Merge branch 'main' into catalpaaa-lora-and-model-dir

.gitignore

@@ -1,26 +1,22 @@
-cache/*
-characters/*
-extensions/silero_tts/outputs/*
-extensions/elevenlabs_tts/outputs/*
-extensions/sd_api_pictures/outputs/*
-logs/*
-loras/*
-models/*
-softprompts/*
-torch-dumps/*
+cache
+characters
+training/datasets
+extensions/silero_tts/outputs
+extensions/elevenlabs_tts/outputs
+extensions/sd_api_pictures/outputs
+logs
+loras
+models
+softprompts
+torch-dumps
 *pycache*
 */*pycache*
 */*/pycache*
 venv/
 .venv/
+repositories
 
 settings.json
 img_bot*
 img_me*
-
-!characters/Example.json
-!characters/Example.png
-!loras/place-your-loras-here.txt
-!models/place-your-models-here.txt
-!softprompts/place-your-softprompts-here.txt
-!torch-dumps/place-your-pt-models-here.txt
+prompts/[0-9]*

README.md

@@ -176,10 +176,10 @@ Optionally, you can use the following command-line flags:
 | `--cai-chat`     | Launch the web UI in chat mode with a style similar to Character.AI's. If the file `img_bot.png` or `img_bot.jpg` exists in the same folder as server.py, this image will be used as the bot's profile picture. Similarly, `img_me.png` or `img_me.jpg` will be used as your profile picture. |
 | `--cpu`          | Use the CPU to generate text.|
 | `--load-in-8bit` | Load the model with 8-bit precision.|
-| `--load-in-4bit` | DEPRECATED: use `--gptq-bits 4` instead. |
-| `--gptq-bits GPTQ_BITS` |  GPTQ: Load a pre-quantized model with specified precision. 2, 3, 4 and 8 (bit) are supported. Currently only works with LLaMA and OPT. |
-| `--gptq-model-type MODEL_TYPE` |  GPTQ: Model type of pre-quantized model. Currently only LLaMa and OPT are supported. |
-| `--gptq-pre-layer GPTQ_PRE_LAYER` |  GPTQ: The number of layers to preload. |
+| `--wbits WBITS`            | GPTQ: Load a pre-quantized model with specified precision in bits. 2, 3, 4 and 8 are supported. |
+| `--model_type MODEL_TYPE`  | GPTQ: Model type of pre-quantized model. Currently only LLaMA and OPT are supported. |
+| `--groupsize GROUPSIZE`    | GPTQ: Group size. |
+| `--pre_layer PRE_LAYER`    | GPTQ: The number of layers to preload. |
 | `--bf16`         | Load the model with bfloat16 precision. Requires NVIDIA Ampere GPU. |
 | `--auto-devices` | Automatically split the model across the available GPU(s) and CPU.|
 | `--disk`         | If the model is too large for your GPU(s) and CPU combined, send the remaining layers to the disk. |

css/chat.css

@@ -23,3 +23,9 @@ div.svelte-362y77>*, div.svelte-362y77>.form>* {
 .pending.svelte-1ed2p3z {
     opacity: 1;
 }
+
+#extensions {
+  padding: 0;
+  padding: 0;
+}
+

css/main.css

@@ -37,12 +37,6 @@
   text-decoration: none !important;
 }
 
-svg {
-  display: unset !important;
-  vertical-align: middle !important;
-  margin: 5px;
-}
-
 ol li p, ul li p {
     display: inline-block;
 }
@@ -54,3 +48,18 @@ ol li p, ul li p {
 .gradio-container-3-18-0 .prose * h1, h2, h3, h4 {
   color: white;
 }
+
+.gradio-container {
+  max-width: 100% !important;
+  padding-top: 0 !important;
+}
+
+#extensions {
+  padding: 15px;
+  padding: 15px;
+}
+
+span.math.inline {
+  font-size: 27px;
+  vertical-align: baseline !important;
+}

css/main.js

@@ -11,7 +11,7 @@ let extensions = document.getElementById('extensions');
 main_parent.addEventListener('click', function(e) {
     // Check if the main element is visible
     if (main.offsetHeight > 0 && main.offsetWidth > 0) {
-        extensions.style.display = 'block';
+        extensions.style.display = 'flex';
     } else {
         extensions.style.display = 'none';
     }

download-model.py

@@ -118,7 +118,7 @@ def get_download_links_from_huggingface(model, branch):
             is_safetensors = re.match("model.*\.safetensors", fname)
             is_pt = re.match(".*\.pt", fname)
             is_tokenizer = re.match("tokenizer.*\.model", fname)
-            is_text = re.match(".*\.(txt|json|py)", fname) or is_tokenizer
+            is_text = re.match(".*\.(txt|json|py|md)", fname) or is_tokenizer
 
             if any((is_pytorch, is_safetensors, is_pt, is_tokenizer, is_text)):
                 if is_text:

extensions/silero_tts/script.py

@@ -26,6 +26,7 @@ current_params = params.copy()
 voices_by_gender = ['en_99', 'en_45', 'en_18', 'en_117', 'en_49', 'en_51', 'en_68', 'en_0', 'en_26', 'en_56', 'en_74', 'en_5', 'en_38', 'en_53', 'en_21', 'en_37', 'en_107', 'en_10', 'en_82', 'en_16', 'en_41', 'en_12', 'en_67', 'en_61', 'en_14', 'en_11', 'en_39', 'en_52', 'en_24', 'en_97', 'en_28', 'en_72', 'en_94', 'en_36', 'en_4', 'en_43', 'en_88', 'en_25', 'en_65', 'en_6', 'en_44', 'en_75', 'en_91', 'en_60', 'en_109', 'en_85', 'en_101', 'en_108', 'en_50', 'en_96', 'en_64', 'en_92', 'en_76', 'en_33', 'en_116', 'en_48', 'en_98', 'en_86', 'en_62', 'en_54', 'en_95', 'en_55', 'en_111', 'en_3', 'en_83', 'en_8', 'en_47', 'en_59', 'en_1', 'en_2', 'en_7', 'en_9', 'en_13', 'en_15', 'en_17', 'en_19', 'en_20', 'en_22', 'en_23', 'en_27', 'en_29', 'en_30', 'en_31', 'en_32', 'en_34', 'en_35', 'en_40', 'en_42', 'en_46', 'en_57', 'en_58', 'en_63', 'en_66', 'en_69', 'en_70', 'en_71', 'en_73', 'en_77', 'en_78', 'en_79', 'en_80', 'en_81', 'en_84', 'en_87', 'en_89', 'en_90', 'en_93', 'en_100', 'en_102', 'en_103', 'en_104', 'en_105', 'en_106', 'en_110', 'en_112', 'en_113', 'en_114', 'en_115']
 voice_pitches = ['x-low', 'low', 'medium', 'high', 'x-high']
 voice_speeds = ['x-slow', 'slow', 'medium', 'fast', 'x-fast']
+streaming_state = shared.args.no_stream # remember if chat streaming was enabled
 
 # Used for making text xml compatible, needed for voice pitch and speed control
 table = str.maketrans({
@@ -77,6 +78,7 @@ def input_modifier(string):
         shared.history['visible'][-1] = [shared.history['visible'][-1][0], shared.history['visible'][-1][1].replace('controls autoplay>','controls>')]
 
     shared.processing_message = "*Is recording a voice message...*"
+    shared.args.no_stream = True # Disable streaming cause otherwise the audio output will stutter and begin anew every time the message is being updated
     return string
 
 def output_modifier(string):
@@ -84,7 +86,7 @@ def output_modifier(string):
     This function is applied to the model outputs.
     """
 
-    global model, current_params
+    global model, current_params, streaming_state
 
     for i in params:
         if params[i] != current_params[i]:
@@ -116,6 +118,7 @@ def output_modifier(string):
             string += f'\n\n{original_string}'
 
     shared.processing_message = "*Is typing...*"
+    shared.args.no_stream = streaming_state # restore the streaming option to the previous value
     return string
 
 def bot_prefix_modifier(string):

modules/GPTQ_loader.py

@@ -14,18 +14,21 @@ import opt
 
 
 def load_quantized(model_name):
-    if not shared.args.gptq_model_type:
+    if not shared.args.model_type:
         # Try to determine model type from model name
-        model_type = model_name.split('-')[0].lower()
-        if model_type not in ('llama', 'opt'):
-            print("Can't determine model type from model name. Please specify it manually using --gptq-model-type "
+        if model_name.lower().startswith(('llama', 'alpaca')):
+            model_type = 'llama'
+        elif model_name.lower().startswith(('opt', 'galactica')):
+            model_type = 'opt'
+        else:
+            print("Can't determine model type from model name. Please specify it manually using --model_type "
                   "argument")
             exit()
     else:
-        model_type = shared.args.gptq_model_type.lower()
+        model_type = shared.args.model_type.lower()
 
     if model_type == 'llama':
-        if not shared.args.gptq_pre_layer:
+        if not shared.args.pre_layer:
             load_quant = llama.load_quant
         else:
             load_quant = llama_inference_offload.load_quant
@@ -35,35 +38,44 @@ def load_quantized(model_name):
         print("Unknown pre-quantized model type specified. Only 'llama' and 'opt' are supported")
         exit()
 
