Add prompt builder + Speech to text

Add prompt builder + Speech to text

Author: svilupp

CHANGELOG.md

@@ -8,15 +8,25 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
+### Fixed
+
+## [0.2.0]
+
+### Added
+
 Chat tab
-- added delete icon to the last message in the conversation (for easy deletion)
-- added a button in "Advanced Settings" to "Fork a conversation" (save to history for reference, but continue from a fresh copy)
-- added a focus when a template is selected (on expand, on template selection, etc)
+- Added delete icon to the last message in the conversation (for easy deletion)
+- Added a button in "Advanced Settings" to "Fork a conversation" (save to history for reference, but continue from a fresh copy)
+- Added a focus when a template is selected (on expand, on template selection, etc)
+- Added a little edit icon for messages (disables the q-popup-edit that was distracting and jumping out too often)
+- Added speech-to-text for the chat (click Record/Stop -> it will paste the text into the chat input and copy it in your clipboard)
 
 Meta-prompting tab
-- Add an experimental meta-prompting experience based on [arxiv](https://arxiv.org/pdf/2401.12954). See the tab "Meta-Prompting" for more details.
+- Added an experimental meta-prompting experience based on [arxiv](https://arxiv.org/pdf/2401.12954). See the tab "Meta-Prompting" for more details.
 
-### Fixed
+Prompt Builder tab
+- Added a tab to generate "prompt templates" that you can then "Apply In Chat" (it jumps to the chat tab and provides template variables to fill in)
+- Allows selecting different models (eg, Opus or more powerful ones) and defining how many samples are provided to give you a choice
 
 ## [0.1.0]
 

Project.toml

@@ -1,9 +1,10 @@
 name = "ProToPortal"
 uuid = "f9496bd6-a3bb-4afc-927d-7268532ebfa9"
 authors = ["J S <[email protected]> and contributors"]
-version = "0.2.0-DEV"
+version = "0.2.0"
 
 [deps]
+Base64 = "2a0f44e3-6c83-55bd-87e4-b1978d98bd5f"
 Dates = "ade2ca70-3891-5945-98fb-dc099432e06a"
 GenieFramework = "a59fdf5c-6bf0-4f5d-949c-a137c9e2f353"
 GenieSession = "03cc5b98-4f21-4eb6-99f2-22eced81f962"

README.md

@@ -85,18 +85,33 @@ For a preview, see the video: `docs/src/videos/screen-capture-code-fixing.mp4`
   - Start "new chat" with conversations automatically saved both on disk and in history.
 
 ### History Tab
+
 - Browse and load past conversations with a simple click.
 - Fork past conversations for continued exploration without altering the original history.
 - View the current session or reload to fetch all saved conversations.
 
 ### Templates Tab
+
 - Explore all available templates with previews and metadata to select the most suitable one.
 - Search functionality for quick filtering by partial names or keywords.
 
 ### Configuration Tab
+
 - Change the default model or add new ones from PromptingTools.
 - Modify the default system prompt used when not employing a template.
 
+### Meta-Prompting Tab
+
+- An experimental meta-prompting experience based on [arxiv](https://arxiv.org/pdf/2401.12954).
+- The model calls different "experts" to solve the provided tasks.
+
+### Prompt Builder Tab
+
+- Generate prompt templates (for use in Chat) from a brief description of a task.
+- Generate multiple templates at once to choose from.
+- Iterate all of them by providing more inputs in the text field.
+- Once you're done, click "Apply in Chat" to jump to the normal chat (use as any other template, eg, fill in variables at the top).
+
 And rich logging in the REPL to see what the GUI is doing under the hood!
 
