Fix silero_tts

Useful for starting a new reply.

Dockerfile

@@ -26,7 +26,7 @@ LABEL maintainer="Your Name <your.email@example.com>"
 LABEL description="Docker image for GPTQ-for-LLaMa and Text Generation WebUI"
 
 RUN apt-get update && \
-    apt-get install --no-install-recommends -y git python3 python3-pip && \
+    apt-get install --no-install-recommends -y git python3 python3-pip make g++ && \
     rm -rf /var/lib/apt/lists/*
 
 RUN --mount=type=cache,target=/root/.cache/pip pip3 install virtualenv

README.md

@@ -1,11 +1,9 @@
 # Text generation web UI
 
-A gradio web UI for running Large Language Models like LLaMA, llama.cpp, GPT-J, OPT, and GALACTICA.
+A gradio web UI for running Large Language Models like LLaMA, llama.cpp, GPT-J, Pythia, OPT, and GALACTICA.
 
 Its goal is to become the [AUTOMATIC1111/stable-diffusion-webui](https://github.com/AUTOMATIC1111/stable-diffusion-webui) of text generation.
 
-[[Try it on Google Colab]](https://colab.research.google.com/github/oobabooga/AI-Notebooks/blob/main/Colab-TextGen-GPU.ipynb)
-
 |![Image1](https://github.com/oobabooga/screenshots/raw/main/qa.png) | ![Image2](https://github.com/oobabooga/screenshots/raw/main/cai3.png) |
 |:---:|:---:|
 |![Image3](https://github.com/oobabooga/screenshots/raw/main/gpt4chan.png) | ![Image4](https://github.com/oobabooga/screenshots/raw/main/galactica.png) |
@@ -15,7 +13,7 @@ Its goal is to become the [AUTOMATIC1111/stable-diffusion-webui](https://github.
 * Dropdown menu for switching between models
 * Notebook mode that resembles OpenAI's playground
 * Chat mode for conversation and role playing
-* Instruct mode compatible with Alpaca and Open Assistant formats **\*NEW!\***
+* Instruct mode compatible with Alpaca, Vicuna, and Open Assistant formats **\*NEW!\***
 * Nice HTML output for GPT-4chan
 * Markdown output for [GALACTICA](https://github.com/paperswithcode/galai), including LaTeX rendering
 * [Custom chat characters](https://github.com/oobabooga/text-generation-webui/wiki/Custom-chat-characters)
@@ -34,7 +32,6 @@ Its goal is to become the [AUTOMATIC1111/stable-diffusion-webui](https://github.
 * [LoRA (loading and training)](https://github.com/oobabooga/text-generation-webui/wiki/Using-LoRAs)
 * Softprompts
 * [Extensions](https://github.com/oobabooga/text-generation-webui/wiki/Extensions)
-* [Google Colab](https://github.com/oobabooga/text-generation-webui/wiki/Running-on-Colab)
 
 ## Installation
 
@@ -73,9 +70,15 @@ On Linux or WSL, it can be automatically installed with these two commands:
 curl -sL "https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh" > "Miniconda3.sh"
 bash Miniconda3.sh
 ```
-
 Source: https://educe-ubc.github.io/conda.html
 
+#### 0.1 (Ubuntu/WSL) Install build tools
+
+```
+sudo apt install build-essential
+```
+
+
 #### 1. Create a new conda environment
 
 ```
@@ -209,7 +212,7 @@ Optionally, you can use the following command-line flags:
 
 | Flag                                        | Description |
 |---------------------------------------------|-------------|
-| `--cpu`                                     | Use the CPU to generate text. |
+| `--cpu`                                     | Use the CPU to generate text. Warning: Training on CPU is extremely slow.|
 | `--auto-devices`                            | Automatically split the model across the available GPU(s) and CPU. |
 |  `--gpu-memory GPU_MEMORY [GPU_MEMORY ...]` | Maxmimum GPU memory in GiB to be allocated per GPU. Example: `--gpu-memory 10` for a single GPU, `--gpu-memory 10 5` for two GPUs. You can also set values in MiB like `--gpu-memory 3500MiB`. |
 | `--cpu-memory CPU_MEMORY`                   | Maximum CPU memory in GiB to allocate for offloaded weights. Same as above.|
@@ -218,6 +221,8 @@ Optionally, you can use the following command-line flags:
 | `--load-in-8bit`                            | Load the model with 8-bit precision.|
 | `--bf16`                                    | Load the model with bfloat16 precision. Requires NVIDIA Ampere GPU. |
 | `--no-cache`                                | Set `use_cache` to False while generating text. This reduces the VRAM usage a bit with a performance cost. |
+| `--xformers`                                | Use xformer's memory efficient attention. This should increase your tokens/s. |
+| `--sdp-attention`                           | Use torch 2.0's sdp attention. |
 
 #### llama.cpp
 
@@ -284,7 +289,9 @@ Check the [wiki](https://github.com/oobabooga/text-generation-webui/wiki/System-
 
 ## Contributing
 
-Pull requests, suggestions, and issue reports are welcome.
+Pull requests, suggestions, and issue reports are welcome. 
+
+You are also welcome to review open pull requests.
 
 Before reporting a bug, make sure that you have:
 

api-example-stream.py

@@ -12,6 +12,11 @@ import string
 
 import websockets
 
+# Note, Gradio may pick a different fn value as the definition of the Gradio app changes.
+# You can always launch the web UI and inspect the websocket stream using your browser's dev tools
+# to determine what value Gradio expects here.
+GRADIO_FN = 29
+
 
 def random_hash():
     letters = string.ascii_lowercase + string.digits
@@ -36,6 +41,10 @@ async def run(context):
         'length_penalty': 1,
         'early_stopping': False,
         'seed': -1,
+        'add_bos_token': True,
+        'truncation_length': 2048,
+        'custom_stopping_strings': [],
+        'ban_eos_token': False
     }
     payload = json.dumps([context, params])
     session = random_hash()
@@ -47,14 +56,14 @@ async def run(context):
                 case "send_hash":
                     await websocket.send(json.dumps({
                         "session_hash": session,
-                        "fn_index": 12
+                        "fn_index": GRADIO_FN
                     }))
                 case "estimation":
                     pass
                 case "send_data":
                     await websocket.send(json.dumps({
                         "session_hash": session,
-                        "fn_index": 12,
+                        "fn_index": GRADIO_FN,
                         "data": [
                             payload
                         ]

api-example.py

@@ -22,10 +22,10 @@ server = "127.0.0.1"
 params = {
     'max_new_tokens': 200,
     'do_sample': True,
-    'temperature': 0.5,
-    'top_p': 0.9,
+    'temperature': 0.72,
+    'top_p': 0.73,
     'typical_p': 1,
-    'repetition_penalty': 1.05,
+    'repetition_penalty': 1.1,
     'encoder_repetition_penalty': 1.0,
     'top_k': 0,
     'min_length': 0,
@@ -35,6 +35,10 @@ params = {
     'length_penalty': 1,
     'early_stopping': False,
     'seed': -1,
+    'add_bos_token': True,
+    'custom_stopping_strings': [],
+    'truncation_length': 2048,
+    'ban_eos_token': False,
 }
 
 # Input prompt

css/html_cai_style.css

@@ -7,11 +7,13 @@
     padding-right: 20px;
     display: flex;
     flex-direction: column-reverse;
+    word-break: break-word;
+    overflow-wrap: anywhere;
 }
 
 .message {
     display: grid;
-    grid-template-columns: 60px 1fr;
+    grid-template-columns: 60px minmax(0, 1fr);
     padding-bottom: 25px;
     font-size: 15px;
     font-family: Helvetica, Arial, sans-serif;
@@ -73,6 +75,13 @@
     display: inline !important;
 }
 
+.message-body code {
+    overflow-x: auto;
+}
+.message-body :not(pre) > code {
+    white-space: normal !important;
+}
+
 .dark .message-body p em {
     color: rgb(138, 138, 138) !important;
 }

css/html_instruct_style.css

@@ -7,6 +7,8 @@
     padding-right: 20px;
     display: flex;
     flex-direction: column-reverse;
+    word-break: break-word;
+    overflow-wrap: anywhere;
 }
 
 .message {
@@ -25,9 +27,7 @@
 .message-body {}
 
 .message-body p {
-    margin-bottom: 0 !important;
     font-size: 15px !important;
-    line-height: 1.428571429 !important;
 }
 
 .message-body li {
@@ -39,6 +39,13 @@
     display: inline !important;
 }
 
+.message-body code {
+    overflow-x: auto;
+}
+.message-body :not(pre) > code {
+    white-space: normal !important;
+}
+
 .dark .message-body p em {
     color: rgb(138, 138, 138) !important;
 }
@@ -51,15 +58,16 @@
   padding: 15px;
   border-radius: 20px;
   background-color: #0000000f;
-  margin-bottom: 17.5px;
+  margin-top: 9px !important;
+  margin-bottom: 18px !important;
 }
 
 .gradio-container .chat .user-message {
   padding: 15px;
   border-radius: 20px;
-  margin-bottom: 17.5px !important;
+  margin-bottom: 9px !important;
 }
 
 .dark .chat .assistant-message {
-  background-color: #ffffff21;
+  background-color: #374151;
 }

css/main.css

@@ -67,3 +67,13 @@ span.math.inline {
 div.svelte-15lo0d8 > *, div.svelte-15lo0d8 > .form > * {
   flex-wrap: nowrap;
 }
+
+.header_bar {
+  background-color: #f7f7f7;
+  margin-bottom: 40px;
+}
+
+.dark .header_bar {
+  border: none !important;
+  background-color: #8080802b;
+}

css/main.js

@@ -1,4 +1,4 @@
-document.getElementById("main").parentNode.childNodes[0].style = "border: none; background-color: #8080802b; margin-bottom: 40px";
+document.getElementById("main").parentNode.childNodes[0].classList.add("header_bar");
 document.getElementById("main").parentNode.style = "padding: 0; margin: 0";
 document.getElementById("main").parentNode.parentNode.parentNode.style = "padding: 0";
 

docker-compose.yml

@@ -6,7 +6,6 @@ services:
       args:
         # specify which cuda version your card supports: https://developer.nvidia.com/cuda-gpus
         TORCH_CUDA_ARCH_LIST: ${TORCH_CUDA_ARCH_LIST}
-        GPTQ_VERSION: ${GPTQ_VERSION}
         WEBUI_VERSION: ${WEBUI_VERSION}
     env_file: .env
     ports:

download-model.py

@@ -19,50 +19,6 @@ import requests
 import tqdm
 from tqdm.contrib.concurrent import thread_map
 
