diff --git a/embd_res/klite.embd b/embd_res/klite.embd index f5f971249..fa441654a 100644 --- a/embd_res/klite.embd +++ b/embd_res/klite.embd @@ -8976,7 +8976,7 @@ Current version indicated by LITEVER below. } }); } - + function kai_scenario_load(storyobj) //retain all current settings, but overwrite story and a few key items { temp_scenario = { @@ -9009,9 +9009,9 @@ Current version indicated by LITEVER below. temp_scenario.opmode = storyobj.savedsettings.opmode; } if(temp_scenario.opmode==1) //story - { + { temp_scenario.gui_type = storyobj.savedsettings.gui_type_story?storyobj.savedsettings.gui_type_story:0; - } + } else if(temp_scenario.opmode==2) //adventure { temp_scenario.adventure_context_mod = storyobj.savedsettings.adventure_context_mod?true:false; @@ -9019,11 +9019,11 @@ Current version indicated by LITEVER below. temp_scenario.gui_type = storyobj.savedsettings.gui_type_adventure?storyobj.savedsettings.gui_type_adventure:0; } else if(temp_scenario.opmode==3) //chat - { - temp_scenario.gui_type = storyobj.savedsettings.gui_type_chat?storyobj.savedsettings.gui_type_chat:0; + { + temp_scenario.gui_type = storyobj.savedsettings.gui_type_chat?storyobj.savedsettings.gui_type_chat:0; } else if(temp_scenario.opmode==4) //instruct - { + { temp_scenario.gui_type = storyobj.savedsettings.gui_type_instruct?storyobj.savedsettings.gui_type_instruct:0; } @@ -10429,7 +10429,7 @@ Current version indicated by LITEVER below. scenarios += ``; scenarios += ``; scenarios += ``; - + for(let i=0;i${scenario_sources[i].name}` } @@ -11712,9 +11712,9 @@ Current version indicated by LITEVER below. } } function fetch_openrouter_balance() - { + { let desired_oai_key = document.getElementById("custom_oai_key").value.trim(); - + let oaiheaders = {}; if(desired_oai_key!="" && desired_oai_key!=dummy_api_key){ oaiheaders["Authorization"] = "Bearer " + desired_oai_key; @@ -12821,7 +12821,7 @@ Current version indicated by LITEVER below. let slotname = (testslot?`Server Slot `+(lbl)+` - `+tsdesc+``:`Server Slot `+(lbl)+` - [ Empty ]`); choices += ``; } - } + } else if(selectedlocation=="3") //local scenario { locationdesc.innerText = "Local Scenario Library stores your story as a Custom Scenario, without any generation settings.\nData is saved to a temporary cache and can be deleted by your browser. To avoid losing data, use the download save button."; @@ -13090,10 +13090,10 @@ Current version indicated by LITEVER below. } else { - import_compressed_story(loadedstorycompressed,false); + import_compressed_story(loadedstorycompressed,false); last_used_saveslot = slot; } - + if(switch_to_corpo) { localsettings.gui_type_chat = 3; @@ -13301,7 +13301,7 @@ Current version indicated by LITEVER below. let selectedslot = document.getElementById("saveslotselecteddropdown").value; let selectedlocation = document.getElementById("saveslotlocationdropdown").value; let islocal = (selectedlocation=="1"); - let isscenario = (selectedlocation=="3"); + let isscenario = (selectedlocation=="3"); download_from_slot(selectedslot,islocal,isscenario); } function delete_from_curr_slot() @@ -14308,7 +14308,7 @@ Current version indicated by LITEVER below. { document.body.classList.add(`theme-${i}`); } - } + } } function update_genimg_button_visiblility() @@ -17612,7 +17612,7 @@ Current version indicated by LITEVER below. pending_context_preinjection += endmatcher; } } - + if (force_new_turn || ends_with_assistant_tag) { if (localsettings.inject_timestamps) { pending_context_preinjection += get_current_timestamp(); @@ -18112,7 +18112,7 @@ Current version indicated by LITEVER below. { seqs.push(me + "\:"); } - if(localsettings.chatopponent!="") + if(localsettings.chatopponent!