+    # Now we are going to try to locate the quantized model file.
     path_to_model = Path(f'models/{model_name}')
-    if path_to_model.name.lower().startswith('llama-7b'):
-        pt_model = f'llama-7b-{shared.args.gptq_bits}bit'
-    elif path_to_model.name.lower().startswith('llama-13b'):
-        pt_model = f'llama-13b-{shared.args.gptq_bits}bit'
-    elif path_to_model.name.lower().startswith('llama-30b'):
-        pt_model = f'llama-30b-{shared.args.gptq_bits}bit'
-    elif path_to_model.name.lower().startswith('llama-65b'):
-        pt_model = f'llama-65b-{shared.args.gptq_bits}bit'
+    found_pts = list(path_to_model.glob("*.pt"))
+    found_safetensors = list(path_to_model.glob("*.safetensors"))
+    pt_path = None
+
+    if len(found_pts) == 1:
+        pt_path = found_pts[0]
+    elif len(found_safetensors) == 1:
+        pt_path = found_safetensors[0]
     else:
-        pt_model = f'{model_name}-{shared.args.gptq_bits}bit'
+        if path_to_model.name.lower().startswith('llama-7b'):
+            pt_model = f'llama-7b-{shared.args.wbits}bit'
+        elif path_to_model.name.lower().startswith('llama-13b'):
+            pt_model = f'llama-13b-{shared.args.wbits}bit'
+        elif path_to_model.name.lower().startswith('llama-30b'):
+            pt_model = f'llama-30b-{shared.args.wbits}bit'
+        elif path_to_model.name.lower().startswith('llama-65b'):
+            pt_model = f'llama-65b-{shared.args.wbits}bit'
+        else:
+            pt_model = f'{model_name}-{shared.args.wbits}bit'
 
-    # Try to find the .safetensors or .pt both in models/ and in the subfolder
-    pt_path = None
-    for path in [Path(p+ext) for ext in ['.safetensors', '.pt'] for p in [f"models/{pt_model}", f"{path_to_model}/{pt_model}"]]:
-        if path.exists():
-            print(f"Found {path}")
-            pt_path = path
-            break
+        # Try to find the .safetensors or .pt both in models/ and in the subfolder
+        for path in [Path(p+ext) for ext in ['.safetensors', '.pt'] for p in [f"models/{pt_model}", f"{path_to_model}/{pt_model}"]]:
+            if path.exists():
+                print(f"Found {path}")
+                pt_path = path
+                break
 
     if not pt_path:
-        print(f"Could not find {pt_model}, exiting...")
+        print("Could not find the quantized model in .pt or .safetensors format, exiting...")
         exit()
 
     # qwopqwop200's offload
-    if shared.args.gptq_pre_layer:
-        model = load_quant(str(path_to_model), str(pt_path), shared.args.gptq_bits, shared.args.gptq_pre_layer)
+    if shared.args.pre_layer:
+        model = load_quant(str(path_to_model), str(pt_path), shared.args.wbits, shared.args.groupsize, shared.args.pre_layer)
     else:
-        model = load_quant(str(path_to_model), str(pt_path), shared.args.gptq_bits)
+        model = load_quant(str(path_to_model), str(pt_path), shared.args.wbits, shared.args.groupsize)
 
         # accelerate offload (doesn't work properly)
         if shared.args.gpu_memory:

modules/LoRA.py

@@ -18,11 +18,11 @@ def add_lora_to_model(lora_name):
 
     # If a LoRA had been previously loaded, or if we want
     # to unload a LoRA, reload the model
-    if shared.lora_name != "None" or lora_name == "None":
+    if shared.lora_name not in ['None', ''] or lora_name in ['None', '']:
         reload_model()
     shared.lora_name = lora_name
 
-    if lora_name != "None":
+    if lora_name not in ['None', '']:
         print(f"Adding the LoRA {lora_name} to the model...")
         params = {}
         if not shared.args.cpu:

modules/callbacks.py

@@ -25,7 +25,7 @@ class _SentinelTokenStoppingCriteria(transformers.StoppingCriteria):
                 if trimmed_sample.shape[-1] < self.sentinel_token_ids[i].shape[-1]:
                     continue
                 for window in trimmed_sample.unfold(0, self.sentinel_token_ids[i].shape[-1], 1):
-                    if torch.all(torch.eq(self.sentinel_token_ids[i], window)):
+                    if torch.all(torch.eq(self.sentinel_token_ids[i][0], window)):
                         return True
         return False
 
@@ -54,7 +54,7 @@ class Iteratorize:
         self.stop_now = False
 
         def _callback(val):
-            if self.stop_now:
+            if self.stop_now or shared.stop_everything:
                 raise ValueError
             self.q.put(val)
 

modules/chat.py

@@ -80,11 +80,7 @@ def extract_message_from_reply(reply, name1, name2, check):
     reply = fix_newlines(reply)
     return reply, next_character_found
 
-def stop_everything_event():
-    shared.stop_everything = True
-
 def chatbot_wrapper(text, max_new_tokens, do_sample, temperature, top_p, typical_p, repetition_penalty, encoder_repetition_penalty, top_k, min_length, no_repeat_ngram_size, num_beams, penalty_alpha, length_penalty, early_stopping, seed, name1, name2, context, check, chat_prompt_size, chat_generation_attempts=1, regenerate=False):
-    shared.stop_everything = False
     just_started = True
     eos_token = '\n' if check else None
     name1_original = name1

modules/extensions.py

@@ -63,8 +63,8 @@ def create_extensions_block():
 