-parser = argparse.ArgumentParser()
-parser.add_argument('MODEL', type=str, default=None, nargs='?')
-parser.add_argument('--branch', type=str, default='main', help='Name of the Git branch to download from.')
-parser.add_argument('--threads', type=int, default=1, help='Number of files to download simultaneously.')
-parser.add_argument('--text-only', action='store_true', help='Only download text files (txt/json).')
-parser.add_argument('--output', type=str, default=None, help='The folder where the model should be saved.')
-parser.add_argument('--clean', action='store_true', help='Does not resume the previous download.')
-parser.add_argument('--check', action='store_true', help='Validates the checksums of model files.')
-args = parser.parse_args()
-
-
-def get_file(url, output_folder):
-    filename = Path(url.rsplit('/', 1)[1])
-    output_path = output_folder / filename
-    if output_path.exists() and not args.clean:
-        # Check if the file has already been downloaded completely
-        r = requests.get(url, stream=True)
-        total_size = int(r.headers.get('content-length', 0))
-        if output_path.stat().st_size >= total_size:
-            return
-        # Otherwise, resume the download from where it left off
-        headers = {'Range': f'bytes={output_path.stat().st_size}-'}
-        mode = 'ab'
-    else:
-        headers = {}
-        mode = 'wb'
-
-    r = requests.get(url, stream=True, headers=headers)
-    with open(output_path, mode) as f:
-        total_size = int(r.headers.get('content-length', 0))
-        block_size = 1024
-        with tqdm.tqdm(total=total_size, unit='iB', unit_scale=True, bar_format='{l_bar}{bar}| {n_fmt:6}/{total_fmt:6} {rate_fmt:6}') as t:
-            for data in r.iter_content(block_size):
-                t.update(len(data))
-                f.write(data)
-
-
-def sanitize_branch_name(branch_name):
-    pattern = re.compile(r"^[a-zA-Z0-9._-]+$")
-    if pattern.match(branch_name):
-        return branch_name
-    else:
-        raise ValueError("Invalid branch name. Only alphanumeric characters, period, underscore and dash are allowed.")
-
 
 def select_model_from_default_options():
     models = {
@@ -110,7 +66,20 @@ EleutherAI/pythia-1.4b-deduped
     return model, branch
 
 
-def get_download_links_from_huggingface(model, branch):
+def sanitize_model_and_branch_names(model, branch):
+    if model[-1] == '/':
+        model = model[:-1]
+    if branch is None:
+        branch = "main"
+    else:
+        pattern = re.compile(r"^[a-zA-Z0-9._-]+$")
+        if not pattern.match(branch):
+            raise ValueError("Invalid branch name. Only alphanumeric characters, period, underscore and dash are allowed.")
+
+    return model, branch
+
+
+def get_download_links_from_huggingface(model, branch, text_only=False):
     base = "https://huggingface.co"
     page = f"/api/models/{model}/tree/{branch}?cursor="
     cursor = b""
@@ -142,14 +111,14 @@ def get_download_links_from_huggingface(model, branch):
             is_tokenizer = re.match("tokenizer.*\.model", fname)
             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 any((is_pytorch, is_safetensors, is_pt, is_ggml, is_tokenizer, is_text)):
                 if 'lfs' in dict[i]:
                     sha256.append([fname, dict[i]['lfs']['oid']])
                 if is_text:
                     links.append(f"https://huggingface.co/{model}/resolve/{branch}/{fname}")
                     classifications.append('text')
                     continue
-                if not args.text_only:
+                if not text_only:
                     links.append(f"https://huggingface.co/{model}/resolve/{branch}/{fname}")
                     if is_safetensors:
                         has_safetensors = True
@@ -177,80 +146,125 @@ def get_download_links_from_huggingface(model, branch):
     return links, sha256, is_lora
 
 
-def download_files(file_list, output_folder, num_threads=8):
-    thread_map(lambda url: get_file(url, output_folder), file_list, max_workers=num_threads, disable=True)
-
-
-if __name__ == '__main__':
-    model = args.MODEL
-    branch = args.branch
-    if model is None:
-        model, branch = select_model_from_default_options()
-    else:
-        if model[-1] == '/':
-            model = model[:-1]
-            branch = args.branch
-        if branch is None:
-            branch = "main"
-        else:
-            try:
-                branch = sanitize_branch_name(branch)
-            except ValueError as err_branch:
-                print(f"Error: {err_branch}")
-                sys.exit()
-
-    links, sha256, is_lora = get_download_links_from_huggingface(model, branch)
-
-    if args.output is not None:
-        base_folder = args.output
-    else:
+def get_output_folder(model, branch, is_lora, base_folder=None):
+    if base_folder is None:
         base_folder = 'models' if not is_lora else 'loras'
 
     output_folder = f"{'_'.join(model.split('/')[-2:])}"
     if branch != 'main':
         output_folder += f'_{branch}'
     output_folder = Path(base_folder) / output_folder
+    return output_folder
+
+
+def get_single_file(url, output_folder, start_from_scratch=False):
+    filename = Path(url.rsplit('/', 1)[1])
+    output_path = output_folder / filename
+    if output_path.exists() and not start_from_scratch:
+        # Check if the file has already been downloaded completely
+        r = requests.get(url, stream=True)
+        total_size = int(r.headers.get('content-length', 0))
+        if output_path.stat().st_size >= total_size:
+            return
+        # Otherwise, resume the download from where it left off
+        headers = {'Range': f'bytes={output_path.stat().st_size}-'}
+        mode = 'ab'
+    else:
+        headers = {}
+        mode = 'wb'
+
+    r = requests.get(url, stream=True, headers=headers)
+    with open(output_path, mode) as f:
+        total_size = int(r.headers.get('content-length', 0))
+        block_size = 1024
+        with tqdm.tqdm(total=total_size, unit='iB', unit_scale=True, bar_format='{l_bar}{bar}| {n_fmt:6}/{total_fmt:6} {rate_fmt:6}') as t:
+            for data in r.iter_content(block_size):
+                t.update(len(data))
+                f.write(data)
+
+
+def start_download_threads(file_list, output_folder, start_from_scratch=False, threads=1):
+    thread_map(lambda url: get_single_file(url, output_folder, start_from_scratch=start_from_scratch), file_list, max_workers=threads, disable=True)
+
+
+def download_model_files(model, branch, links, sha256, output_folder, start_from_scratch=False, threads=1):
+    # Creating the folder and writing the metadata
+    if not output_folder.exists():
+        output_folder.mkdir()
+    with open(output_folder / 'huggingface-metadata.txt', 'w') as f:
+        f.write(f'url: https://huggingface.co/{model}\n')
+        f.write(f'branch: {branch}\n')
+        f.write(f'download date: {str(datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S"))}\n')
+        sha256_str = ''
+        for i in range(len(sha256)):
+            sha256_str += f'    {sha256[i][1]} {sha256[i][0]}\n'
+        if sha256_str != '':
+            f.write(f'sha256sum:\n{sha256_str}')
+
+    # Downloading the files
+    print(f"Downloading the model to {output_folder}")
+    start_download_threads(links, output_folder, start_from_scratch=start_from_scratch, threads=threads)
+
+
+def check_model_files(model, branch, links, sha256, output_folder):
+    # Validate the checksums
+    validated = True
+    for i in range(len(sha256)):
+        fpath = (output_folder / sha256[i][0])
+
+        if not fpath.exists():
+            print(f"The following file is missing: {fpath}")
+            validated = False
+            continue
+
+        with open(output_folder / sha256[i][0], "rb") as f:
+            bytes = f.read()
+            file_hash = hashlib.sha256(bytes).hexdigest()
+            if file_hash != sha256[i][1]:
+                print(f'Checksum failed: {sha256[i][0]}  {sha256[i][1]}')
+                validated = False
+            else:
+                print(f'Checksum validated: {sha256[i][0]}  {sha256[i][1]}')
+
+    if validated:
+        print('[+] Validated checksums of all model files!')
+    else:
+        print('[-] Invalid checksums. Rerun download-model.py with the --clean flag.')
+
+
+if __name__ == '__main__':
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument('MODEL', type=str, default=None, nargs='?')
+    parser.add_argument('--branch', type=str, default='main', help='Name of the Git branch to download from.')
+    parser.add_argument('--threads', type=int, default=1, help='Number of files to download simultaneously.')
+    parser.add_argument('--text-only', action='store_true', help='Only download text files (txt/json).')
+    parser.add_argument('--output', type=str, default=None, help='The folder where the model should be saved.')
+    parser.add_argument('--clean', action='store_true', help='Does not resume the previous download.')
+    parser.add_argument('--check', action='store_true', help='Validates the checksums of model files.')
+    args = parser.parse_args()
+
+    branch = args.branch
+    model = args.MODEL
+    if model is None:
+        model, branch = select_model_from_default_options()
+
+    # Cleaning up the model/branch names
+    try:
+        model, branch = sanitize_model_and_branch_names(model, branch)
+    except ValueError as err_branch:
+        print(f"Error: {err_branch}")
+        sys.exit()
+
+    # Getting the download links from Hugging Face
+    links, sha256, is_lora = get_download_links_from_huggingface(model, branch, text_only=args.text_only)
+
+    # Getting the output folder
+    output_folder = get_output_folder(model, branch, is_lora, base_folder=args.output)
 
