Merge pull request #21 from henk717/united

Update to current

aiserver.py

@@ -351,6 +351,8 @@ class vars:
     lazy_load   = True  # Whether or not to use torch_lazy_loader.py for transformers models in order to reduce CPU memory usage
     use_colab_tpu = os.environ.get("COLAB_TPU_ADDR", "") != "" or os.environ.get("TPU_NAME", "") != ""  # Whether or not we're in a Colab TPU instance or Kaggle TPU instance and are going to use the TPU rather than the CPU
     revision    = None
+    output_streaming = False
+    token_stream_queue = [] # Queue for the token streaming
 
 utils.vars = vars
 
@@ -800,6 +802,7 @@ def savesettings():
     js["fulldeterminism"] = vars.full_determinism
     js["autosave"]    = vars.autosave
     js["welcome"]     = vars.welcome
+    js["output_streaming"] = vars.output_streaming
 
     if(vars.seed_specified):
         js["seed"]    = vars.seed
@@ -911,6 +914,8 @@ def processsettings(js):
         vars.newlinemode = js["newlinemode"]
     if("welcome" in js):
         vars.welcome = js["welcome"]
+    if("output_streaming" in js):
+        vars.output_streaming = js["output_streaming"]
     
     if("seed" in js):
         vars.seed = js["seed"]
@@ -943,12 +948,20 @@ def processsettings(js):
 
 def check_for_sp_change():
     while(True):
-        time.sleep(0.1)
+        time.sleep(0.05)
+
         if(vars.sp_changed):
             with app.app_context():
                 emit('from_server', {'cmd': 'spstatitems', 'data': {vars.spfilename: vars.spmeta} if vars.allowsp and len(vars.spfilename) else {}}, namespace=None, broadcast=True)
             vars.sp_changed = False
 
+        if(vars.output_streaming and vars.token_stream_queue):
+            # If emit blocks, waiting for it to complete before clearing could
+            # introduce a race condition that drops tokens.
+            queued_tokens = list(vars.token_stream_queue)
+            vars.token_stream_queue.clear()
+            socketio.emit("from_server", {"cmd": "streamtoken", "data": queued_tokens}, namespace=None, broadcast=True)
+
 socketio.start_background_task(check_for_sp_change)
 
 def spRequest(filename):
@@ -1044,6 +1057,7 @@ def general_startup(override_args=None):
     parser.add_argument("--no_aria2", action='store_true', default=False, help="Prevents KoboldAI from using aria2 to download huggingface models more efficiently, in case aria2 is causing you issues")
     parser.add_argument("--lowmem", action='store_true', help="Extra Low Memory loading for the GPU, slower but memory does not peak to twice the usage")
     parser.add_argument("--savemodel", action='store_true', help="Saves the model to the models folder even if --colab is used (Allows you to save models to Google Drive)")
+    parser.add_argument("--customsettings", help="Preloads arguements from json file. You only need to provide the location of the json file. Use customsettings.json template file. It can be renamed if you wish so that you can store multiple configurations. Leave any settings you want as default as null. Any values you wish to set need to be in double quotation marks")
     #args: argparse.Namespace = None
     if "pytest" in sys.modules and override_args is None:
         args = parser.parse_args([])
@@ -1057,6 +1071,14 @@ def general_startup(override_args=None):
     else:
         args = parser.parse_args()
 
+    if args.customsettings:
+        f = open (args.customsettings)
+        importedsettings = json.load(f)
+        for items in importedsettings:
+            if importedsettings[items] is not None:
+                setattr(args, items, importedsettings[items])            
+        f.close()
+
     vars.model = args.model;
     vars.revision = args.revision
 
@@ -1533,6 +1555,27 @@ def patch_transformers():
     new_init.old_init = transformers.generation_logits_process.NoBadWordsLogitsProcessor.__init__
     transformers.generation_logits_process.NoBadWordsLogitsProcessor.__init__ = new_init
 