="") { //NOTE: we do not add our opponents name as a stop sequence IF thinking is forced //the model just gets too confused and is likely to repeat the tag. @@ -22500,14 +22500,14 @@ Current version indicated by LITEVER below. } function classic_delete_turn(idx) - { + { edit_chunk_save(idx,"",true); } function classic_edit_turn(idx) { let chatunits = []; //todo: eventually refactor let currctx = concat_gametext(false, "", "", "", false); - currctx = replace_instruct_placeholders(currctx); + currctx = replace_instruct_placeholders(currctx); if(localsettings.opmode==3) { chatunits = repack_chat_history(currctx); @@ -22527,20 +22527,20 @@ Current version indicated by LITEVER below. let currmsg = chatunits[idx].msg; currmsg = stash_image_placeholders(currmsg, false); - + inputBoxOkCancel("Quickly edit a previous turn message","Quick Edit Turn",currmsg,"(Delete Turn)",()=>{ let userinput = getInputBoxValue(); edit_chunk_save(idx,userinput,true); },()=>{ //do nothing on cancel - },true,true); + },true,true); } function classic_click_turn(button,idx,toggleopen = true) { let incomplete_resp = (synchro_pending_stream != "" || pending_response_id != ""); if (incomplete_resp) { return; } - + function handleOutsideClickClassicTurn(event) { popupMenu.style.display = 'none'; document.removeEventListener('click', handleOutsideClickClassicTurn); @@ -22548,7 +22548,7 @@ Current version indicated by LITEVER below. const rect = button.getBoundingClientRect(); const popupMenu = document.getElementById('classicturnmenu'); - + let popcontent = ``; popcontent += ``; popcontent += ``; @@ -22843,7 +22843,7 @@ Current version indicated by LITEVER below. let chatunits = []; let currctx = concat_gametext(false, "", "", "", false); - currctx = replace_instruct_placeholders(currctx); + currctx = replace_instruct_placeholders(currctx); if(localsettings.opmode==3) { chatunits = repack_chat_history(currctx); @@ -22960,7 +22960,7 @@ Current version indicated by LITEVER below. let idx = modified_turn; let chatunits = []; let currctx = concat_gametext(false, "", "", "", false); - currctx = replace_instruct_placeholders(currctx); + currctx = replace_instruct_placeholders(currctx); if(localsettings.opmode==3) { chatunits = repack_chat_history(currctx); @@ -22977,7 +22977,7 @@ Current version indicated by LITEVER below. { chatunits = repack_instruct_history(currctx, allow_blank); } - + let needsave = false; let existing_msg_compare = stash_image_placeholders(chatunits[idx].msg,false); @@ -23077,7 +23077,7 @@ Current version indicated by LITEVER below. let textarea = document.getElementById("corpo_edit_inp"); let modified_message = textarea.value; edit_chunk_save(corpo_editing_turn, modified_message); - + } function corpo_edit_chunk_resend(idx) { @@ -25052,7 +25052,7 @@ Current version indicated by LITEVER below. //refactor - repack as turns let chatunits = []; - input = replace_instruct_placeholders(input); + input = replace_instruct_placeholders(input); if(localsettings.opmode==3) //chat mode { chatunits = repack_chat_history(input); @@ -27848,8 +27848,8 @@ Current version indicated by LITEVER below. - - + +
Slot Storage Option diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu index 33e2ced5b..6080ac74d 100644 --- a/ggml/src/ggml-cuda/ggml-cuda.cu +++ b/ggml/src/ggml-cuda/ggml-cuda.cu @@ -208,7 +208,7 @@ static ggml_cuda_device_info ggml_cuda_init() { //#else // GGML_LOG_INFO("%s: GGML_CUDA_FORCE_CUBLAS: no\n", __func__); //#endif // GGML_CUDA_FORCE_CUBLAS - + GGML_LOG_INFO("%s: found %d " GGML_CUDA_NAME " devices:\n", __func__, info.device_count); std::vector> turing_devices_without_mma; diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp index caf50297e..845fa3515 100644 --- a/gpttype_adapter.cpp +++ b/gpttype_adapter.cpp @@ -1525,7 +1525,7 @@ void sample_entropy(llama_token_data_array * cur_p, float min_temp, float max_te // Only apply smoothing if smoothing_factor is > 0. Do not change base implementation otherwise. if (smoothing_factor > 0 && cur_p->size > 1) { sample_softmax(cur_p); - float h = cur_p->data[0].logit; // Find the maximum logit for h to be added after the transformation + float h = cur_p->data[0].logit; // Find the maximum logit for h to be added after the transformation // Apply the modified quadratic transformation using the smoothing_factor and smoothing_curve for (size_t i = 0; i < cur_p->size; ++i) { float logit_shifted = cur_p->data[i].logit - h; diff --git a/koboldcpp.py b/koboldcpp.py index 102212fe9..0d84312f7 100644 --- a/koboldcpp.py +++ b/koboldcpp.py @@ -2365,7 +2365,7 @@ def is_ipv6_supported(): except Exception: return False -def format_jinja(messages, tools): +def format_jinja(messages, tools): try: def strftime_now(format='%Y-%m-%d %H:%M:%S'): return datetime.now().strftime(format) @@ -2389,7 +2389,7 @@ def format_jinja(messages, tools): def remove_outer_tags(inputstr): try: - stripped = inputstr.strip() + stripped = inputstr.strip() match = re.match(r'^<([^\s<>]+)>(.