     if args.check:
-        # Validate the checksums
-        validated = True
-        for i in range(len(sha256)):
-            fpath = (output_folder / sha256[i][0])
-
-            if not fpath.exists():
-                print(f"The following file is missing: {fpath}")
-                validated = False
-                continue
-
-            with open(output_folder / sha256[i][0], "rb") as f:
-                bytes = f.read()
-                file_hash = hashlib.sha256(bytes).hexdigest()
-                if file_hash != sha256[i][1]:
-                    print(f'Checksum failed: {sha256[i][0]}  {sha256[i][1]}')
-                    validated = False
-                else:
-                    print(f'Checksum validated: {sha256[i][0]}  {sha256[i][1]}')
-
-        if validated:
-            print('[+] Validated checksums of all model files!')
-        else:
-            print('[-] Invalid checksums. Rerun download-model.py with the --clean flag.')
-
+        # Check previously downloaded files
+        check_model_files(model, branch, links, sha256, output_folder)
     else:
-
-        # Creating the folder and writing the metadata
-        if not output_folder.exists():
-            output_folder.mkdir()
-        with open(output_folder / 'huggingface-metadata.txt', 'w') as f:
-            f.write(f'url: https://huggingface.co/{model}\n')
-            f.write(f'branch: {branch}\n')
-            f.write(f'download date: {str(datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S"))}\n')
-            sha256_str = ''
-            for i in range(len(sha256)):
-                sha256_str += f'    {sha256[i][1]} {sha256[i][0]}\n'
-            if sha256_str != '':
-                f.write(f'sha256sum:\n{sha256_str}')
-
-        # Downloading the files
-        print(f"Downloading the model to {output_folder}")
-        download_files(links, output_folder, args.threads)
+        # Download files
+        download_model_files(model, branch, links, sha256, output_folder, threads=args.threads)

extensions/api/script.py

@@ -57,12 +57,15 @@ class Handler(BaseHTTPRequestHandler):
                 'length_penalty': float(body.get('length_penalty', 1)),
                 'early_stopping': bool(body.get('early_stopping', False)),
                 'seed': int(body.get('seed', -1)),
+                'add_bos_token': int(body.get('add_bos_token', True)),
+                'custom_stopping_strings': body.get('custom_stopping_strings', []),
+                'truncation_length': int(body.get('truncation_length', 2048)),
+                'ban_eos_token': bool(body.get('ban_eos_token', False)),
             }
 
             generator = generate_reply(
                 prompt,
                 generate_params,
-                stopping_strings=body.get('stopping_strings', []),
             )
 
             answer = ''
@@ -78,6 +81,19 @@ class Handler(BaseHTTPRequestHandler):
                 }]
             })
             self.wfile.write(response.encode('utf-8'))
+        elif self.path == '/api/v1/token-count':
+            # Not compatible with KoboldAI api
+            self.send_response(200)
+            self.send_header('Content-Type', 'application/json')
+            self.end_headers()
+
+            tokens = encode(body['prompt'])[0]
+            response = json.dumps({
+                'results': [{
+                    'tokens': len(tokens)
+                }]
+            })
+            self.wfile.write(response.encode('utf-8'))
         else:
             self.send_error(404)
 

extensions/character_bias/script.py

@@ -1,8 +1,23 @@
 import gradio as gr
+import os
+
+# get the current directory of the script
+current_dir = os.path.dirname(os.path.abspath(__file__))
+
+# check if the bias_options.txt file exists, if not, create it
+bias_file = os.path.join(current_dir, "bias_options.txt")
+if not os.path.isfile(bias_file):
+    with open(bias_file, "w") as f:
+        f.write("*I am so happy*\n*I am so sad*\n*I am so excited*\n*I am so bored*\n*I am so angry*")
+
+# read bias options from the text file
+with open(bias_file, "r") as f:
+    bias_options = [line.strip() for line in f.readlines()]
 
 params = {
     "activate": True,
     "bias string": " *I am so happy*",
+    "use custom string": False,
 }
 
 
@@ -11,7 +26,6 @@ def input_modifier(string):
     This function is applied to your text inputs before
     they are fed into the model.
     """
-
     return string
 
 
@@ -19,7 +33,6 @@ def output_modifier(string):
     """
     This function is applied to the model outputs.
     """
-
     return string
 
 
@@ -29,9 +42,11 @@ def bot_prefix_modifier(string):
     the prefix text for the Bot and can be used to bias its
     behavior.
     """
-
     if params['activate']:
-        return f'{string} {params["bias string"].strip()} '
+        if params['use custom string']:
+            return f'{string} {params["custom string"].strip()} '
+        else:
+            return f'{string} {params["bias string"].strip()} '
     else:
         return string
 
@@ -39,8 +54,29 @@ def bot_prefix_modifier(string):
 def ui():
     # Gradio elements
     activate = gr.Checkbox(value=params['activate'], label='Activate character bias')
-    string = gr.Textbox(value=params["bias string"], label='Character bias')
+    dropdown_string = gr.Dropdown(choices=bias_options, value=params["bias string"], label='Character bias', info='To edit the options in this dropdown edit the "bias_options.txt" file')
+    use_custom_string = gr.Checkbox(value=False, label='Use custom bias textbox instead of dropdown')
+    custom_string = gr.Textbox(value="", placeholder="Enter custom bias string", label="Custom Character Bias", info='To use this textbox activate the checkbox above')
 
     # Event functions to update the parameters in the backend
-    string.change(lambda x: params.update({"bias string": x}), string, None)
+    def update_bias_string(x):
+        if x:
+            params.update({"bias string": x})
+        else:
+            params.update({"bias string": dropdown_string.get()})
+        return x
+
+    def update_custom_string(x):
+        params.update({"custom string": x})
+
+    dropdown_string.change(update_bias_string, dropdown_string, None)
+    custom_string.change(update_custom_string, custom_string, None)
     activate.change(lambda x: params.update({"activate": x}), activate, None)
+    use_custom_string.change(lambda x: params.update({"use custom string": x}), use_custom_string, None)
+
+    # Group elements together depending on the selected option
+    def bias_string_group():
+        if use_custom_string.value:
+            return gr.Group([use_custom_string, custom_string])
+        else:
+            return dropdown_string

extensions/send_pictures/script.py

@@ -25,7 +25,7 @@ def caption_image(raw_image):
 
 
 def generate_chat_picture(picture, name1, name2):
-    text = f'*{name1} sends {name2} a picture that contains the following: "{caption_image(picture)}"*'
+    text = f'*{name1} sends {name2} a picture that contains the following: “{caption_image(picture)}”*'
     # lower the resolution of sent images for the chat, otherwise the log size gets out of control quickly with all the base64 values in visible history
     picture.thumbnail((300, 300))
     buffer = BytesIO()

extensions/silero_tts/script.py

@@ -165,13 +165,13 @@ def ui():
     convert_arr = [convert_confirm, convert, convert_cancel]
     convert.click(lambda: [gr.update(visible=True), gr.update(visible=False), gr.update(visible=True)], None, convert_arr)
     convert_confirm.click(lambda: [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False)], None, convert_arr)
-    convert_confirm.click(remove_tts_from_history, [shared.gradio[k] for k in ['name1', 'name2', 'Chat mode']], shared.gradio['display'])
+    convert_confirm.click(remove_tts_from_history, [shared.gradio[k] for k in ['name1', 'name2', 'mode']], shared.gradio['display'])
     convert_confirm.click(lambda: chat.save_history(timestamp=False), [], [], show_progress=False)
     convert_cancel.click(lambda: [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False)], None, convert_arr)
 
     # Toggle message text in history
     show_text.change(lambda x: params.update({"show_text": x}), show_text, None)
-    show_text.change(toggle_text_in_history, [shared.gradio[k] for k in ['name1', 'name2', 'Chat mode']], shared.gradio['display'])
+    show_text.change(toggle_text_in_history, [shared.gradio[k] for k in ['name1', 'name2', 'mode']], shared.gradio['display'])
     show_text.change(lambda: chat.save_history(timestamp=False), [], [], show_progress=False)
 
     # Event functions to update the parameters in the backend

modules/GPTQ_loader.py

@@ -100,10 +100,10 @@ def load_quantized(model_name):
     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]
+    if len(found_pts) > 0:
+        pt_path = found_pts[-1]
+    elif len(found_safetensors) > 0:
+        pt_path = found_safetensors[-1]
     else:
         if path_to_model.name.lower().startswith('llama-7b'):
             pt_model = f'llama-7b-{shared.args.wbits}bit'
@@ -119,13 +119,14 @@ def load_quantized(model_name):
         # Try to find the .safetensors or .pt both in the model dir and in the subfolder
         for path in [Path(p + ext) for ext in ['.safetensors', '.pt'] for p in [f"{shared.args.model_dir}/{pt_model}", f"{path_to_model}/{pt_model}"]]:
             if path.exists():
-                print(f"Found {path}")
                 pt_path = path
                 break
 
     if not pt_path:
         print("Could not find the quantized model in .pt or .safetensors format, exiting...")
         exit()
+    else:
+        print(f"Found the following quantized model: {pt_path}")
 
     # qwopqwop200's offload
     if model_type == 'llama' and shared.args.pre_layer:

modules/chat.py

@@ -18,47 +18,53 @@ from modules.text_generation import (encode, generate_reply,
                                      get_max_prompt_length)
 
 
-def generate_chat_prompt(user_input, max_new_tokens, name1, name2, context, chat_prompt_size, **kwargs):
-    is_instruct = kwargs['is_instruct'] if 'is_instruct' in kwargs else False
-    end_of_turn = kwargs['end_of_turn'] if 'end_of_turn' in kwargs else ''
+def generate_chat_prompt(user_input, state, **kwargs):
     impersonate = kwargs['impersonate'] if 'impersonate' in kwargs else False
+    _continue = kwargs['_continue'] if '_continue' in kwargs else False
     also_return_rows = kwargs['also_return_rows'] if 'also_return_rows' in kwargs else False
-    rows = [f"{context.strip()}\n"]
+    is_instruct = state['mode'] == 'instruct'
+    rows = [f"{state['context'].strip()}\n"]
 