     # Creating the extension ui elements
     if should_display_ui:
-        with gr.Box(elem_id="extensions"):
-            gr.Markdown("Extensions")
+        with gr.Column(elem_id="extensions"):
             for extension, name in iterator():
+                gr.Markdown(f"\n### {name}")
                 if hasattr(extension, "ui"):
                     extension.ui()

modules/models.py

@@ -44,7 +44,7 @@ def load_model(model_name):
     shared.is_RWKV = model_name.lower().startswith('rwkv-')
 
     # Default settings
-    if not any([shared.args.cpu, shared.args.load_in_8bit, shared.args.gptq_bits, shared.args.auto_devices, shared.args.disk, shared.args.gpu_memory is not None, shared.args.cpu_memory is not None, shared.args.deepspeed, shared.args.flexgen, shared.is_RWKV]):
+    if not any([shared.args.cpu, shared.args.load_in_8bit, shared.args.wbits, shared.args.auto_devices, shared.args.disk, shared.args.gpu_memory is not None, shared.args.cpu_memory is not None, shared.args.deepspeed, shared.args.flexgen, shared.is_RWKV]):
         if any(size in shared.model_name.lower() for size in ('13b', '20b', '30b')):
             model = AutoModelForCausalLM.from_pretrained(Path(f"{shared.args.model_dir}/{shared.model_name}"), device_map='auto', load_in_8bit=True)
         else:
@@ -95,7 +95,7 @@ def load_model(model_name):
         return model, tokenizer
 
     # Quantized model
-    elif shared.args.gptq_bits > 0:
+    elif shared.args.wbits > 0:
         from modules.GPTQ_loader import load_quantized
 
         model = load_quantized(model_name)

modules/shared.py

@@ -52,7 +52,8 @@ settings = {
         'default': 'Common sense questions and answers\n\nQuestion: \nFactual answer:',
         '^(gpt4chan|gpt-4chan|4chan)': '-----\n--- 865467536\nInput text\n--- 865467537\n',
         '(rosey|chip|joi)_.*_instruct.*': 'User: \n',
-        'oasst-*': '<|prompter|>Write a story about future of AI development<|endoftext|><|assistant|>'
+        'oasst-*': '<|prompter|>Write a story about future of AI development<|endoftext|><|assistant|>',
+        'alpaca-*': "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n### Instruction:\nWrite a poem about the transformers Python library. \nMention the word \"large language models\" in that poem.\n### Response:\n",
     },
     'lora_prompts': {
         'default': 'Common sense questions and answers\n\nQuestion: \nFactual answer:',
@@ -78,10 +79,15 @@ parser.add_argument('--chat', action='store_true', help='Launch the web UI in ch
 parser.add_argument('--cai-chat', action='store_true', help='Launch the web UI in chat mode with a style similar to Character.AI\'s. If the file img_bot.png or img_bot.jpg exists in the same folder as server.py, this image will be used as the bot\'s profile picture. Similarly, img_me.png or img_me.jpg will be used as your profile picture.')
 parser.add_argument('--cpu', action='store_true', help='Use the CPU to generate text.')
 parser.add_argument('--load-in-8bit', action='store_true', help='Load the model with 8-bit precision.')
-parser.add_argument('--load-in-4bit', action='store_true', help='DEPRECATED: use --gptq-bits 4 instead.')
-parser.add_argument('--gptq-bits', type=int, default=0, help='GPTQ: Load a pre-quantized model with specified precision. 2, 3, 4 and 8bit are supported. Currently only works with LLaMA and OPT.')
-parser.add_argument('--gptq-model-type', type=str, help='GPTQ: Model type of pre-quantized model. Currently only LLaMa and OPT are supported.')
-parser.add_argument('--gptq-pre-layer', type=int, default=0, help='GPTQ: The number of layers to preload.')
+
+parser.add_argument('--gptq-bits', type=int, default=0, help='DEPRECATED: use --wbits instead.')
+parser.add_argument('--gptq-model-type', type=str, help='DEPRECATED: use --model_type instead.')
+parser.add_argument('--gptq-pre-layer', type=int, default=0, help='DEPRECATED: use --pre_layer instead.')
+parser.add_argument('--wbits', type=int, default=0, help='GPTQ: Load a pre-quantized model with specified precision in bits. 2, 3, 4 and 8 are supported.')
+parser.add_argument('--model_type', type=str, help='GPTQ: Model type of pre-quantized model. Currently only LLaMA and OPT are supported.')
+parser.add_argument('--groupsize', type=int, default=-1, help='GPTQ: Group size.')
+parser.add_argument('--pre_layer', type=int, default=0, help='GPTQ: The number of layers to preload.')
+
 parser.add_argument('--bf16', action='store_true', help='Load the model with bfloat16 precision. Requires NVIDIA Ampere GPU.')
 parser.add_argument('--auto-devices', action='store_true', help='Automatically split the model across the available GPU(s) and CPU.')
 parser.add_argument('--disk', action='store_true', help='If the model is too large for your GPU(s) and CPU combined, send the remaining layers to the disk.')
@@ -112,6 +118,8 @@ parser.add_argument("--lora-dir", type=str, default='loras/', help="Path to dire
 args = parser.parse_args()
 