+    class TokenStreamer(StoppingCriteria):
+        # A StoppingCriteria is used here because it seems to run after
+        # everything has been evaluated score-wise. 
+        def __init__(self, tokenizer):
+            self.tokenizer = tokenizer
+
+        def __call__(
+            self,
+            input_ids: torch.LongTensor,
+            scores: torch.FloatTensor,
+            **kwargs,
+        ) -> bool:
+            # Do not intermingle multiple generations' outputs!
+            if(vars.numseqs > 1):
+                return False
+
+            tokenizer_text = utils.decodenewlines(tokenizer.decode(input_ids[0, -1]))
+
+            vars.token_stream_queue.append(tokenizer_text)
+            return False
+
 
     # Sets up dynamic world info scanner
     class DynamicWorldInfoScanCriteria(StoppingCriteria):
@@ -1587,7 +1630,10 @@ def patch_transformers():
             tokenizer=tokenizer,
             excluded_world_info=self.kai_scanner_excluded_world_info,
         )
+        token_streamer = TokenStreamer(tokenizer=tokenizer)
+
         stopping_criteria.insert(0, self.kai_scanner)
+        stopping_criteria.insert(0, token_streamer)
         return stopping_criteria
     transformers.generation_utils.GenerationMixin._get_stopping_criteria = new_get_stopping_criteria
 
@@ -2689,6 +2735,7 @@ def lua_has_setting(setting):
         "rmspch",
         "adsnsp",
         "singleline",
+        "output_streaming"
     )
 
 #==================================================================#
@@ -2720,6 +2767,7 @@ def lua_get_setting(setting):
     if(setting in ("frmtrmspch", "rmspch")): return vars.formatoptns["frmttrmspch"]
     if(setting in ("frmtadsnsp", "adsnsp")): return vars.formatoptns["frmtadsnsp"]
     if(setting in ("frmtsingleline", "singleline")): return vars.formatoptns["singleline"]
+    if(setting == "output_streaming"): return vars.output_streaming
 
 #==================================================================#
 #  Set the setting with the given name if it exists
@@ -2756,6 +2804,7 @@ def lua_set_setting(setting, v):
     if(setting in ("frmtrmspch", "rmspch")): vars.formatoptns["frmttrmspch"] = v
     if(setting in ("frmtadsnsp", "adsnsp")): vars.formatoptns["frmtadsnsp"] = v
     if(setting in ("frmtsingleline", "singleline")): vars.formatoptns["singleline"] = v
+    if(setting == "output_streaming"): vars.output_streaming = v
 
 #==================================================================#
 #  Get contents of memory
@@ -3468,6 +3517,10 @@ def get_message(msg):
         vars.full_determinism = msg['data']
         settingschanged()
         refresh_settings()
+    elif(msg['cmd'] == 'setoutputstreaming'):
+        vars.output_streaming = msg['data']
+        settingschanged()
+        refresh_settings()
     elif(not vars.host and msg['cmd'] == 'importwi'):
         wiimportrequest()
     elif(msg['cmd'] == 'debug'):
@@ -4665,6 +4718,7 @@ def refresh_settings():
     emit('from_server', {'cmd': 'updatefrmtrmspch', 'data': vars.formatoptns["frmtrmspch"]}, broadcast=True)
     emit('from_server', {'cmd': 'updatefrmtadsnsp', 'data': vars.formatoptns["frmtadsnsp"]}, broadcast=True)
     emit('from_server', {'cmd': 'updatesingleline', 'data': vars.formatoptns["singleline"]}, broadcast=True)
+    emit('from_server', {'cmd': 'updateoutputstreaming', 'data': vars.output_streaming}, broadcast=True)
     
     # Allow toggle events again
     emit('from_server', {'cmd': 'allowtoggle', 'data': True}, broadcast=True)
@@ -5838,7 +5892,7 @@ def importgame():
 def importAidgRequest(id):    
     exitModes()
     