 ## Screenshots

app.jl

@@ -1,4 +1,5 @@
 module App
+using Base64
 using ProToPortal
 using PromptingTools
 const PT = PromptingTools
@@ -69,9 +70,19 @@ const HISTORY_SAVE = get(ENV, "PROTO_HISTORY_SAVE", true)
     @in chat_auto_reply_count = 0
     # chat
     @in conv_displayed = Dict{Symbol, Any}[]
+    @in chat_edit_show = false
+    @in chat_edit_content = ""
+    @in chat_edit_save = false
+    @in chat_edit_index = 0
+    @in is_recording = false
+    @in audio_chunks = []
+    @in mediaRecorder = nothing
+    @in channel_ = nothing
     # Enter text
     @in chat_question = ""
     @out chat_disabled = false
+    @out chat_question_tokens = ""
+    @out chat_convo_tokens = ""
     # Select template
     @in chat_advanced_expanded = false
     @in chat_template_expanded = false
@@ -92,6 +103,16 @@ const HISTORY_SAVE = get(ENV, "PROTO_HISTORY_SAVE", true)
     @in meta_rounds_current = 0
     @in meta_displayed = Dict{Symbol, Any}[]
     @in meta_rm_last_msg = false
+    ## Prompt Builder
+    @in builder_apply = false
+    @in builder_submit = false
+    @in builder_reset = false
+    @in builder_disabled = false
+    @in builder_question = ""
+    @in builder_tabs = Dict{Symbol, Any}[]
+    @in builder_tab = "tab1"
+    @in builder_model = isempty(PT.GROQ_API_KEY) ? "gpt4t" : "gllama370"
+    @in builder_samples = 3
     # Template browser
     @in template_filter = ""
     @in template_submit = false
@@ -231,6 +252,34 @@ const HISTORY_SAVE = get(ENV, "PROTO_HISTORY_SAVE", true)
         chat_reset = true
         conv_displayed = conv_displayed_temp
     end
+    @onchange conv_displayed begin
+        chat_convo_tokens = if isempty(conv_displayed)
+            ""
+        elseif PT.isaimessage(conv_displayed[end][:message])
+            msg = conv_displayed[end][:message]
+            "Tokens: $(sum(msg.tokens)), Cost: \$$(round(msg.cost;digits=2))"
+        else
+            ""
+        end
+    end
+    ## Chat Speech-to-text
+    @onchange fileuploads begin
+        if !isempty(fileuploads)
+            @info "File was uploaded: " fileuploads["path"]
+            filename = base64encode(fileuploads["name"])
+            try
+                fn_new = fileuploads["path"] * ".wav"
+                mv(fileuploads["path"], fn_new; force = true)
+                chat_question = openai_whisper(fn_new)
+                rm(fn_new; force = true)
+                Base.run(__model__, "this.copyToClipboardText(this.chat_question);")
+            catch e
+                @error "Error processing file: $e"
+                notify(__model__, "Error processing file: $(fileuploads["name"])")
+            end
+            fileuploads = Dict{AbstractString, AbstractString}()
+        end
+    end
     ### Meta-prompting
     @onbutton meta_submit begin
         meta_disabled = true
@@ -272,6 +321,71 @@ const HISTORY_SAVE = get(ENV, "PROTO_HISTORY_SAVE", true)
         pop!(meta_displayed)
         meta_displayed = meta_displayed
     end
+    ### Prompt Builder
+    @onbutton builder_submit begin
+        builder_disabled = true
+        @info "> Prompt Builder Triggered - generating $(builder_samples) samples"
+        first_run = isempty(builder_tabs)
+        for i in 1:builder_samples
+            if first_run
+                ## Generate the first version
+                conv_current = send_to_model(
+                    :PromptGeneratorBasic; task = builder_question, model = builder_model)
+                new_sample = Dict(:name => "tab$(i)",
+                    :label => "Sample $(i)",
+                    :display => [msg2display(msg; id)
+                                 for (id, msg) in enumerate(conv_current)])
+                ## add new sample
+                builder_tabs = push!(builder_tabs, new_sample)
+            else
+                ## Generate the future iterations
+                current_tab = builder_tabs[i]
+                conv = prepare_conversation(
+                    current_tab[:display]; question = builder_question)
+                conv_current = send_to_model(
+                    conv; model = builder_model)
+                ## update the tab
+                current_tab[:display] = [msg2display(msg; id)
+                                         for (id, msg) in enumerate(conv_current)]
+                builder_tabs[i] = current_tab
+                builder_tabs = builder_tabs
+            end
+        end
+
+        builder_disabled, builder_question = false, ""
+    end
+    @onbutton builder_reset begin
+        @info "> Prompt Builder Reset!"
+        builder_tabs = empty!(builder_tabs)
+        builder_disabled, builder_question = false, ""
+    end
+    @onbutton builder_apply begin
+        @info "> Applying Prompt Builder!"
+        builder_msg = filter(x -> x[:name] == builder_tab, builder_tabs) |> only
+        aimsg = builder_msg[:display][end][:message]