 # Provisional, this will be deleted later
-if args.load_in_4bit:
-    print("Warning: --load-in-4bit is deprecated and will be removed. Use --gptq-bits 4 instead.\n")
-    args.gptq_bits = 4
+deprecated_dict = {'gptq_bits': ['wbits', 0], 'gptq_model_type': ['model_type', None], 'gptq_pre_layer': ['prelayer', 0]}
+for k in deprecated_dict:
+    if eval(f"args.{k}") != deprecated_dict[k][1]:
+        print(f"Warning: --{k} is deprecated and will be removed. Use --{deprecated_dict[k][0]} instead.")
+        exec(f"args.{deprecated_dict[k][0]} = args.{k}")

modules/text_generation.py

@@ -99,9 +99,13 @@ def set_manual_seed(seed):
         if torch.cuda.is_available():
             torch.cuda.manual_seed_all(seed)
 
+def stop_everything_event():
+    shared.stop_everything = True
+
 def generate_reply(question, max_new_tokens, do_sample, temperature, top_p, typical_p, repetition_penalty, encoder_repetition_penalty, top_k, min_length, no_repeat_ngram_size, num_beams, penalty_alpha, length_penalty, early_stopping, seed, eos_token=None, stopping_strings=[]):
     clear_torch_cache()
     set_manual_seed(seed)
+    shared.stop_everything = False
     t0 = time.time()
 
     original_question = question
@@ -236,8 +240,6 @@ def generate_reply(question, max_new_tokens, do_sample, temperature, top_p, typi
                         break
                     yield formatted_outputs(reply, shared.model_name)
 
-                yield formatted_outputs(reply, shared.model_name)
-
         # Stream the output naively for FlexGen since it doesn't support 'stopping_criteria'
         else:
             for i in range(max_new_tokens//8+1):
@@ -270,5 +272,5 @@ def generate_reply(question, max_new_tokens, do_sample, temperature, top_p, typi
         traceback.print_exc()
     finally:
         t1 = time.time()
-        print(f"Output generated in {(t1-t0):.2f} seconds ({(len(output)-len(original_input_ids[0]))/(t1-t0):.2f} tokens/s, {len(output)-len(original_input_ids[0])} tokens)")
+        print(f"Output generated in {(t1-t0):.2f} seconds ({(len(output)-len(original_input_ids[0]))/(t1-t0):.2f} tokens/s, {len(output)-len(original_input_ids[0])} tokens, context {len(original_input_ids[0])})")
         return

prompts/Alpaca.txt

@@ -0,0 +1,6 @@
+Below is an instruction that describes a task. Write a response that appropriately completes the request.
+### Instruction:
+Write a poem about the transformers Python library. 
+Mention the word "large language models" in that poem.
+### Response:
+

prompts/Open Assistant.txt

@@ -0,0 +1 @@
+<|prompter|>Write a story about future of AI development<|endoftext|><|assistant|>

prompts/QA.txt

@@ -0,0 +1,4 @@
+Common sense questions and answers
+
+Question: 
+Factual answer:

requirements.txt

@@ -1,7 +1,7 @@
 accelerate==0.17.1
 bitsandbytes==0.37.1
 flexgen==0.1.7
-gradio==3.18.0
+gradio==3.23.0
 markdown
 numpy
 peft==0.2.0

server.py

@@ -4,6 +4,7 @@ import re
 import sys
 import time
 import zipfile
+from datetime import datetime
 from pathlib import Path
 
 import gradio as gr
@@ -15,7 +16,8 @@ import modules.ui as ui
 from modules.html_generator import generate_chat_html
 from modules.LoRA import add_lora_to_model
 from modules.models import load_model, load_soft_prompt
-from modules.text_generation import clear_torch_cache, generate_reply
+from modules.text_generation import (clear_torch_cache, generate_reply,
+                                     stop_everything_event)
 
 # Loading custom settings
 settings_file = None
@@ -38,6 +40,13 @@ def get_available_models():
 def get_available_presets():
     return sorted(set(map(lambda x : '.'.join(str(x.name).split('.')[:-1]), Path('presets').glob('*.txt'))), key=str.lower)
 
+def get_available_prompts():
+    prompts = []
+    prompts += sorted(set(map(lambda x : '.'.join(str(x.name).split('.')[:-1]), Path('prompts').glob('[0-9]*.txt'))), key=str.lower, reverse=True)
+    prompts += sorted(set(map(lambda x : '.'.join(str(x.name).split('.')[:-1]), Path('prompts').glob('*.txt'))), key=str.lower)
+    prompts += ['None']
+    return prompts
+
 def get_available_characters():
     return ['None'] + sorted(set(map(lambda x : '.'.join(str(x.name).split('.')[:-1]), Path('characters').glob('*.json'))), key=str.lower)
 
@@ -50,12 +59,17 @@ def get_available_softprompts():
 def get_available_loras():
     return ['None'] + sorted([item.name for item in list(Path('shared.args.lora_dir').glob('*')) if not item.name.endswith(('.txt', '-np', '.pt', '.json'))], key=str.lower)
 