-    urlformat = "https://prompts.aidg.club/api/"
+    urlformat = "https://aetherroom.club/api/"
     req = requests.get(urlformat+id)
 
     if(req.status_code == 200):

customsettings_template.json

@@ -0,0 +1 @@
+{"aria2_port":null, "breakmodel":null, "breakmodel_disklayers":null, "breakmodel_gpulayers":null, "breakmodel_layers":null, "colab":null, "configname":null, "cpu":null, "host":null, "localtunnel":null, "lowmem":null, "model":null, "ngrok":null, "no_aria2":null, "noaimenu":null, "nobreakmodel":null, "override_delete":null, "override_rename":null, "path":null, "port":null, "quiet":null, "remote":null, "revision":null, "savemodel":null, "unblock":null}

gensettings.py

@@ -251,7 +251,18 @@ gensettingstf = [
 	"step": 1,
 	"default": 0,
   "tooltip": "Show debug info"
-	}
+	},
+    {
+	"uitype": "toggle",
+	"unit": "bool",
+	"label": "Token Streaming",
+	"id": "setoutputstreaming",
+	"min": 0,
+	"max": 1,
+	"step": 1,
+	"default": 0,
+  "tooltip": "Shows outputs to you as they are made. Does not work with more than one gens per action."
+	},
 ]
 
 gensettingsik =[{
@@ -415,9 +426,9 @@ formatcontrols = [{
     "tooltip": "Remove special characters (@,#,%,^, etc)"
     },
     {
-    "label": "Add sentence spacing",
+    "label": "Automatic spacing",
     "id": "frmtadsnsp",
-    "tooltip": "If the last action ended with punctuation, add a space to the beginning of the next action."
+    "tooltip": "Add spaces automatically if needed"
     },
     {
     "label": "Single Line",

install_requirements.bat

@@ -14,7 +14,7 @@ if exist miniconda3\ (
   echo This is required if you are switching modes, or if you get dependency errors in the game.
   echo 1. Yes
   echo 2. No
-  SET /P D=Type the number of the desired option and then press ENTER: 
+  SET /P D=Type the number of the desired option and then press ENTER:
 ) ELSE (
 	SET D=Workaround
 )
@@ -25,7 +25,7 @@ echo Which installation mode would you like?
 echo 1. Temporary Drive Letter (Mounts the folder as drive B:, more stable and portable)
 echo 2. Subfolder (Traditional method, can't run in folder paths that contain spaces)
 echo.
-SET /P M=Type the number of the desired option and then press ENTER: 
+SET /P M=Type the number of the desired option and then press ENTER:
 IF %M%==1 GOTO drivemap
 IF %M%==2 GOTO subfolder
 ECHO Incorrect choice
@@ -35,7 +35,7 @@ GOTO MODE
 :drivemap
 echo 3 > loader.settings
 subst B: /D >nul
-mkdir miniconda3 
+mkdir miniconda3
 subst B: miniconda3
 SET TEMP=B:\
 SET TMP=B:\
@@ -44,8 +44,7 @@ copy loader.settings B:\loader.settings
 copy disconnect-kobold-drive.bat B:\disconnect-kobold-drive.bat
 B:
 umamba.exe create -r B:\python\ -n base
-IF %B%==1 umamba.exe install --no-shortcuts -r B:\python\ -n base -f "%~dp0\environments\huggingface.yml" -y --always-copy
-IF %B%==2 umamba.exe install --no-shortcuts -r B:\python\ -n base -f "%~dp0\environments\finetuneanon.yml" -y --always-copy
+umamba.exe install --no-shortcuts -r B:\python\ -n base -f "%~dp0\environments\huggingface.yml" -y --always-copy
 umamba.exe -r B:\ clean -a -y
 rd B:\Python\pkgs /S /Q
 subst B: /d
@@ -57,8 +56,7 @@ echo 2 > loader.settings
 SET TEMP=%~DP0MINICONDA3
 SET TMP=%~DP0MINICONDA3
 umamba.exe create -r miniconda3\ -n base
-IF %B%==1 umamba.exe install --no-shortcuts -r miniconda3 -n base -f environments\huggingface.yml -y --always-copy
-IF %B%==2 umamba.exe install --no-shortcuts -r miniconda3 -n base -f environments\finetuneanon.yml -y --always-copy
+umamba.exe install --no-shortcuts -r miniconda3 -n base -f environments\huggingface.yml -y --always-copy
 umamba.exe clean -a -y
 rd miniconda3\Python\pkgs /S /Q
 pause