+        instructions, inputs = parse_builder(aimsg)
+        if isempty(instructions) && isempty(inputs)
+            notify(__model__, "Parsing failed! Retry...")
+        else
+            conv_current = if isempty(inputs)
+                notify(__model__, "Parsing failed! Expect bad results / edit as needed!")
+                ## slap all instructions into user message
+                [PT.SystemMessage(system_prompt), PT.UserMessage(instructions)]
+            else
+                ## turn into sytem and user message
+                [PT.SystemMessage(instructions), PT.UserMessage(inputs)]
+            end
+            conv_displayed = [msg2display(msg; id)
+                              for (id, msg) in enumerate(conv_current)]
+            ## show the variables to fill in by the user -- use the last message / UserMessage
+            chat_template_expanded = true
+            chat_template_variables = [Dict(:id => id, :variable => String(sym),
+                                           :content => "")
+                                       for (id, sym) in enumerate(conv_current[end].variables)]
+            ## change page to chat
+            selected_page = "chat"
+        end
+    end
     ### Template browsing behavior
     @onbutton template_submit begin
         @info "> Template filter: $template_filter"
@@ -306,10 +420,7 @@ const HISTORY_SAVE = get(ENV, "PROTO_HISTORY_SAVE", true)
         end
     end
 end
-## TODO: add cost tracking on configuration pages + token tracking
-## TODO: add RAG/knowledge loading from folder or URL
-# Required for the JS events
-
+### JAVASCRIPT SECTION ### 
 # set focus to the first variable when it changes
 @watch begin
     raw"""
@@ -330,6 +441,16 @@ end
             console.log("woowza");
         }
     },
+    // saves edits made in the chat dialog
+    saveEdits(index) {
+        this.chat_edit_show = false;
+        this.conv_displayed[this.chat_edit_index].content = this.chat_edit_content;
+        this.chat_edit_content = "";
+    },
+    updateLengthChat() {
+        const tokens = Math.round(this.chat_question.length / 3.5);
+        this.chat_question_tokens = `Approx. tokens: ${tokens}`;
+    },
     focusTemplateSelect() {
         this.$nextTick(() => {
             this.$refs.tpl_select.focus();
@@ -387,6 +508,56 @@ end
         el.select();                                    // Select the <textarea> content
         document.execCommand('copy');                   // Copy - only works as a result of a user action (e.g. click events)
         document.body.removeChild(el);                  // Remove the <textarea> element
+    },
+    copyToClipboardText(str) {
+        const el = document.createElement('textarea');  // Create a <textarea> element
+        el.value = str;                                 // Set its value to the string that you want copied
+        el.setAttribute('readonly', '');                // Make it readonly to be tamper-proof
+        el.style.position = 'absolute';                 
+        el.style.left = '-9999px';                      // Move outside the screen to make it invisible
+        document.body.appendChild(el);                  // Append the <textarea> element to the HTML document
+        el.select();                                    // Select the <textarea> content
+        document.execCommand('copy');                   // Copy - only works as a result of a user action (e.g. click events)
+        document.body.removeChild(el);                  // Remove the <textarea> element
+    },
+    async toggleRecording() {
+        if (!this.is_recording) {
+          this.startRecording()
+        } else {
+          this.stopRecording()
+        }
+    },
+    async startRecording() {
+      navigator.mediaDevices.getUserMedia({ audio: true })
+        .then(stream => {
+          this.is_recording = true
+          this.mediaRecorder = new MediaRecorder(stream);
+          this.mediaRecorder.start();
+          this.mediaRecorder.onstop = () => {
+            const audioBlob = new Blob(this.audio_chunks, { type: 'audio/wav' });
+            this.is_recording = false;
+
+            // upload via uploader
+            const file = new File([audioBlob], 'test.wav');
+            this.$refs.uploader.addFiles([file], 'test.wav');
+            this.$refs.uploader.upload(); // Trigger the upload
+            console.log("Uploaded WAV");
+            this.$refs.uploader.reset();
+            this.audio_chunks=[];
+
+          };
+          this.mediaRecorder.ondataavailable = event => {
+            this.audio_chunks.push(event.data);
+          };
+        })
+        .catch(error => console.error('Error accessing microphone:', error));
+    },
+    stopRecording() {
+      if (this.mediaRecorder) {
+        this.mediaRecorder.stop();
+      } else {
+        console.error('MediaRecorder is not initialized');
+      }
     }
     """
 end