+def unload_model():
+    shared.model = shared.tokenizer = None
+    clear_torch_cache()
+
 def load_model_wrapper(selected_model):
     if selected_model != shared.model_name:
         shared.model_name = selected_model
-        shared.model = shared.tokenizer = None
-        clear_torch_cache()
-        shared.model, shared.tokenizer = load_model(shared.model_name)
+
+        unload_model()
+        if selected_model != '':
+            shared.model, shared.tokenizer = load_model(shared.model_name)
 
     return selected_model
 
@@ -93,7 +107,7 @@ def load_preset_values(preset_menu, return_dict=False):
     if return_dict:
         return generate_params
     else:
-        return preset_menu, generate_params['do_sample'], generate_params['temperature'], generate_params['top_p'], generate_params['typical_p'], generate_params['repetition_penalty'], generate_params['encoder_repetition_penalty'], generate_params['top_k'], generate_params['min_length'], generate_params['no_repeat_ngram_size'], generate_params['num_beams'], generate_params['penalty_alpha'], generate_params['length_penalty'], generate_params['early_stopping']
+        return generate_params['do_sample'], generate_params['temperature'], generate_params['top_p'], generate_params['typical_p'], generate_params['repetition_penalty'], generate_params['encoder_repetition_penalty'], generate_params['top_k'], generate_params['min_length'], generate_params['no_repeat_ngram_size'], generate_params['num_beams'], generate_params['penalty_alpha'], generate_params['length_penalty'], generate_params['early_stopping']
 
 def upload_soft_prompt(file):
     with zipfile.ZipFile(io.BytesIO(file)) as zf:
@@ -118,11 +132,45 @@ def create_model_and_preset_menus():
                 shared.gradio['preset_menu'] = gr.Dropdown(choices=available_presets, value=default_preset if not shared.args.flexgen else 'Naive', label='Generation parameters preset')
                 ui.create_refresh_button(shared.gradio['preset_menu'], lambda : None, lambda : {'choices': get_available_presets()}, 'refresh-button')
 
+def save_prompt(text):
+    fname = f"{datetime.now().strftime('%Y-%m-%d-%H:%M:%S')}.txt"
+    with open(Path(f'prompts/{fname}'), 'w', encoding='utf-8') as f:
+        f.write(text)
+    return f"Saved to prompts/{fname}"
+
+def load_prompt(fname):
+    if fname in ['None', '']:
+        return ''
+    else:
+        with open(Path(f'prompts/{fname}.txt'), 'r', encoding='utf-8') as f:
+            return f.read()
+        
+def create_prompt_menus():
+    with gr.Row():
+        with gr.Column():
+            with gr.Row():
+                shared.gradio['prompt_menu'] = gr.Dropdown(choices=get_available_prompts(), value='None', label='Prompt')
+                ui.create_refresh_button(shared.gradio['prompt_menu'], lambda : None, lambda : {'choices': get_available_prompts()}, 'refresh-button')
+
+        with gr.Column():
+            with gr.Column():
+                shared.gradio['save_prompt'] = gr.Button('Save prompt')
+                shared.gradio['status'] = gr.Markdown('Ready')
+
+    shared.gradio['prompt_menu'].change(load_prompt, [shared.gradio['prompt_menu']], [shared.gradio['textbox']], show_progress=False)
+    shared.gradio['save_prompt'].click(save_prompt, [shared.gradio['textbox']], [shared.gradio['status']], show_progress=False)
+
 def create_settings_menus(default_preset):
     generate_params = load_preset_values(default_preset if not shared.args.flexgen else 'Naive', return_dict=True)
 
     with gr.Row():
         with gr.Column():
+            create_model_and_preset_menus()
+        with gr.Column():
+            shared.gradio['seed'] = gr.Number(value=-1, label='Seed (-1 for random)')
+
+    with gr.Row():
+        with gr.Column():
             with gr.Box():
                 gr.Markdown('Custom generation parameters ([reference](https://huggingface.co/docs/transformers/main_classes/text_generation#transformers.GenerationConfig))')
                 with gr.Row():
@@ -151,12 +199,6 @@ def create_settings_menus(default_preset):
                         shared.gradio['length_penalty'] = gr.Slider(-5, 5, value=generate_params['length_penalty'], label='length_penalty')
                 shared.gradio['early_stopping'] = gr.Checkbox(value=generate_params['early_stopping'], label='early_stopping')
 