     # Finding the maximum prompt size
+    chat_prompt_size = state['chat_prompt_size']
     if shared.soft_prompt:
         chat_prompt_size -= shared.soft_prompt_tensor.shape[1]
-    max_length = min(get_max_prompt_length(max_new_tokens), chat_prompt_size)
+    max_length = min(get_max_prompt_length(state), chat_prompt_size)
 
     if is_instruct:
-        prefix1 = f"{name1}\n"
-        prefix2 = f"{name2}\n"
+        prefix1 = f"{state['name1']}\n"
+        prefix2 = f"{state['name2']}\n"
     else:
-        prefix1 = f"{name1}: "
-        prefix2 = f"{name2}: "
+        prefix1 = f"{state['name1']}: "
+        prefix2 = f"{state['name2']}: "
 
     i = len(shared.history['internal']) - 1
-    while i >= 0 and len(encode(''.join(rows), max_new_tokens)[0]) < max_length:
-        rows.insert(1, f"{prefix2}{shared.history['internal'][i][1].strip()}{end_of_turn}\n")
+    while i >= 0 and len(encode(''.join(rows))[0]) < max_length:
+        if _continue and i == len(shared.history['internal']) - 1:
+            rows.insert(1, f"{prefix2}{shared.history['internal'][i][1]}")
+        else:
+            rows.insert(1, f"{prefix2}{shared.history['internal'][i][1].strip()}{state['end_of_turn']}\n")
         string = shared.history['internal'][i][0]
         if string not in ['', '<|BEGIN-VISIBLE-CHAT|>']:
-            rows.insert(1, f"{prefix1}{string.strip()}{end_of_turn}\n")
+            rows.insert(1, f"{prefix1}{string.strip()}{state['end_of_turn']}\n")
         i -= 1
 
     if impersonate:
         rows.append(f"{prefix1.strip() if not is_instruct else prefix1}")
         limit = 2
+    elif _continue:
+        limit = 3
     else:
         # Adding the user message
         user_input = fix_newlines(user_input)
         if len(user_input) > 0:
-            rows.append(f"{prefix1}{user_input}{end_of_turn}\n")
+            rows.append(f"{prefix1}{user_input}{state['end_of_turn']}\n")
 
         # Adding the Character prefix
         rows.append(apply_extensions(f"{prefix2.strip() if not is_instruct else prefix2}", "bot_prefix"))
         limit = 3
 
-    while len(rows) > limit and len(encode(''.join(rows), max_new_tokens)[0]) >= max_length:
+    while len(rows) > limit and len(encode(''.join(rows))[0]) >= max_length:
         rows.pop(1)
     prompt = ''.join(rows)
 
@@ -68,16 +74,26 @@ def generate_chat_prompt(user_input, max_new_tokens, name1, name2, context, chat
         return prompt
 
 
-def extract_message_from_reply(reply, name1, name2, stop_at_newline):
-    next_character_found = False
+def get_stopping_strings(state):
+    if state['mode'] == 'instruct':
+        stopping_strings = [f"\n{state['name1']}", f"\n{state['name2']}"]
+    else:
+        stopping_strings = [f"\n{state['name1']}:", f"\n{state['name2']}:"]
+    stopping_strings += state['custom_stopping_strings']
+    return stopping_strings
 
-    if stop_at_newline:
+
+def extract_message_from_reply(reply, state):
+    next_character_found = False
+    stopping_strings = get_stopping_strings(state)
+
+    if state['stop_at_newline']:
         lines = reply.split('\n')
         reply = lines[0].strip()
         if len(lines) > 1:
             next_character_found = True
     else:
-        for string in [f"\n{name1}:", f"\n{name2}:"]:
+        for string in stopping_strings:
             idx = reply.find(string)
             if idx != -1:
                 reply = reply[:idx]
@@ -86,7 +102,7 @@ def extract_message_from_reply(reply, name1, name2, stop_at_newline):
         # If something like "\nYo" is generated just before "\nYou:"
         # is completed, trim it
         if not next_character_found:
-            for string in [f"\n{name1}:", f"\n{name2}:"]:
+            for string in stopping_strings:
                 for j in range(len(string) - 1, 0, -1):
                     if reply[-j:] == string[:j]:
                         reply = reply[:-j]
@@ -99,20 +115,15 @@ def extract_message_from_reply(reply, name1, name2, stop_at_newline):
     return reply, next_character_found
 
 
-def chatbot_wrapper(text, generate_state, name1, name2, context, mode, end_of_turn, regenerate=False):
-    if mode == 'instruct':
-        stopping_strings = [f"\n{name1}", f"\n{name2}"]
-    else:
-        stopping_strings = [f"\n{name1}:", f"\n{name2}:"]
+def chatbot_wrapper(text, state, regenerate=False, _continue=False):
 
     # Defining some variables
     cumulative_reply = ''
+    last_reply = [shared.history['internal'][-1][1], shared.history['visible'][-1][1]] if _continue else None
     just_started = True
-    name1_original = name1
     visible_text = custom_generate_chat_prompt = None
-    eos_token = '\n' if generate_state['stop_at_newline'] else None
-    if 'pygmalion' in shared.model_name.lower():
-        name1 = "You"
+    eos_token = '\n' if state['stop_at_newline'] else None
+    stopping_strings = get_stopping_strings(state)
 
     # Check if any extension wants to hijack this function call
     for extension, _ in extensions_module.iterator():
@@ -124,28 +135,29 @@ def chatbot_wrapper(text, generate_state, name1, name2, context, mode, end_of_tu
 
     if visible_text is None:
         visible_text = text
-    text = apply_extensions(text, "input")
+    if not _continue:
+        text = apply_extensions(text, "input")
 
     # Generating the prompt
-    kwargs = {'end_of_turn': end_of_turn, 'is_instruct': mode == 'instruct'}
+    kwargs = {'_continue': _continue}
     if custom_generate_chat_prompt is None:
-        prompt = generate_chat_prompt(text, generate_state['max_new_tokens'], name1, name2, context, generate_state['chat_prompt_size'], **kwargs)
+        prompt = generate_chat_prompt(text, state, **kwargs)
     else:
-        prompt = custom_generate_chat_prompt(text, generate_state['max_new_tokens'], name1, name2, context, generate_state['chat_prompt_size'], **kwargs)
+        prompt = custom_generate_chat_prompt(text, state, **kwargs)
 
     # Yield *Is typing...*
-    if not regenerate:
+    if not any((regenerate, _continue)):
         yield shared.history['visible'] + [[visible_text, shared.processing_message]]
 
     # Generate
-    for i in range(generate_state['chat_generation_attempts']):
+    for i in range(state['chat_generation_attempts']):
         reply = None
-        for reply in generate_reply(f"{prompt}{' ' if len(cumulative_reply) > 0 else ''}{cumulative_reply}", generate_state, eos_token=eos_token, stopping_strings=stopping_strings):
+        for reply in generate_reply(f"{prompt}{' ' if len(cumulative_reply) > 0 else ''}{cumulative_reply}", state, eos_token=eos_token, stopping_strings=stopping_strings):
             reply = cumulative_reply + reply
 
             # Extracting the reply
-            reply, next_character_found = extract_message_from_reply(reply, name1, name2, generate_state['stop_at_newline'])
-            visible_reply = re.sub("(<USER>|<user>|{{user}})", name1_original, reply)
+            reply, next_character_found = extract_message_from_reply(reply, state)
+            visible_reply = re.sub("(<USER>|<user>|{{user}})", state['name1'], reply)
             visible_reply = apply_extensions(visible_reply, "output")
 
             # We need this global variable to handle the Stop event,
@@ -154,11 +166,17 @@ def chatbot_wrapper(text, generate_state, name1, name2, context, mode, end_of_tu
                 return shared.history['visible']
             if just_started:
                 just_started = False
-                shared.history['internal'].append(['', ''])
-                shared.history['visible'].append(['', ''])
+                if not _continue:
+                    shared.history['internal'].append(['', ''])
+                    shared.history['visible'].append(['', ''])
 
-            shared.history['internal'][-1] = [text, reply]
-            shared.history['visible'][-1] = [visible_text, visible_reply]
+            if _continue:
+                sep = list(map(lambda x: ' ' if len(x) > 0 and x[-1] != ' ' else '', last_reply))
+                shared.history['internal'][-1] = [text, f'{last_reply[0]}{sep[0]}{reply}']
+                shared.history['visible'][-1] = [visible_text, f'{last_reply[1]}{sep[1]}{visible_reply}']
+            else:
+                shared.history['internal'][-1] = [text, reply]
+                shared.history['visible'][-1] = [visible_text, visible_reply]
             if not shared.args.no_stream:
                 yield shared.history['visible']
             if next_character_found:
@@ -170,28 +188,22 @@ def chatbot_wrapper(text, generate_state, name1, name2, context, mode, end_of_tu
     yield shared.history['visible']
 
 
-def impersonate_wrapper(text, generate_state, name1, name2, context, mode, end_of_turn):
-    if mode == 'instruct':
-        stopping_strings = [f"\n{name1}", f"\n{name2}"]
-    else:
-        stopping_strings = [f"\n{name1}:", f"\n{name2}:"]
+def impersonate_wrapper(text, state):
 
     # Defining some variables
     cumulative_reply = ''
-    eos_token = '\n' if generate_state['stop_at_newline'] else None
-    if 'pygmalion' in shared.model_name.lower():
-        name1 = "You"
-
-    prompt = generate_chat_prompt(text, generate_state['max_new_tokens'], name1, name2, context, generate_state['chat_prompt_size'], impersonate=True, end_of_turn=end_of_turn)
+    eos_token = '\n' if state['stop_at_newline'] else None
+    prompt = generate_chat_prompt(text, state, impersonate=True)
+    stopping_strings = get_stopping_strings(state)
 
     # Yield *Is typing...*
     yield shared.processing_message
 
-    for i in range(generate_state['chat_generation_attempts']):
+    for i in range(state['chat_generation_attempts']):
         reply = None
-        for reply in generate_reply(f"{prompt}{' ' if len(cumulative_reply) > 0 else ''}{cumulative_reply}", generate_state, eos_token=eos_token, stopping_strings=stopping_strings):
+        for reply in generate_reply(f"{prompt}{' ' if len(cumulative_reply) > 0 else ''}{cumulative_reply}", state, eos_token=eos_token, stopping_strings=stopping_strings):
             reply = cumulative_reply + reply
-            reply, next_character_found = extract_message_from_reply(reply, name1, name2, generate_state['stop_at_newline'])
+            reply, next_character_found = extract_message_from_reply(reply, state)
             yield reply
             if next_character_found:
                 break
@@ -202,22 +214,32 @@ def impersonate_wrapper(text, generate_state, name1, name2, context, mode, end_o
     yield reply
 