docs/src/index.md

@@ -11,14 +11,11 @@ hero:
     alt: Portal Icon
   actions:
     - theme: brand
-      text: Get Started
-      link: /getting_started
+      text: Introduction
+      link: /introduction
     - theme: alt
-      text: How It Works
-      link: /how_it_works
-    - theme: alt
-      text: F.A.Q.
-      link: /frequently_asked_questions
+      text: Reference
+      link: /reference
     - theme: alt
       text: View on GitHub
       link: https://github.com/svilupp/ProToPortal.jl

docs/src/introduction.md

@@ -92,6 +92,18 @@ For a preview, see the video: `docs/src/videos/screen-capture-code-fixing.mp4`
 - Change the default model or add new ones from PromptingTools.
 - Modify the default system prompt used when not employing a template.
 
+### Meta-Prompting Tab
+
+- An experimental meta-prompting experience based on [arxiv](https://arxiv.org/pdf/2401.12954).
+- The model calls different "experts" to solve the provided tasks.
+
+### Prompt Builder Tab
+
+- Generate prompt templates (for use in Chat) from a brief description of a task.
+- Generate multiple templates at once to choose from.
+- Iterate all of them by providing more inputs in the text field.
+- Once you're done, click "Apply in Chat" to jump to the normal chat (use as any other template, eg, fill in variables at the top).
+
 And rich logging in the REPL to see what the GUI is doing under the hood!
 
 ## Alternatives

main.jl

@@ -1,10 +1,9 @@
 using Pkg
 Pkg.activate(".")
 using GenieFramework
-ENV["GENIE_HOST"] = "0.0.0.0"
+ENV["GENIE_HOST"] = "127.0.0.1"
 ENV["PORT"] = "8000"
-ENV["GENIE_ENV"] = "prod"
-## include("app.jl") # hack for hot-reloading when fixing things
-Genie.Generator.write_secrets_file();
+## ENV["GENIE_ENV"] = "prod"
+include("app.jl") # hack for hot-reloading when fixing things
 Genie.loadapp();
-up(async=false);
+up(async = true);

src/ProToPortal.jl

@@ -10,6 +10,8 @@ using PromptingTools.Experimental.AgentTools: aicodefixer_feedback, airetry!, AI
 const AT = PromptingTools.Experimental.AgentTools
 
 using GenieFramework
+using GenieFramework.JSON3
+using GenieFramework.Stipple.HTTP
 using GenieSession
 using GenieFramework.StippleUI.Layouts: layout
 using GenieFramework.StippleUI.API: kw
@@ -19,7 +21,7 @@ export Genie, Server, up, down
 
 export msg2display, display2msg, template_variables, update_message, update_message!
 export render_messages, render_template_messages
-export conversation2transcript, parse_critic, load_conversations_from_dir
+export conversation2transcript, parse_critic, parse_builder, load_conversations_from_dir
 include("utils.jl")
 
 export save_conversation
@@ -28,11 +30,15 @@ include("serialization.jl")
 export flash, flash_has_message
 include("flash.jl")
 
+export openai_whisper
+include("speech_to_text.jl")
+
 export messagecard, templatecard
 include("components.jl")
 
 include("view_chat.jl")
 include("view_meta.jl")
+include("view_builder.jl")
 
 export ui, ui_login
 include("view.jl")