*?)\s*$', stripped, re.DOTALL) # Try angle brackets first if match: return match.group(2).strip() @@ -2399,10 +2399,10 @@ def remove_outer_tags(inputstr): return stripped # If no match, return original string except Exception: return stripped - -def normalize_tool_call(obj): # Normalize various tool call formats to OpenAI format + +def normalize_tool_call(obj): # Normalize various tool call formats to OpenAI format if "type" in obj and "function" in obj: # Already in OpenAI format - return obj + return obj if "name" in obj and ("arguments" in obj or "parameters" in obj): args = obj.get("arguments", obj.get("parameters", {})) return { @@ -2422,13 +2422,13 @@ def normalize_tool_call(obj): # Normalize various tool call formats to OpenAI fo "arguments": func.get("arguments", func.get("parameters", {})) } } - + return obj # Used to parse json for openai tool calls def extract_json_from_string(input_string): parsed_json = None - input_string = remove_outer_tags(input_string) #if we detected wrapper tags, remove them + input_string = remove_outer_tags(input_string) #if we detected wrapper tags, remove them try: # First check if model exported perfect json parsed_json = json.loads(input_string) @@ -2740,8 +2740,8 @@ ws ::= | " " | "\n" [ \t]{0,20} attachedaudid = 0 jinja_output = None jinjatools = genparams.get('tools', []) - if use_jinja and cached_chat_template: - jinja_output = format_jinja(messages_array,jinjatools) + if use_jinja and cached_chat_template: + jinja_output = format_jinja(messages_array,jinjatools) if jinja_output: messages_string = jinja_output if jinjatools and len(jinjatools)>0: @@ -2825,7 +2825,7 @@ ws ::= | " " | "\n" [ \t]{0,20} elif message['role'] == "tool": messages_string += tools_message_end messages_string += assistant_message_gen - + genparams["prompt"] = messages_string if len(images_added)>0: genparams["images"] = images_added @@ -3499,7 +3499,7 @@ Change Mode
response_body = embedded_kailite else: response_body = (f"Embedded KoboldAI Lite is not found.
You will have to connect via the main KoboldAI client, or use this URL to connect.").encode() - + elif self.path in ["/noscript", "/noscript?"] or self.path.startswith(('/noscript?','noscript?')): #it's possible for the root url to have ?params without / self.noscript_webui() @@ -3702,7 +3702,7 @@ Change Mode
elif self.path.endswith(('/.well-known/serviceinfo')): response_body = (json.dumps({"version":"0.2","software":{"name":"KoboldCpp","version":KcppVersion,"repository":"https://github.com/LostRuins/koboldcpp","homepage":"https://github.com/LostRuins/koboldcpp","logo":"https://raw.githubusercontent.com/LostRuins/koboldcpp/refs/heads/concedo/niko.ico"},"api":{"koboldai":{"name":"KoboldAI API","rel_url":"/api","documentation":"https://lite.koboldai.net/koboldcpp_api","version":KcppVersion},"openai":{"name":"OpenAI API","rel_url ":"/v1","documentation":"https://openai.com/documentation/api","version":KcppVersion}}}).encode()) - elif self.path=="/props": + elif self.path=="/props": response_body = (json.dumps({ "chat_template": cached_chat_template, "id": 0, @@ -3734,16 +3734,16 @@ Change Mode
response_body = embedded_kcpp_docs else: response_body = ("KoboldCpp API is running!\n\nAPI usage reference can be found at the wiki: https://github.com/LostRuins/koboldcpp/wiki").encode() - + elif self.path=="/lcpp": content_type = 'text/html' # IMPORTANT: svelte needs a patch to accept this as a non-redirect path. Search for `r.pathname === e + "/index.html"` and add desired path there. if supports_gzip and embedded_lcpp_ui_gz is not None: response_body = embedded_lcpp_ui_gz - content_encoding = 'gzip' + content_encoding = 'gzip' else: response_body = ("Llama.cpp UI is not available. Please use the KoboldAI Lite UI instead.").encode() - + elif self.path.startswith(("/sdui")): content_type = 'text/html' if supports_gzip and embedded_kcpp_sdui_gz is not None: @@ -3752,7 +3752,7 @@ Change Mode