 
-def cai_chatbot_wrapper(text, generate_state, name1, name2, context, mode, end_of_turn):
-    for history in chatbot_wrapper(text, generate_state, name1, name2, context, mode, end_of_turn):
-        yield chat_html_wrapper(history, name1, name2, mode)
+def cai_chatbot_wrapper(text, state):
+    for history in chatbot_wrapper(text, state):
+        yield chat_html_wrapper(history, state['name1'], state['name2'], state['mode'])
 
 
-def regenerate_wrapper(text, generate_state, name1, name2, context, mode, end_of_turn):
+def regenerate_wrapper(text, state):
     if (len(shared.history['visible']) == 1 and not shared.history['visible'][0][0]) or len(shared.history['internal']) == 0:
-        yield chat_html_wrapper(shared.history['visible'], name1, name2, mode)
+        yield chat_html_wrapper(shared.history['visible'], state['name1'], state['name2'], state['mode'])
     else:
         last_visible = shared.history['visible'].pop()
         last_internal = shared.history['internal'].pop()
         # Yield '*Is typing...*'
-        yield chat_html_wrapper(shared.history['visible'] + [[last_visible[0], shared.processing_message]], name1, name2, mode)
-        for history in chatbot_wrapper(last_internal[0], generate_state, name1, name2, context, mode, end_of_turn, regenerate=True):
+        yield chat_html_wrapper(shared.history['visible'] + [[last_visible[0], shared.processing_message]], state['name1'], state['name2'], state['mode'])
+        for history in chatbot_wrapper(last_internal[0], state, regenerate=True):
             shared.history['visible'][-1] = [last_visible[0], history[-1][1]]
-            yield chat_html_wrapper(shared.history['visible'], name1, name2, mode)
+            yield chat_html_wrapper(shared.history['visible'], state['name1'], state['name2'], state['mode'])
+
+
+def continue_wrapper(text, state):
+    if (len(shared.history['visible']) == 1 and not shared.history['visible'][0][0]) or len(shared.history['internal']) == 0:
+        yield chat_html_wrapper(shared.history['visible'], state['name1'], state['name2'], state['mode'])
+    else:
+        # Yield ' ...'
+        yield chat_html_wrapper(shared.history['visible'][:-1] + [[shared.history['visible'][-1][0], shared.history['visible'][-1][1] + ' ...']], state['name1'], state['name2'], state['mode'])
+        for history in chatbot_wrapper(shared.history['internal'][-1][0], state, _continue=True):
+            yield chat_html_wrapper(shared.history['visible'], state['name1'], state['name2'], state['mode'])
 
 
 def remove_last_message(name1, name2, mode):
@@ -245,6 +267,21 @@ def replace_last_reply(text, name1, name2, mode):
     return chat_html_wrapper(shared.history['visible'], name1, name2, mode)
 
 
+def send_dummy_message(text, name1, name2, mode):
+    shared.history['visible'].append([text, ''])
+    shared.history['internal'].append([apply_extensions(text, "input"), ''])
+    return chat_html_wrapper(shared.history['visible'], name1, name2, mode)
+
+
+def send_dummy_reply(text, name1, name2, mode):
+    if len(shared.history['visible']) > 0 and not shared.history['visible'][-1][1] == '':
+        shared.history['visible'].append(['', ''])
+        shared.history['internal'].append(['', ''])
+    shared.history['visible'][-1][1] = text
+    shared.history['internal'][-1][1] = apply_extensions(text, "input")
+    return chat_html_wrapper(shared.history['visible'], name1, name2, mode)
+
+
 def clear_html():
     return chat_html_wrapper([], "", "")
 
@@ -257,6 +294,9 @@ def clear_chat_log(name1, name2, greeting, mode):
         shared.history['internal'] += [['<|BEGIN-VISIBLE-CHAT|>', greeting]]
         shared.history['visible'] += [['', apply_extensions(greeting, "output")]]
 
+    # Save cleared logs
+    save_history(mode)
+
     return chat_html_wrapper(shared.history['visible'], name1, name2, mode)
 
 
@@ -301,15 +341,23 @@ def tokenize_dialogue(dialogue, name1, name2, mode):
     return history
 
 
-def save_history(timestamp=True):
-    if timestamp:
-        fname = f"{shared.character}_{datetime.now().strftime('%Y%m%d-%H%M%S')}.json"
+def save_history(mode, timestamp=False):
+    # Instruct mode histories should not be saved as if
+    # Alpaca or Vicuna were characters
+    if mode == 'instruct':
+        if not timestamp:
+            return
+        fname = f"Instruct_{datetime.now().strftime('%Y%m%d-%H%M%S')}.json"
     else:
-        fname = f"{shared.character}_persistent.json"
+        if timestamp:
+            fname = f"{shared.character}_{datetime.now().strftime('%Y%m%d-%H%M%S')}.json"
+        else:
+            fname = f"{shared.character}_persistent.json"
     if not Path('logs').exists():
         Path('logs').mkdir()
     with open(Path(f'logs/{fname}'), 'w', encoding='utf-8') as f:
         f.write(json.dumps({'data': shared.history['internal'], 'data_visible': shared.history['visible']}, indent=2))
+
     return Path(f'logs/{fname}')
 
 
@@ -323,16 +371,6 @@ def load_history(file, name1, name2):
                 shared.history['visible'] = j['data_visible']
             else:
                 shared.history['visible'] = copy.deepcopy(shared.history['internal'])
-        # Compatibility with Pygmalion AI's official web UI
-        elif 'chat' in j:
-            shared.history['internal'] = [':'.join(x.split(':')[1:]).strip() for x in j['chat']]
-            if len(j['chat']) > 0 and j['chat'][0].startswith(f'{name2}:'):
-                shared.history['internal'] = [['<|BEGIN-VISIBLE-CHAT|>', shared.history['internal'][0]]] + [[shared.history['internal'][i], shared.history['internal'][i + 1]] for i in range(1, len(shared.history['internal']) - 1, 2)]
-                shared.history['visible'] = copy.deepcopy(shared.history['internal'])
-                shared.history['visible'][0][0] = ''
-            else:
-                shared.history['internal'] = [[shared.history['internal'][i], shared.history['internal'][i + 1]] for i in range(0, len(shared.history['internal']) - 1, 2)]
-                shared.history['visible'] = copy.deepcopy(shared.history['internal'])
     except:
         shared.history['internal'] = tokenize_dialogue(file, name1, name2)
         shared.history['visible'] = copy.deepcopy(shared.history['internal'])
@@ -368,8 +406,6 @@ def generate_pfp_cache(character):
 
 def load_character(character, name1, name2, mode):
     shared.character = character
-    shared.history['internal'] = []
-    shared.history['visible'] = []
     context = greeting = end_of_turn = ""
     greeting_field = 'greeting'
     picture = None
@@ -414,13 +450,22 @@ def load_character(character, name1, name2, mode):
         greeting = shared.settings['greeting']
         end_of_turn = shared.settings['end_of_turn']
 
-    if Path(f'logs/{shared.character}_persistent.json').exists():
-        load_history(open(Path(f'logs/{shared.character}_persistent.json'), 'rb').read(), name1, name2)
-    elif greeting != "":
-        shared.history['internal'] += [['<|BEGIN-VISIBLE-CHAT|>', greeting]]
-        shared.history['visible'] += [['', apply_extensions(greeting, "output")]]
+    if mode != 'instruct':
+        shared.history['internal'] = []
+        shared.history['visible'] = []
 
-    return name1, name2, picture, greeting, context, end_of_turn, chat_html_wrapper(shared.history['visible'], name1, name2, mode, reset_cache=True)
+        if Path(f'logs/{shared.character}_persistent.json').exists():
+            load_history(open(Path(f'logs/{shared.character}_persistent.json'), 'rb').read(), name1, name2)
+        else:
+            # Insert greeting if it exists
+            if greeting != "":
+                shared.history['internal'] += [['<|BEGIN-VISIBLE-CHAT|>', greeting]]
+                shared.history['visible'] += [['', apply_extensions(greeting, "output")]]
+
+            # Create .json log files since they don't already exist
+            save_history(mode)
+
+    return name1, name2, picture, greeting, context, end_of_turn, chat_html_wrapper(shared.history['visible'], name1, name2, mode)
 
 
 def load_default_history(name1, name2):
@@ -468,4 +513,4 @@ def upload_your_profile_picture(img, name1, name2, mode):
         img.save(Path('cache/pfp_me.png'))
         print('Profile picture saved to "cache/pfp_me.png"')
 
-    return chat_html_wrapper(shared.history['visible'], name1, name2, mode, reset_cache=True)
+    return chat_html_wrapper(shared.history['visible'], name1, name2, mode, reset_cache=True)

modules/html_generator.py

@@ -164,10 +164,9 @@ def generate_instruct_html(history):
 def generate_cai_chat_html(history, name1, name2, reset_cache=False):
     output = f'<style>{cai_css}</style><div class="chat" id="chat">'
 
