vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-10-03 03:25:40 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

bump default ctx to 16k

Browse source
This commit is contained in:
Concedo 2026-09-21 01:32:59 +08:00
parent 635f05b657
commit 34b81e2ada
2 changed files with 8 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

10
embd_res/klite.embd
View file

@ -4609,7 +4609,7 @@ Current version indicated by LITEVER below.
second_ep_model:"gpt2",
second_ep_url:"",
max_context_length: (localflag?12288:6144),
max_context_length: (localflag?16384:6144),
max_length: (localflag?2048:768),
last_maxctx: 0,
auto_ctxlen: true,
@ -15188,12 +15188,12 @@ Current version indicated by LITEVER below.
document.getElementById("max_context_length_slide").max = ep_maxctx;
document.getElementById("max_context_length_slide_label").innerText = ep_maxctx;
}
if(ep_maxctx && ep_maxctx>=12288 && document.getElementById("max_length_slide").max < Math.floor(ep_maxctx/4))
if(ep_maxctx && ep_maxctx>=16384 && document.getElementById("max_length_slide").max < Math.floor(ep_maxctx/4))
{
document.getElementById("max_length_slide").max = Math.floor(ep_maxctx/4);
document.getElementById("max_length_slide_label").innerText = Math.floor(ep_maxctx/4);
}
if(localflag && ep_maxctx>=12288 && localsettings.max_context_length<ep_maxctx && (localsettings.last_maxctx!=ep_maxctx || localsettings.max_context_length==defaultsettings.max_context_length))
if(localflag && ep_maxctx>=16384 && localsettings.max_context_length<ep_maxctx && (localsettings.last_maxctx!=ep_maxctx || localsettings.max_context_length==defaultsettings.max_context_length))
{
localsettings.max_context_length = ep_maxctx;
localsettings.last_maxctx = ep_maxctx;
@ -31942,11 +31942,11 @@ Current version indicated by LITEVER below.
<input title="Context Size" inputmode="numeric" class="justifyright push-right" id="max_context_length" style="width: 8ch;" oninput="
document.getElementById('max_context_length_slide').value = this.value;">
</div>
<div><input title="Context Size Slider" type="range" min="512" max="12288" step="8" id="max_context_length_slide" oninput="
<div><input title="Context Size Slider" type="range" min="512" max="16384" step="8" id="max_context_length_slide" oninput="
document.getElementById('max_context_length').value = this.value;"></div>
<div class="settingminmax">
<div class="justifyleft">512</div>
<div class="justifyright" id="max_context_length_slide_label">12288</div>
<div class="justifyright" id="max_context_length_slide_label">16384</div>
</div>
<div id="auto_ctxlen_panel" class="settinglabel">
<div class="justifyleft" title="Automatically lowers settings if incompatible with existing workers">Auto-Adjust Limits </div>

5
koboldcpp.py
View file

@ -8837,6 +8837,7 @@ def show_gui():
batchsize_text = ["Don't Batch","16","32","64","128","256","512","1024","2048","4096"]
ubatchsize_text = ["Match Batch Size","16","32","64","128","256","512","1024","2048","4096"]
contextsize_text = ["256", "512", "1024", "2048", "3072", "4096", "5120", "6144", "7168", "8192", "9216", "10240", "11264", "12288", "13312", "14336", "15360", "16384", "18432", "20480", "22528", "24576", "26624", "28672", "30720", "32768", "36864", "40960", "45056", "49152", "53248", "57344", "61440", "65536", "73728", "81920", "90112", "98304", "106496", "114688", "122880", "131072", "147456", "163840", "180224", "196608", "212992", "229376", "245760", "262144" ]
default_contextsize_index = contextsize_text.index(str(default_maxctx))
quantkv_text = ["f16","bf16","q8_0","q5_1","q4_0"]
if not any(runopts):
@ -9579,7 +9580,7 @@ def show_gui():
makecheckbox(quick_tab, name, properties[0], int(idx/2) + 20, idx % 2, tooltiptxt=properties[1])
# context size
makeslider(quick_tab, "Context Size:", contextsize_text, context_var, 40, width=280, set=17, tooltip="What is the maximum context size to support. Model specific. You cannot exceed it.\nLarger contexts require more memory, and not all models support it.")
makeslider(quick_tab, "Context Size:", contextsize_text, context_var, 40, width=280, set=default_contextsize_index, tooltip="What is the maximum context size to support. Model specific. You cannot exceed it.\nLarger contexts require more memory, and not all models support it.")
# load model
makefileentry(quick_tab, "GGUF Text Model:", "Select GGUF or GGML Model File", model_var, 50, 280, onchoosefile=on_picked_model_file,tooltiptxt="Select a GGUF or GGML model file on disk to be loaded.")
@ -9671,7 +9672,7 @@ def show_gui():
cacheslots_entry, cacheslots_label = makelabelentry(context_tab, "CacheSlots:", smartcacheslots_var, row=5, padx=(300), singleline=True, tooltip="Number of slots for smartcache",labelpadx=(220))
# context size
makeslider(context_tab, "Context Size:",contextsize_text, context_var, 18, width=280, set=17,tooltip="What is the maximum context size to support. Model specific. You cannot exceed it.\nLarger contexts require more memory, and not all models support it.")
makeslider(context_tab, "Context Size:",contextsize_text, context_var, 18, width=280, set=default_contextsize_index,tooltip="What is the maximum context size to support. Model specific. You cannot exceed it.\nLarger contexts require more memory, and not all models support it.")
context_var.trace_add("write", changed_gpulayers_estimate)
makelabelentry(context_tab, "Default Gen Amt:", defaultgenamt_var, row=20, padx=(120), singleline=True, tooltip="How many tokens to generate by default, if not specified. Must be smaller than context size. Usually, your frontend GUI will override this.")
makelabelentry(context_tab, "Prompt Limit:", genlimit_var, row=20, padx=(300), singleline=True, tooltip="If set, restricts max output tokens to this limit regardless of API request. Set to 0 to disable.",labelpadx=(210))