diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp index bbfd5c792..20f25dc26 100644 --- a/gpttype_adapter.cpp +++ b/gpttype_adapter.cpp @@ -3059,6 +3059,7 @@ static void PrepareMediaEmbds(const int nctx, const std::vector & media_int } else { + media_composite_image_signature = ""; //force invalidate printf("\nWarning: Vision Image excluded - Context size too low or not enough clip tokens! (needed %d)\nImage will be IGNORED! You probably want to relaunch with a larger context size!\n",cliptokensneeded); } media_objects[i].mediachunks.push_back(chunk); @@ -3112,6 +3113,7 @@ static void PrepareMediaEmbds(const int nctx, const std::vector & media_int } else { + media_composite_image_signature = ""; //force invalidate printf("\nWarning: Audio Embd excluded - Context size too low or not enough clip tokens! (needed %d)\nAudio will be IGNORED! You probably want to relaunch with a larger context size!\n",cliptokensneeded); } @@ -3315,7 +3317,7 @@ generation_outputs gpttype_generate(const generation_inputs inputs) media_composite_image_signature = new_media_composite; if(debugmode==1 && !is_quiet) { - printf("\nLLAVA images changed, existing cache invalidated"); + printf("\nAttached media changed, existing multimodal cache invalidated"); } media_data_changed = true; } @@ -3520,7 +3522,7 @@ generation_outputs gpttype_generate(const generation_inputs inputs) { if(last_media_mem.size() + kcpp_data->n_predict + 4 > nctx) { - printf("\nWarning: Too many LLaVA tokens, max context exceeded! They will be ignored!\n"); + printf("\nWarning: Too many multimodal tokens, max context exceeded! They will be ignored!\n"); } else { @@ -4266,7 +4268,7 @@ generation_outputs gpttype_generate(const generation_inputs inputs) { PrepareMediaEmbds(nctx, media_intro); media_embds_built = true; - printf("\nSomehow vision embd was not prepared (maybe no fast forward), rebuilding it...\n"); + printf("\nSomehow media embeds was not prepared (maybe no fast forward), rebuilding it...\n"); } //if partial batch, dispatch existing first @@ -4301,11 +4303,11 @@ generation_outputs gpttype_generate(const generation_inputs inputs) auto evr = llama_decode(llama_ctx_v4, batch.batch); if(evr!=0) { - printf("\nError when appending llava intro: %d\n",evr); + printf("\nError when appending media intro: %d\n",evr); } else { - printf("\rProcessing LLaVa Intro (%d tokens)",introsize); + printf("\rProcessing Media Intro (%d tokens)",introsize); } n_past += introsize; llavatokensevaled += introsize; @@ -4340,7 +4342,7 @@ generation_outputs gpttype_generate(const generation_inputs inputs) if(!err) { media_composite_image_signature = ""; //force invalidate - fprintf(stderr, "\nFailed to eval llava image at %d!\n",n_past); + fprintf(stderr, "\nFailed to eval media tokens at %d!\n",n_past); output.text = nullptr; output.status = 0; output.prompt_tokens = output.completion_tokens = 0; @@ -4370,7 +4372,7 @@ generation_outputs gpttype_generate(const generation_inputs inputs) if(llavatokenscounted!=llavatokensevaled) { media_composite_image_signature = ""; //force invalidate - fprintf(stderr, "\nLLAVA image tokens mismatch at %d! (%d vs %d tokens)\n",n_past,llavatokenscounted,llavatokensevaled); + fprintf(stderr, "\nMedia tokens mismatch at %d! (%d vs %d tokens)\n",n_past,llavatokenscounted,llavatokensevaled); output.text = nullptr; output.status = 0; output.prompt_tokens = output.completion_tokens = 0; diff --git a/klite.embd b/klite.embd index dfc542f56..2beefb51f 100644 --- a/klite.embd +++ b/klite.embd @@ -14734,11 +14734,11 @@ Current version indicated by LITEVER below. function self_upload_file_dispatch(data,filename) { - const maxSize = 20 * 1024 * 1024; // approx 20MB limit + const maxSize = 30 * 1024 * 1024; // approx 30MB limit const dlen = (data.length*0.75); const mbs = Math.ceil(dlen/1024/1024); if (dlen > maxSize) { - msgbox(`Selected file exceeds 20MB size limit!\nSelected file was ${mbs}MB. Please try a smaller file.`, "File Too Large"); + msgbox(`Selected file exceeds 30MB size limit!\nSelected file was ${mbs}MB. Please try a smaller file.`, "File Too Large"); return; } diff --git a/koboldcpp.py b/koboldcpp.py index b41e10655..0df5daa23 100644 --- a/koboldcpp.py +++ b/koboldcpp.py @@ -5124,7 +5124,7 @@ def show_gui(): ctk.CTkButton(model_tab, width=70, text = "HF Search", command = model_searcher ).grid(row=1,column=0, stick="nw", padx=370, pady=2) makefileentry(model_tab, "Text Lora:", "Select Lora File",lora_var, 3,width=160,singlerow=True,tooltiptxt="Select an optional GGML Text LoRA adapter to use.\nLeave blank to skip.") makelabelentry(model_tab, "Multiplier: ", loramult_var, 3, 50,padx=390,singleline=True,tooltip="Scale multiplier for Text LoRA Strength. Default is 1.0", labelpadx=330) - makefileentry(model_tab, "Vision mmproj:", "Select Vision mmproj File", mmproj_var, 7,width=280,singlerow=True,tooltiptxt="Select a mmproj file to use for vision models like LLaVA.\nLeave blank to skip.") + makefileentry(model_tab, "Mmproj File:", "Select Audio or Vision mmproj File", mmproj_var, 7,width=280,singlerow=True,tooltiptxt="Select a mmproj file to use for multimodal models for vision and audio recognition.\nLeave blank to skip.") makecheckbox(model_tab, "Vision Force CPU", mmprojcpu_var, 9, tooltiptxt="Force CLIP for Vision mmproj always on CPU.") makelabelentry(model_tab, "Vision MaxRes:", visionmaxres_var, 9, padx=320, singleline=True, tooltip=f"Clamp MMProj vision maximum allowed resolution. Allowed values are between 512 to 2048 px (default {default_visionmaxres}).", labelpadx=220) makefileentry(model_tab, "Draft Model:", "Select Speculative Text Model File", draftmodel_var, 11,width=280,singlerow=True,tooltiptxt="Select a draft text model file to use for speculative decoding.\nLeave blank to skip.")