-    # The time.time() is to prevent the brower from caching the image
-    suffix = f"?{time.time()}" if reset_cache else f"?{name2}"
-    img_bot = f'<img src="file/cache/pfp_character.png{suffix}">' if Path("cache/pfp_character.png").exists() else ''
-    img_me = f'<img src="file/cache/pfp_me.png{suffix}">' if Path("cache/pfp_me.png").exists() else ''
+    # We use ?name2 and ?time.time() to force the browser to reset caches
+    img_bot = f'<img src="file/cache/pfp_character.png?{name2}">' if Path("cache/pfp_character.png").exists() else ''
+    img_me = f'<img src="file/cache/pfp_me.png?{time.time() if reset_cache else ""}">' if Path("cache/pfp_me.png").exists() else ''
 
     for i, _row in enumerate(history[::-1]):
         row = [convert_to_markdown(entry) for entry in _row]

modules/llama_attn_hijack.py

@@ -0,0 +1,176 @@
+import math
+import sys
+import torch
+import torch.nn as nn
+import transformers.models.llama.modeling_llama
+
+from typing import Optional
+from typing import Tuple
+
+import modules.shared as shared
+
+
+if shared.args.xformers:
+    try:
+        import xformers.ops
+    except Exception:
+        print("🔴 xformers not found! Please install it before trying to use it.", file=sys.stderr)
+
+
+def hijack_llama_attention():
+    if shared.args.xformers:
+        transformers.models.llama.modeling_llama.LlamaAttention.forward = xformers_forward
+        print("Replaced attention with xformers_attention")
+    elif shared.args.sdp_attention:
+        transformers.models.llama.modeling_llama.LlamaAttention.forward = sdp_attention_forward
+        print("Replaced attention with sdp_attention")
+
+
+def xformers_forward(
+    self,
+    hidden_states: torch.Tensor,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_value: Optional[Tuple[torch.Tensor]] = None,
+    output_attentions: bool = False,
+    use_cache: bool = False,
+) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+    bsz, q_len, _ = hidden_states.size()
+
+    query_states = self.q_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+    key_states = self.k_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+    value_states = self.v_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+
+    kv_seq_len = key_states.shape[-2]
+    if past_key_value is not None:
+        kv_seq_len += past_key_value[0].shape[-2]
+    cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+    query_states, key_states = transformers.models.llama.modeling_llama.apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+    # [bsz, nh, t, hd]
+
+    if past_key_value is not None:
+        # reuse k, v, self_attention
+        key_states = torch.cat([past_key_value[0], key_states], dim=2)
+        value_states = torch.cat([past_key_value[1], value_states], dim=2)
+
+    past_key_value = (key_states, value_states) if use_cache else None
+
+    #We only apply xformers optimizations if we don't need to output the whole attention matrix
+    if not output_attentions:
+        dtype = query_states.dtype
+
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+        
+        #This is a nasty hack. We know attention_mask in transformers is either LowerTriangular or all Zeros.
+        #We therefore check if one element in the upper triangular portion is zero. If it is, then the mask is all zeros.
+        if attention_mask is None or attention_mask[0, 0, 0, 1] == 0:
+            # input and output should be of form (bsz, q_len, num_heads, head_dim)
+            attn_output = xformers.ops.memory_efficient_attention(query_states, key_states, value_states, attn_bias=None)
+        else:
+            # input and output should be of form (bsz, q_len, num_heads, head_dim)
+            attn_output = xformers.ops.memory_efficient_attention(query_states, key_states, value_states, attn_bias=xformers.ops.LowerTriangularMask())
+        attn_weights = None
+    else:
+        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz * self.num_heads, q_len, kv_seq_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+            attn_weights = attn_weights + attention_mask
+            attn_weights = torch.max(attn_weights, torch.tensor(torch.finfo(attn_weights.dtype).min))
+
+        # upcast attention to fp32
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        attn_output = attn_output.transpose(1, 2)
+
+    attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+
+    attn_output = self.o_proj(attn_output)
+
+    return attn_output, attn_weights, past_key_value
+
+
+def sdp_attention_forward(
+    self,
+    hidden_states: torch.Tensor,
+    attention_mask: Optional[torch.Tensor] = None,
+    position_ids: Optional[torch.LongTensor] = None,
+    past_key_value: Optional[Tuple[torch.Tensor]] = None,
+    output_attentions: bool = False,
+    use_cache: bool = False,
+) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+    bsz, q_len, _ = hidden_states.size()
+
+    query_states = self.q_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+    key_states = self.k_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+    value_states = self.v_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+
+    kv_seq_len = key_states.shape[-2]
+    if past_key_value is not None:
+        kv_seq_len += past_key_value[0].shape[-2]
+    cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+    query_states, key_states = transformers.models.llama.modeling_llama.apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
+    # [bsz, nh, t, hd]
+
+    if past_key_value is not None:
+        # reuse k, v, self_attention
+        key_states = torch.cat([past_key_value[0], key_states], dim=2)
+        value_states = torch.cat([past_key_value[1], value_states], dim=2)
+
+    past_key_value = (key_states, value_states) if use_cache else None
+
+    #We only apply sdp attention if we don't need to output the whole attention matrix
+    if not output_attentions:
+        attn_output = torch.nn.functional.scaled_dot_product_attention(query_states, key_states, value_states, attn_mask=attention_mask, is_causal=False)
+        attn_weights = None
+    else:
+        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz * self.num_heads, q_len, kv_seq_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+            attn_weights = attn_weights + attention_mask
+            attn_weights = torch.max(attn_weights, torch.tensor(torch.finfo(attn_weights.dtype).min))
+
+        # upcast attention to fp32
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+    attn_output = attn_output.transpose(1, 2)
+    attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+
+    attn_output = self.o_proj(attn_output)
+
+    return attn_output, attn_weights, past_key_value

modules/models.py

@@ -14,6 +14,7 @@ from transformers import (AutoConfig, AutoModelForCausalLM, AutoTokenizer,
                           BitsAndBytesConfig, LlamaTokenizer)
 
 import modules.shared as shared
+from modules import llama_attn_hijack
 
 transformers.logging.set_verbosity_error()
 
@@ -169,14 +170,25 @@ def load_model(model_name):
 
         model = AutoModelForCausalLM.from_pretrained(checkpoint, **params)
 
+    # Hijack attention with xformers
+    if any((shared.args.xformers, shared.args.sdp_attention)):
+        llama_attn_hijack.hijack_llama_attention()
+
     # Loading the tokenizer
     if any((k in shared.model_name.lower() for k in ['gpt4chan', 'gpt-4chan'])) and Path(f"{shared.args.model_dir}/gpt-j-6B/").exists():
         tokenizer = AutoTokenizer.from_pretrained(Path(f"{shared.args.model_dir}/gpt-j-6B/"))
     elif type(model) is transformers.LlamaForCausalLM:
         tokenizer = LlamaTokenizer.from_pretrained(Path(f"{shared.args.model_dir}/{shared.model_name}/"), clean_up_tokenization_spaces=True)
+        # Leaving this here until the LLaMA tokenizer gets figured out.
+        # For some people this fixes things, for others it causes an error.
+        try:
+            tokenizer.eos_token_id = 2
+            tokenizer.bos_token_id = 1
+            tokenizer.pad_token_id = 0
+        except:
+            pass
     else:
         tokenizer = AutoTokenizer.from_pretrained(Path(f"{shared.args.model_dir}/{shared.model_name}/"))
-    tokenizer.truncation_side = 'left'
 
     print(f"Loaded the model in {(time.time()-t0):.2f} seconds.")
     return model, tokenizer

modules/shared.py

@@ -34,7 +34,13 @@ settings = {
     'context': 'This is a conversation with your Assistant. The Assistant is very helpful and is eager to chat with you and answer your questions.',
     'greeting': 'Hello there!',
     'end_of_turn': '',
+    'custom_stopping_strings': '',
     'stop_at_newline': False,
+    'add_bos_token': True,
+    'ban_eos_token': False,
+    'truncation_length': 2048,
+    'truncation_length_min': 0,
+    'truncation_length_max': 4096,
     'chat_prompt_size': 2048,
     'chat_prompt_size_min': 0,
     'chat_prompt_size_max': 2048,
@@ -44,7 +50,7 @@ settings = {
     'default_extensions': [],
     'chat_default_extensions': ["gallery"],
     'presets': {
-        'default': 'NovelAI-Sphinx Moth',
+        'default': 'Default',
         '.*(alpaca|llama)': "LLaMA-Precise",
         '.*pygmalion': 'NovelAI-Storywriter',
         '.*RWKV': 'Naive',
@@ -89,7 +95,7 @@ parser.add_argument('--extensions', type=str, nargs="+", help='The list of exten
 parser.add_argument('--verbose', action='store_true', help='Print the prompts to the terminal.')
 
 # Accelerate/transformers
-parser.add_argument('--cpu', action='store_true', help='Use the CPU to generate text.')
+parser.add_argument('--cpu', action='store_true', help='Use the CPU to generate text. Warning: Training on CPU is extremely slow.')
 parser.add_argument('--auto-devices', action='store_true', help='Automatically split the model across the available GPU(s) and CPU.')
 parser.add_argument('--gpu-memory', type=str, nargs="+", help='Maxmimum GPU memory in GiB to be allocated per GPU. Example: --gpu-memory 10 for a single GPU, --gpu-memory 10 5 for two GPUs. You can also set values in MiB like --gpu-memory 3500MiB.')
 parser.add_argument('--cpu-memory', type=str, help='Maximum CPU memory in GiB to allocate for offloaded weights. Same as above.')
@@ -98,6 +104,8 @@ parser.add_argument('--disk-cache-dir', type=str, default="cache", help='Directo
 parser.add_argument('--load-in-8bit', action='store_true', help='Load the model with 8-bit precision.')
 parser.add_argument('--bf16', action='store_true', help='Load the model with bfloat16 precision. Requires NVIDIA Ampere GPU.')
 parser.add_argument('--no-cache', action='store_true', help='Set use_cache to False while generating text. This reduces the VRAM usage a bit at a performance cost.')
+parser.add_argument('--xformers', action='store_true', help="Use xformer's memory efficient attention. This should increase your tokens/s.")
+parser.add_argument('--sdp-attention', action='store_true', help="Use torch 2.0's sdp attention.")
 