static/application.js

@@ -78,6 +78,7 @@ var rs_accept;
 var rs_close;
 var seqselmenu;
 var seqselcontents;
+var stream_preview;
 
 var storyname = null;
 var memorymode = false;
@@ -103,6 +104,7 @@ var gamestate = "";
 var gamesaved = true;
 var modelname = null;
 var model = "";
+var ignore_stream = false;
 
 // This is true iff [we're in macOS and the browser is Safari] or [we're in iOS]
 var using_webkit_patch = true;
@@ -888,6 +890,7 @@ function formatChunkInnerText(chunk) {
 }
 
 function dosubmit(disallow_abort) {
+	ignore_stream = false;
 	submit_start = Date.now();
 	var txt = input_text.val().replace(/\u00a0/g, " ");
 	if((disallow_abort || gamestate !== "wait") && !memorymode && !gamestarted && ((!adventure || !action_mode) && txt.trim().length == 0)) {
@@ -902,6 +905,7 @@ function dosubmit(disallow_abort) {
 }
 
 function _dosubmit() {
+	ignore_stream = false;
 	var txt = submit_throttle.txt;
 	var disallow_abort = submit_throttle.disallow_abort;
 	submit_throttle = null;
@@ -2082,6 +2086,15 @@ function unbindGametext() {
 	gametext_bound = false;
 }
 
+function endStream() {
+	// Clear stream, the real text is about to be displayed.
+	ignore_stream = true;
+	if (stream_preview) {
+		stream_preview.remove();
+		stream_preview = null;
+	}
+}
+
 function update_gpu_layers() {
 	var gpu_layers
 	gpu_layers = 0;
@@ -2258,6 +2271,21 @@ $(document).ready(function(){
 				active_element.focus();
 			})();
 			$("body").addClass("connected");
+		} else if (msg.cmd == "streamtoken") {
+			// Sometimes the stream_token messages will come in too late, after
+			// we have recieved the full text. This leads to some stray tokens
+			// appearing after the output. To combat this, we only allow tokens
+			// to be displayed after requesting and before recieving text.
+			if (ignore_stream) return;
+			if (!$("#setoutputstreaming")[0].checked) return;
+
+			if (!stream_preview) {
+				stream_preview = document.createElement("span");
+				game_text.append(stream_preview);
+			}
+
+			stream_preview.innerText += msg.data.join("");
+			scrollToBottom();
 		} else if(msg.cmd == "updatescreen") {
 			var _gamestarted = gamestarted;
 			gamestarted = msg.gamestarted;
@@ -2307,7 +2335,11 @@ $(document).ready(function(){
 			} else if (!empty_chunks.has(index.toString())) {
 				// Append at the end
 				unbindGametext();
-				var lc = game_text[0].lastChild;
+
+				// game_text can contain things other than chunks (stream
+				// preview), so we use querySelector to get the last chunk.
+				var lc = game_text[0].querySelector("chunk:last-of-type");
+
 				if(lc.tagName === "CHUNK" && lc.lastChild !== null && lc.lastChild.tagName === "BR") {
 					lc.removeChild(lc.lastChild);
 				}
@@ -2323,7 +2355,11 @@ $(document).ready(function(){
 			var element = game_text.children('#n' + index);
 			if(element.length) {
 				unbindGametext();
-				if((element[0].nextSibling === null || element[0].nextSibling.nodeType !== 1 || element[0].nextSibling.tagName !== "CHUNK") && element[0].previousSibling !== null && element[0].previousSibling.tagName === "CHUNK") {
+				if(
+					(element[0].nextSibling === null || element[0].nextSibling.nodeType !== 1 || element[0].nextSibling.tagName !== "CHUNK")
+					&& element[0].previousSibling !== null
+					&& element[0].previousSibling.tagName === "CHUNK"
+				) {
 					element[0].previousSibling.appendChild(document.createElement("br"));
 				}
 				element.remove();  // Remove the chunk
@@ -2333,6 +2369,7 @@ $(document).ready(function(){
 		} else if(msg.cmd == "setgamestate") {
 			// Enable or Disable buttons
 			if(msg.data == "ready") {
+				endStream();
 				enableSendBtn();
 				enableButtons([button_actmem, button_actwi, button_actback, button_actfwd, button_actretry]);
 				hideWaitAnimation();
@@ -2519,6 +2556,9 @@ $(document).ready(function(){
 		} else if(msg.cmd == "updatesingleline") {
 			// Update toggle state
 			$("#singleline").prop('checked', msg.data).change();
+		} else if(msg.cmd == "updateoutputstreaming") {
+			// Update toggle state
+			$("#setoutputstreaming").prop('checked', msg.data).change();
 		} else if(msg.cmd == "allowtoggle") {
 			// Allow toggle change states to propagate
 			allowtoggle = msg.data;
@@ -2914,6 +2954,7 @@ $(document).ready(function(){
 	});
 	