-            shared.gradio['seed'] = gr.Number(value=-1, label='Seed (-1 for random)')
-
-    with gr.Row():
-        shared.gradio['preset_menu_mirror'] = gr.Dropdown(choices=available_presets, value=default_preset if not shared.args.flexgen else 'Naive', label='Generation parameters preset')
-        ui.create_refresh_button(shared.gradio['preset_menu_mirror'], lambda : None, lambda : {'choices': get_available_presets()}, 'refresh-button')
-
     with gr.Row():
         shared.gradio['lora_menu'] = gr.Dropdown(choices=available_loras, value=shared.lora_name, label='LoRA')
         ui.create_refresh_button(shared.gradio['lora_menu'], lambda : None, lambda : {'choices': get_available_loras()}, 'refresh-button')
@@ -171,8 +213,7 @@ def create_settings_menus(default_preset):
             shared.gradio['upload_softprompt'] = gr.File(type='binary', file_types=['.zip'])
 
     shared.gradio['model_menu'].change(load_model_wrapper, [shared.gradio['model_menu']], [shared.gradio['model_menu']], show_progress=True)
-    shared.gradio['preset_menu'].change(load_preset_values, [shared.gradio['preset_menu']], [shared.gradio[k] for k in ['preset_menu_mirror', 'do_sample', 'temperature', 'top_p', 'typical_p', 'repetition_penalty', 'encoder_repetition_penalty', 'top_k', 'min_length', 'no_repeat_ngram_size', 'num_beams', 'penalty_alpha', 'length_penalty', 'early_stopping']])
-    shared.gradio['preset_menu_mirror'].change(load_preset_values, [shared.gradio['preset_menu_mirror']], [shared.gradio[k] for k in ['preset_menu', 'do_sample', 'temperature', 'top_p', 'typical_p', 'repetition_penalty', 'encoder_repetition_penalty', 'top_k', 'min_length', 'no_repeat_ngram_size', 'num_beams', 'penalty_alpha', 'length_penalty', 'early_stopping']])
+    shared.gradio['preset_menu'].change(load_preset_values, [shared.gradio['preset_menu']], [shared.gradio[k] for k in ['do_sample', 'temperature', 'top_p', 'typical_p', 'repetition_penalty', 'encoder_repetition_penalty', 'top_k', 'min_length', 'no_repeat_ngram_size', 'num_beams', 'penalty_alpha', 'length_penalty', 'early_stopping']])
     shared.gradio['lora_menu'].change(load_lora_wrapper, [shared.gradio['lora_menu']], [shared.gradio['lora_menu'], shared.gradio['textbox']], show_progress=True)
     shared.gradio['softprompts_menu'].change(load_soft_prompt, [shared.gradio['softprompts_menu']], [shared.gradio['softprompts_menu']], show_progress=True)
     shared.gradio['upload_softprompt'].upload(upload_soft_prompt, [shared.gradio['upload_softprompt']], [shared.gradio['softprompts_menu']])
@@ -237,8 +278,9 @@ if shared.args.lora:
 
 # Default UI settings
 default_preset = shared.settings['presets'][next((k for k in shared.settings['presets'] if re.match(k.lower(), shared.model_name.lower())), 'default')]
-default_text = shared.settings['lora_prompts'][next((k for k in shared.settings['lora_prompts'] if re.match(k.lower(), shared.lora_name.lower())), 'default')]
-if default_text == '':
+if shared.lora_name != "None":
+    default_text = shared.settings['lora_prompts'][next((k for k in shared.settings['lora_prompts'] if re.match(k.lower(), shared.lora_name.lower())), 'default')]
+else:
     default_text = shared.settings['prompts'][next((k for k in shared.settings['prompts'] if re.match(k.lower(), shared.model_name.lower())), 'default')]
 title ='Text generation web UI'
 description = '\n\n# Text generation lab\nGenerate text using Large Language Models.\n'
@@ -259,8 +301,8 @@ def create_interface():
                     shared.gradio['display'] = gr.Chatbot(value=shared.history['visible']).style(color_map=("#326efd", "#212528"))
                 shared.gradio['textbox'] = gr.Textbox(label='Input')
                 with gr.Row():
-                    shared.gradio['Stop'] = gr.Button('Stop', elem_id="stop")
                     shared.gradio['Generate'] = gr.Button('Generate')
+                    shared.gradio['Stop'] = gr.Button('Stop', elem_id="stop")
                 with gr.Row():
                     shared.gradio['Impersonate'] = gr.Button('Impersonate')
                     shared.gradio['Regenerate'] = gr.Button('Regenerate')
@@ -273,8 +315,6 @@ def create_interface():
                     shared.gradio['Clear history-confirm'] = gr.Button('Confirm', variant="stop", visible=False)
                     shared.gradio['Clear history-cancel'] = gr.Button('Cancel', visible=False)
 
-                create_model_and_preset_menus()
-
             with gr.Tab("Character", elem_id="chat-settings"):
                 shared.gradio['name1'] = gr.Textbox(value=shared.settings[f'name1{suffix}'], lines=1, label='Your name')
                 shared.gradio['name2'] = gr.Textbox(value=shared.settings[f'name2{suffix}'], lines=1, label='Bot\'s name')
@@ -327,7 +367,7 @@ def create_interface():
             gen_events.append(shared.gradio['textbox'].submit(eval(function_call), shared.input_params, shared.gradio['display'], show_progress=shared.args.no_stream))
             gen_events.append(shared.gradio['Regenerate'].click(chat.regenerate_wrapper, shared.input_params, shared.gradio['display'], show_progress=shared.args.no_stream))
             gen_events.append(shared.gradio['Impersonate'].click(chat.impersonate_wrapper, shared.input_params, shared.gradio['textbox'], show_progress=shared.args.no_stream))
-            shared.gradio['Stop'].click(chat.stop_everything_event, [], [], cancels=gen_events, queue=False)
+            shared.gradio['Stop'].click(stop_everything_event, [], [], queue=False, cancels=gen_events if shared.args.no_stream else None)
 