 # llama.cpp
 parser.add_argument('--threads', type=int, default=0, help='Number of threads to use in llama.cpp.')

modules/text_generation.py

@@ -1,3 +1,4 @@
+import random
 import re
 import time
 import traceback
@@ -14,35 +15,45 @@ from modules.html_generator import generate_4chan_html, generate_basic_html
 from modules.models import clear_torch_cache, local_rank
 
 
-def get_max_prompt_length(tokens):
-    max_length = 2048 - tokens
+def get_max_prompt_length(state):
+    max_length = state['truncation_length'] - state['max_new_tokens']
     if shared.soft_prompt:
         max_length -= shared.soft_prompt_tensor.shape[1]
     return max_length
 
 
-def encode(prompt, tokens_to_generate=0, add_special_tokens=True):
+def encode(prompt, add_special_tokens=True, add_bos_token=True, truncation_length=None):
     if any((shared.is_RWKV, shared.is_llamacpp)):
         input_ids = shared.tokenizer.encode(str(prompt))
         input_ids = np.array(input_ids).reshape(1, len(input_ids))
         return input_ids
     else:
-        input_ids = shared.tokenizer.encode(str(prompt), return_tensors='pt', truncation=True, max_length=get_max_prompt_length(tokens_to_generate), add_special_tokens=add_special_tokens)
+        input_ids = shared.tokenizer.encode(str(prompt), return_tensors='pt', add_special_tokens=add_special_tokens)
 
+        # This is a hack for making replies more creative.
+        if not add_bos_token and input_ids[0][0] == shared.tokenizer.bos_token_id:
+            input_ids = input_ids[:, 1:]
+
+        # Llama adds this extra token when the first character is '\n', and this
+        # compromises the stopping criteria, so we just remove it
         if type(shared.tokenizer) is transformers.LlamaTokenizer and input_ids[0][0] == 29871:
             input_ids = input_ids[:, 1:]
 
-        if shared.args.cpu:
-            return input_ids
-        elif shared.args.flexgen:
-            return input_ids.numpy()
-        elif shared.args.deepspeed:
-            return input_ids.to(device=local_rank)
-        elif torch.has_mps:
-            device = torch.device('mps')
-            return input_ids.to(device)
-        else:
-            return input_ids.cuda()
+    # Handling truncation
+    if truncation_length is not None:
+        input_ids = input_ids[:, -truncation_length:]
+
+    if any((shared.is_RWKV, shared.is_llamacpp, shared.args.cpu)):
+        return input_ids
+    elif shared.args.flexgen:
+        return input_ids.numpy()
+    elif shared.args.deepspeed:
+        return input_ids.to(device=local_rank)
+    elif torch.has_mps:
+        device = torch.device('mps')
+        return input_ids.to(device)
+    else:
+        return input_ids.cuda()
 
 
 def decode(output_ids):
@@ -62,9 +73,8 @@ def generate_softprompt_input_tensors(input_ids):
     # filler_input_ids += shared.model.config.bos_token_id # setting dummy input_ids to bos tokens
     return inputs_embeds, filler_input_ids
 
+
 # Removes empty replies from gpt4chan outputs
-
-
 def fix_gpt4chan(s):
     for i in range(10):
         s = re.sub("--- [0-9]*\n>>[0-9]*\n---", "---", s)
@@ -72,9 +82,8 @@ def fix_gpt4chan(s):
         s = re.sub("--- [0-9]*\n\n\n---", "---", s)
     return s
 
+
 # Fix the LaTeX equations in galactica
-
-
 def fix_galactica(s):
     s = s.replace(r'\[', r'$')
     s = s.replace(r'\]', r'$')
@@ -101,19 +110,22 @@ def formatted_outputs(reply, model_name):
 
 
 def set_manual_seed(seed):
-    if seed != -1:
-        torch.manual_seed(seed)
-        if torch.cuda.is_available():
-            torch.cuda.manual_seed_all(seed)
+    seed = int(seed)
+    if seed == -1:
+        seed = random.randint(1, 2**31)
+    torch.manual_seed(seed)
+    if torch.cuda.is_available():
+        torch.cuda.manual_seed_all(seed)
+    return seed
 
 
 def stop_everything_event():
     shared.stop_everything = True
 
 
-def generate_reply(question, generate_state, eos_token=None, stopping_strings=[]):
+def generate_reply(question, state, eos_token=None, stopping_strings=[]):
     clear_torch_cache()
-    set_manual_seed(generate_state['seed'])
+    seed = set_manual_seed(state['seed'])
     shared.stop_everything = False
     generate_params = {}
     t0 = time.time()
@@ -121,15 +133,17 @@ def generate_reply(question, generate_state, eos_token=None, stopping_strings=[]
     original_question = question
     if not shared.is_chat():
         question = apply_extensions(question, 'input')
-    if shared.args.verbose:
-        print(f'\n\n{question}\n--------------------\n')
 
     # These models are not part of Hugging Face, so we handle them
     # separately and terminate the function call earlier
     if any((shared.is_RWKV, shared.is_llamacpp)):
+
+        if shared.args.verbose:
+            print(f'\n\n{question}\n--------------------\n')
+
         for k in ['temperature', 'top_p', 'top_k', 'repetition_penalty']:
-            generate_params[k] = generate_state[k]
-        generate_params['token_count'] = generate_state['max_new_tokens']
+            generate_params[k] = state[k]
+        generate_params['token_count'] = state['max_new_tokens']
         try:
             if shared.args.no_stream:
                 reply = shared.model.generate(context=question, **generate_params)
@@ -155,33 +169,40 @@ def generate_reply(question, generate_state, eos_token=None, stopping_strings=[]
             t1 = time.time()
             original_tokens = len(encode(original_question)[0])
             new_tokens = len(encode(output)[0]) - original_tokens
-            print(f'Output generated in {(t1-t0):.2f} seconds ({new_tokens/(t1-t0):.2f} tokens/s, {new_tokens} tokens, context {original_tokens})')
+            print(f'Output generated in {(t1-t0):.2f} seconds ({new_tokens/(t1-t0):.2f} tokens/s, {new_tokens} tokens, context {original_tokens}, seed {seed})')
             return
 
-    input_ids = encode(question, generate_state['max_new_tokens'])
+    input_ids = encode(question, add_bos_token=state['add_bos_token'], truncation_length=get_max_prompt_length(state))
     original_input_ids = input_ids
     output = input_ids[0]
 
+    if shared.args.verbose:
+        print(f'\n\n{decode(input_ids[0])}\n--------------------\n')
+
     cuda = not any((shared.args.cpu, shared.args.deepspeed, shared.args.flexgen))
     eos_token_ids = [shared.tokenizer.eos_token_id] if shared.tokenizer.eos_token_id is not None else []
     if eos_token is not None:
         eos_token_ids.append(int(encode(eos_token)[0][-1]))
+
+    # Handling the stopping strings
     stopping_criteria_list = transformers.StoppingCriteriaList()
-    if type(stopping_strings) is list and len(stopping_strings) > 0:
-        t = [encode(string, 0, add_special_tokens=False) for string in stopping_strings]
-        stopping_criteria_list.append(_SentinelTokenStoppingCriteria(sentinel_token_ids=t, starting_idx=len(input_ids[0])))
+    for st in [stopping_strings, state['custom_stopping_strings']]:
+        if type(st) is list and len(st) > 0:
+            sentinel_token_ids = [encode(string, add_special_tokens=False) for string in st]
+            stopping_criteria_list.append(_SentinelTokenStoppingCriteria(sentinel_token_ids=sentinel_token_ids, starting_idx=len(input_ids[0])))
+            break
 
     if not shared.args.flexgen:
         for k in ['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']:
-            generate_params[k] = generate_state[k]
+            generate_params[k] = state[k]
         generate_params['eos_token_id'] = eos_token_ids
         generate_params['stopping_criteria'] = stopping_criteria_list
-        if shared.args.no_stream:
-            generate_params['min_length'] = 0
+        if state['ban_eos_token']:
+            generate_params['suppress_tokens'] = [shared.tokenizer.eos_token_id]
     else:
         for k in ['max_new_tokens', 'do_sample', 'temperature']:
-            generate_params[k] = generate_state[k]
-        generate_params['stop'] = generate_state['eos_token_ids'][-1]
+            generate_params[k] = state[k]
+        generate_params['stop'] = state['eos_token_ids'][-1]
         if not shared.args.no_stream:
             generate_params['max_new_tokens'] = 8
 
@@ -244,7 +265,7 @@ def generate_reply(question, generate_state, eos_token=None, stopping_strings=[]
 
         # Stream the output naively for FlexGen since it doesn't support 'stopping_criteria'
         else:
-            for i in range(generate_state['max_new_tokens'] // 8 + 1):
+            for i in range(state['max_new_tokens'] // 8 + 1):
                 clear_torch_cache()
                 with torch.no_grad():
                     output = shared.model.generate(**generate_params)[0]
@@ -276,5 +297,5 @@ def generate_reply(question, generate_state, eos_token=None, stopping_strings=[]
         t1 = time.time()
         original_tokens = len(original_input_ids[0])
         new_tokens = len(output) - original_tokens
-        print(f'Output generated in {(t1-t0):.2f} seconds ({new_tokens/(t1-t0):.2f} tokens/s, {new_tokens} tokens, context {original_tokens})')
+        print(f'Output generated in {(t1-t0):.2f} seconds ({new_tokens/(t1-t0):.2f} tokens/s, {new_tokens} tokens, context {original_tokens}, seed {seed})')
         return

modules/training.py

@@ -238,7 +238,7 @@ def do_train(lora_name: str, micro_batch_size: int, batch_size: int, epochs: int
             warmup_steps=100,
             num_train_epochs=epochs,
             learning_rate=actual_lr,
-            fp16=True,
+            fp16=False if shared.args.cpu else True,
             logging_steps=20,
             evaluation_strategy="steps" if eval_data is not None else "no",
             save_strategy="steps",
@@ -248,7 +248,8 @@ def do_train(lora_name: str, micro_batch_size: int, batch_size: int, epochs: int
             save_total_limit=3,
             load_best_model_at_end=True if eval_data is not None else False,
             # TODO: Enable multi-device support
-            ddp_find_unused_parameters=None
+            ddp_find_unused_parameters=None,
+            no_cuda=shared.args.cpu
         ),
         data_collator=transformers.DataCollatorForLanguageModeling(shared.tokenizer, mlm=False),
         callbacks=list([Callbacks()])

presets/Default.txt

@@ -1,7 +1,6 @@
 do_sample=True
-top_p=0.5
-top_k=40
-temperature=0.7
+top_p=0.95
+top_k=50
+temperature=1
 repetition_penalty=1.2
 typical_p=1.0
-early_stopping=False

requirements.txt

@@ -1,10 +1,10 @@
 accelerate==0.18.0
-bitsandbytes==0.37.2
 datasets
 flexgen==0.1.7
 gradio==3.24.1
 markdown
 numpy
+Pillow>=9.5.0
 peft==0.2.0
 requests
 rwkv==0.7.3
@@ -13,3 +13,6 @@ sentencepiece
 pyyaml
 tqdm
 git+https://github.com/huggingface/transformers
+bitsandbytes==0.37.2; platform_system != "Windows"
+llama-cpp-python==0.1.30; platform_system != "Windows"
+https://github.com/abetlen/llama-cpp-python/releases/download/v0.1.30/llama_cpp_python-0.1.30-cp310-cp310-win_amd64.whl; platform_system == "Windows"

server.py

@@ -2,11 +2,14 @@ import os
 
 os.environ['GRADIO_ANALYTICS_ENABLED'] = 'False'
 