 	button_actretry.on("click", function(ev) {
+		ignore_stream = false;
 		hideMessage();
 		socket.send({'cmd': 'retry', 'chatname': chatmode ? chat_name.val() : undefined, 'data': ''});
 		hidegenseqs();
@@ -3160,6 +3201,7 @@ $(document).ready(function(){
 	});
 	
 	rs_accept.on("click", function(ev) {
+		ignore_stream = false;
 		hideMessage();
 		socket.send({'cmd': 'rndgame', 'memory': $("#rngmemory").val(), 'data': topic.val()});
 		hideRandomStoryPopup();
@@ -3233,6 +3275,32 @@ $(document).ready(function(){
 			return true;
 		}
 	});
+
+	// Shortcuts
+	$(window).keydown(function (ev) {
+		// Only ctrl prefixed (for now)
+		if (!ev.ctrlKey) return;
+
+		let handled = true;
+		switch (ev.key) {
+			// Ctrl+Z - Back
+			case "z":
+				button_actback.click();
+				break;
+			// Ctrl+Y - Forward
+			case "y":
+				button_actfwd.click();
+				break;
+			// Ctrl+E - Retry
+			case "e":
+				button_actretry.click();
+				break;
+			default:
+				handled = false;
+		}
+
+		if (handled) ev.preventDefault();
+	});
 });
 
 

templates/index.html

@@ -76,7 +76,7 @@
 									<div class="dropdown-menu">
 										<a class="dropdown-item" href="#" id="btn_import">AI Dungeon Adventure</a>
 										<a class="dropdown-item" href="#" id="btn_importwi">AI Dungeon World Info</a>
-										<a class="dropdown-item" href="#" id="btn_impaidg">aidg.club Prompt</a>
+										<a class="dropdown-item" href="#" id="btn_impaidg">aetherroom.club Prompt</a>
 									</div>
 								</li>
 								<li class="nav-item">
@@ -233,7 +233,7 @@
 				<div class="popuptitletext">Enter the Prompt Number</div>
 			</div>
 			<div class="aidgpopuplistheader">
-				(4-digit number at the end of aidg.club URL)
+				(4-digit number at the end of aetherroom.club URL)
 			</div>
 			<div class="aidgpopupcontent">
 				<input class="form-control" type="text" placeholder="Prompt Number" id="aidgpromptnum">

utils.py

@@ -210,7 +210,7 @@ def aria2_hook(pretrained_model_name_or_path: str, force_download=False, cache_d
     def is_cached(url):
         try:
             transformers.file_utils.get_from_cache(url, cache_dir=cache_dir, local_files_only=True)
-        except FileNotFoundError:
+        except (FileNotFoundError, transformers.file_utils.EntryNotFoundError):
             return False
         return True
     while True:  # Try to get the huggingface.co URL of the model's pytorch_model.bin or pytorch_model.bin.index.json file