             shared.gradio['Copy last reply'].click(chat.send_last_reply_to_input, [], shared.gradio['textbox'], show_progress=shared.args.no_stream)
             shared.gradio['Replace last reply'].click(chat.replace_last_reply, [shared.gradio['textbox'], shared.gradio['name1'], shared.gradio['name2']], shared.gradio['display'], show_progress=shared.args.no_stream)
@@ -368,19 +408,29 @@ def create_interface():
 
         elif shared.args.notebook:
             with gr.Tab("Text generation", elem_id="main"):
-                with gr.Tab('Raw'):
-                    shared.gradio['textbox'] = gr.Textbox(value=default_text, lines=25)
-                with gr.Tab('Markdown'):
-                    shared.gradio['markdown'] = gr.Markdown()
-                with gr.Tab('HTML'):
-                    shared.gradio['html'] = gr.HTML()
-
                 with gr.Row():
-                    shared.gradio['Stop'] = gr.Button('Stop')
-                    shared.gradio['Generate'] = gr.Button('Generate')
-                shared.gradio['max_new_tokens'] = gr.Slider(minimum=shared.settings['max_new_tokens_min'], maximum=shared.settings['max_new_tokens_max'], step=1, label='max_new_tokens', value=shared.settings['max_new_tokens'])
+                    with gr.Column(scale=4):
+                        with gr.Tab('Raw'):
+                            shared.gradio['textbox'] = gr.Textbox(value=default_text, elem_id="textbox", lines=25)
+                        with gr.Tab('Markdown'):
+                            shared.gradio['markdown'] = gr.Markdown()
+                        with gr.Tab('HTML'):
+                            shared.gradio['html'] = gr.HTML()
+
+                        with gr.Row():
+                            with gr.Column():
+                                with gr.Row():
+                                    shared.gradio['Generate'] = gr.Button('Generate')
+                                    shared.gradio['Stop'] = gr.Button('Stop')
+                            with gr.Column():
+                                pass
+
+                    with gr.Column(scale=1):
+                        gr.HTML('<div style="padding-bottom: 13px"></div>')
+                        shared.gradio['max_new_tokens'] = gr.Slider(minimum=shared.settings['max_new_tokens_min'], maximum=shared.settings['max_new_tokens_max'], step=1, label='max_new_tokens', value=shared.settings['max_new_tokens'])
+
+                        create_prompt_menus()
 
-                create_model_and_preset_menus()
             with gr.Tab("Parameters", elem_id="parameters"):
                 create_settings_menus(default_preset)
 
@@ -388,7 +438,7 @@ def create_interface():
             output_params = [shared.gradio[k] for k in ['textbox', 'markdown', 'html']]
             gen_events.append(shared.gradio['Generate'].click(generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream, api_name='textgen'))
             gen_events.append(shared.gradio['textbox'].submit(generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream))
-            shared.gradio['Stop'].click(None, None, None, cancels=gen_events)
+            shared.gradio['Stop'].click(stop_everything_event, [], [], queue=False, cancels=gen_events if shared.args.no_stream else None)
             shared.gradio['interface'].load(None, None, None, _js=f"() => {{{ui.main_js}}}")
 
         else:
@@ -404,7 +454,7 @@ def create_interface():
                             with gr.Column():
                                 shared.gradio['Stop'] = gr.Button('Stop')
 
-                        create_model_and_preset_menus()
+                        create_prompt_menus()
 
                     with gr.Column():
                         with gr.Tab('Raw'):
@@ -413,6 +463,7 @@ def create_interface():
                             shared.gradio['markdown'] = gr.Markdown()
                         with gr.Tab('HTML'):
                             shared.gradio['html'] = gr.HTML()
+
             with gr.Tab("Parameters", elem_id="parameters"):
                 create_settings_menus(default_preset)
 
@@ -421,7 +472,7 @@ def create_interface():
             gen_events.append(shared.gradio['Generate'].click(generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream, api_name='textgen'))
             gen_events.append(shared.gradio['textbox'].submit(generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream))
             gen_events.append(shared.gradio['Continue'].click(generate_reply, [shared.gradio['output_textbox']] + shared.input_params[1:], output_params, show_progress=shared.args.no_stream))
-            shared.gradio['Stop'].click(None, None, None, cancels=gen_events)
+            shared.gradio['Stop'].click(stop_everything_event, [], [], queue=False, cancels=gen_events if shared.args.no_stream else None)
             shared.gradio['interface'].load(None, None, None, _js=f"() => {{{ui.main_js}}}")
 
         with gr.Tab("Interface mode", elem_id="interface-mode"):