elif embedded_kcpp_sdui is not None: response_body = embedded_kcpp_sdui else: - response_body = ("KoboldCpp API is running, but KCPP SDUI is not loaded").encode() + response_body = ("KoboldCpp API is running, but KCPP SDUI is not loaded").encode() elif self.path=="/v1": content_type = 'text/html' @@ -4389,14 +4389,14 @@ Change Mode
try: toolsdata_res = gendat['choices'][0]['message']['tool_calls'] if toolsdata_res and len(toolsdata_res)>0: - toolsdata_res[0]["index"] = 0 # need to add an index for OWUI + toolsdata_res[0]["index"] = 0 # need to add an index for OWUI except Exception: toolsdata_res = [] try: content_text = gendat['choices'][0]['message'].get('content', None) except Exception: content_text = None - + # Send role chunk first chunk_role = json.dumps({ "id": "koboldcpp", @@ -5722,13 +5722,13 @@ def show_gui(): quantkv_var.trace_add("write", toggleflashattn) makecheckbox(tokens_tab, "No BOS Token", nobostoken_var, 43, tooltiptxt="Prevents BOS token from being added at the start of any prompt. Usually NOT recommended for most models.") makecheckbox(tokens_tab, "Enable Guidance", enableguidance_var, 43,padx=(200 if corrupt_scaler else 140), tooltiptxt="Enables the use of Classifier-Free-Guidance, which allows the use of negative prompts. Has performance and memory impact.") - def togglejinja(a,b,c): + def togglejinja(a,b,c): if jinja_var.get()==1: jinjatoolsbox.grid() else: jinja_tools_var.set(0) jinjatoolsbox.grid_remove() - changed_gpulayers_estimate() + changed_gpulayers_estimate() makecheckbox(tokens_tab, "Use Jinja", jinja_var, row=45, command=togglejinja, tooltiptxt="Enables using jinja chat template formatting for chat completions endpoint. Other endpoints are unaffected.") jinjatoolsbox = makecheckbox(tokens_tab, "Jinja for Tools", jinja_tools_var, row=45 ,padx=(200 if corrupt_scaler else 140), tooltiptxt="Allows jinja even with tool calls. If unchecked, jinja will be disabled when tools are used.") jinja_var.trace_add("write", togglejinja) @@ -5736,7 +5736,7 @@ def show_gui(): makelabelentry(tokens_tab, "MoE CPU Layers:", moecpu_var, row=55, padx=(490 if corrupt_scaler else 320), singleline=True, tooltip="Force Mixture of Experts (MoE) weights of the first N layers to the CPU.\nSetting it higher than GPU layers has no effect.", labelpadx=(300 if corrupt_scaler else 210)) makelabelentry(tokens_tab, "Override KV:", override_kv_var, row=57, padx=(220 if corrupt_scaler else 120), singleline=True, width=150, tooltip="Override metadata value by key. Separate multiple values with commas. Format is name=type:value. Types: int, float, bool, str") makelabelentry(tokens_tab, "Override Tensors:", override_tensors_var, row=59, padx=(220 if corrupt_scaler else 120), singleline=True, width=150, tooltip="Override selected backend for specific tensors matching tensor_name_regex_pattern=buffer_type, same as in llama.cpp.") - + # Model Tab model_tab = tabcontent["Loaded Files"] @@ -7397,7 +7397,7 @@ def kcpp_main_process(launch_args, g_memory=None, gui_launcher=False): if isinstance(args.chatcompletionsadapter, str) and os.path.exists(args.chatcompletionsadapter): ccadapter_path = os.path.abspath(args.chatcompletionsadapter) elif isinstance(args.chatcompletionsadapter, str) and adapt_dir: - filename = args.chatcompletionsadapter + filename = args.chatcompletionsadapter if not filename.endswith(".json"): filename += ".json" #strip to just the filename @@ -7717,7 +7717,7 @@ def kcpp_main_process(launch_args, g_memory=None, gui_launcher=False): if not loadok: exitcounter = 999 exit_with_error(3,"Could not load text model: " + modelname) - + # The chat completions adapter is a list that needs derivation from chat templates # Try to derive chat completions adapter from chat template, now that we have the model loaded if not args.nomodel and args.model_param: diff --git a/model_adapter.cpp b/model_adapter.cpp index 1c565f4b4..61382cbaf 100644 --- a/model_adapter.cpp +++ b/model_adapter.cpp @@ -268,7 +268,7 @@ std::string gguf_get_model_arch(const std::string & gguf_filename) } } } - } + } else if (vocabsiz>=51864 && vocabsiz<=51865) { printf("\nWhisper model detected - you should load it as a whisper model instead, not a text model!\n");