Key.</span><br><br>
<input class="form-control" type="text" id="custom_claude_endpoint" placeholder="Claude API URL" value="">
<input class="form-control" type="password" id="custom_claude_key" placeholder="Claude API Key (Required)" value="" onfocus="focus_api_keys()" onblur="blur_api_keys()"><br>