+import importlib
 import io
 import json
+import os
 import re
 import sys
 import time
+import traceback
 import zipfile
 from datetime import datetime
 from pathlib import Path
@@ -21,6 +24,7 @@ from modules.LoRA import add_lora_to_model
 from modules.models import load_model, load_soft_prompt, unload_model
 from modules.text_generation import generate_reply, stop_everything_event
 
+
 # Loading custom settings
 settings_file = None
 if shared.args.settings is not None and Path(shared.args.settings).exists():
@@ -172,6 +176,34 @@ def create_prompt_menus():
     shared.gradio['save_prompt'].click(save_prompt, [shared.gradio['textbox']], [shared.gradio['status']], show_progress=False)
 
 
+def download_model_wrapper(repo_id):
+    try:
+        downloader = importlib.import_module("download-model")
+
+        model = repo_id
+        branch = "main"
+        check = False
+
+        yield ("Cleaning up the model/branch names")
+        model, branch = downloader.sanitize_model_and_branch_names(model, branch)
+
+        yield ("Getting the download links from Hugging Face")
+        links, sha256, is_lora = downloader.get_download_links_from_huggingface(model, branch, text_only=False)
+
+        yield ("Getting the output folder")
+        output_folder = downloader.get_output_folder(model, branch, is_lora)
+
+        if check:
+            yield ("Checking previously downloaded files")
+            downloader.check_model_files(model, branch, links, sha256, output_folder)
+        else:
+            yield (f"Downloading files to {output_folder}")
+            downloader.download_model_files(model, branch, links, sha256, output_folder, threads=1)
+            yield ("Done!")
+    except:
+        yield traceback.format_exc()
+
+
 def create_model_menus():
     with gr.Row():
         with gr.Column():
@@ -182,16 +214,26 @@ def create_model_menus():
             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')
+    with gr.Row():
+        with gr.Column():
+            with gr.Row():
+                with gr.Column():
+                    shared.gradio['custom_model_menu'] = gr.Textbox(label="Download custom model or LoRA",
+                                                                    info="Enter Hugging Face username/model path, e.g: facebook/galactica-125m")
+                with gr.Column():
+                    shared.gradio['download_button'] = gr.Button("Download")
+                    shared.gradio['download_status'] = gr.Markdown()
+        with gr.Column():
+            pass
 
     shared.gradio['model_menu'].change(load_model_wrapper, shared.gradio['model_menu'], shared.gradio['model_menu'], show_progress=True)
     shared.gradio['lora_menu'].change(load_lora_wrapper, shared.gradio['lora_menu'], shared.gradio['lora_menu'], show_progress=True)
+    shared.gradio['download_button'].click(download_model_wrapper, shared.gradio['custom_model_menu'], shared.gradio['download_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)
-    for k in ['max_new_tokens', 'seed', 'stop_at_newline', 'chat_prompt_size', 'chat_generation_attempts']:
-        generate_params[k] = shared.settings[k]
-    shared.gradio['generate_state'] = gr.State(generate_params)
 
     with gr.Row():
         with gr.Column():
@@ -204,24 +246,24 @@ def create_settings_menus(default_preset):
     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))')
+                gr.Markdown('Custom generation parameters ([click here to view technical documentation](https://huggingface.co/docs/transformers/main_classes/text_generation#transformers.GenerationConfig))')
                 with gr.Row():
                     with gr.Column():
-                        shared.gradio['temperature'] = gr.Slider(0.01, 1.99, value=generate_params['temperature'], step=0.01, label='temperature')
-                        shared.gradio['top_p'] = gr.Slider(0.0, 1.0, value=generate_params['top_p'], step=0.01, label='top_p')
-                        shared.gradio['top_k'] = gr.Slider(0, 200, value=generate_params['top_k'], step=1, label='top_k')
-                        shared.gradio['typical_p'] = gr.Slider(0.0, 1.0, value=generate_params['typical_p'], step=0.01, label='typical_p')
+                        shared.gradio['temperature'] = gr.Slider(0.01, 1.99, value=generate_params['temperature'], step=0.01, label='temperature', info='Primary factor to control randomness of outputs. 0 = deterministic (only the most likely token is used). Higher value = more randomness.')
+                        shared.gradio['top_p'] = gr.Slider(0.0, 1.0, value=generate_params['top_p'], step=0.01, label='top_p', info='If not set to 1, select tokens with probabilities adding up to less than this number. Higher value = higher range of possible random results.')
+                        shared.gradio['top_k'] = gr.Slider(0, 200, value=generate_params['top_k'], step=1, label='top_k', info='Similar to top_p, but select instead only the top_k most likely tokens. Higher value = higher range of possible random results.')
+                        shared.gradio['typical_p'] = gr.Slider(0.0, 1.0, value=generate_params['typical_p'], step=0.01, label='typical_p', info='If not set to 1, select only tokens that are at least this much more likely to appear than random tokens, given the prior text.')
                     with gr.Column():
-                        shared.gradio['repetition_penalty'] = gr.Slider(1.0, 1.5, value=generate_params['repetition_penalty'], step=0.01, label='repetition_penalty')
-                        shared.gradio['encoder_repetition_penalty'] = gr.Slider(0.8, 1.5, value=generate_params['encoder_repetition_penalty'], step=0.01, label='encoder_repetition_penalty')
-                        shared.gradio['no_repeat_ngram_size'] = gr.Slider(0, 20, step=1, value=generate_params['no_repeat_ngram_size'], label='no_repeat_ngram_size')
-                        shared.gradio['min_length'] = gr.Slider(0, 2000, step=1, value=generate_params['min_length'] if shared.args.no_stream else 0, label='min_length', interactive=shared.args.no_stream)
+                        shared.gradio['repetition_penalty'] = gr.Slider(1.0, 1.5, value=generate_params['repetition_penalty'], step=0.01, label='repetition_penalty', info='Exponential penalty factor for repeating prior tokens. 1 means no penalty, higher value = less repetition, lower value = more repetition.')
+                        shared.gradio['encoder_repetition_penalty'] = gr.Slider(0.8, 1.5, value=generate_params['encoder_repetition_penalty'], step=0.01, label='encoder_repetition_penalty', info='Also known as the "Hallucinations filter". Used to penalize tokens that are *not* in the prior text. Higher value = more likely to stay in context, lower value = more likely to diverge.')
+                        shared.gradio['no_repeat_ngram_size'] = gr.Slider(0, 20, step=1, value=generate_params['no_repeat_ngram_size'], label='no_repeat_ngram_size', info='If not set to 0, specifies the length of token sets that are completely blocked from repeating at all. Higher values = blocks larger phrases, lower values = blocks words or letters from repeating. Only 0 or high values are a good idea in most cases.')
+                        shared.gradio['min_length'] = gr.Slider(0, 2000, step=1, value=generate_params['min_length'], label='min_length', info='Minimum generation length in tokens.')
                 shared.gradio['do_sample'] = gr.Checkbox(value=generate_params['do_sample'], label='do_sample')
         with gr.Column():
             with gr.Box():
                 gr.Markdown('Contrastive search')
                 shared.gradio['penalty_alpha'] = gr.Slider(0, 5, value=generate_params['penalty_alpha'], label='penalty_alpha')
-            with gr.Box():
+
                 gr.Markdown('Beam search (uses a lot of VRAM)')
                 with gr.Row():
                     with gr.Column():
@@ -230,6 +272,13 @@ 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')
 
+            with gr.Group():
+                with gr.Row():
+                    shared.gradio['add_bos_token'] = gr.Checkbox(value=shared.settings['add_bos_token'], label='Add the bos_token to the beginning of prompts', info='Disabling this can make the replies more creative.')
+                    shared.gradio['ban_eos_token'] = gr.Checkbox(value=shared.settings['ban_eos_token'], label='Ban the eos_token', info='This forces the model to never end the generation prematurely.')
+                shared.gradio['truncation_length'] = gr.Slider(value=shared.settings['truncation_length'], minimum=shared.settings['truncation_length_min'], maximum=shared.settings['truncation_length_max'], step=1, label='Truncate the prompt up to this length', info='The leftmost tokens are removed if the prompt exceeds this length. Most models require this to be at most 2048.')
+                shared.gradio['custom_stopping_strings'] = gr.Textbox(lines=1, value=shared.settings["custom_stopping_strings"] or None, label='Custom stopping strings', info='In addition to the defaults. Written between "" and separated by commas. For instance: "\\nYour Assistant:", "\\nThe assistant:"')
+
     with gr.Accordion('Soft prompt', open=False):
         with gr.Row():
             shared.gradio['softprompts_menu'] = gr.Dropdown(choices=available_softprompts, value='None', label='Soft prompt')
@@ -239,7 +288,7 @@ def create_settings_menus(default_preset):
         with gr.Row():
             shared.gradio['upload_softprompt'] = gr.File(type='binary', file_types=['.zip'])
 
-    shared.gradio['preset_menu'].change(load_preset_values, [shared.gradio[k] for k in ['preset_menu', 'generate_state']], [shared.gradio[k] for k in ['generate_state', '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[k] for k in ['preset_menu', 'interface_state']], [shared.gradio[k] for k in ['interface_state', '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['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'])
 
@@ -312,6 +361,21 @@ else:
 title = 'Text generation web UI'
 
 
+def list_interface_input_elements(chat=False):
+    elements = ['max_new_tokens', 'seed', 'temperature', 'top_p', 'top_k', 'typical_p', 'repetition_penalty', 'encoder_repetition_penalty', 'no_repeat_ngram_size', 'min_length', 'do_sample', 'penalty_alpha', 'num_beams', 'length_penalty', 'early_stopping', 'add_bos_token', 'ban_eos_token', 'truncation_length', 'custom_stopping_strings']
+    if chat:
+        elements += ['name1', 'name2', 'greeting', 'context', 'end_of_turn', 'chat_prompt_size', 'chat_generation_attempts', 'stop_at_newline', 'mode']
+    return elements
+
+
+def gather_interface_values(*args):
+    output = {}
+    for i, element in enumerate(shared.input_elements):
+        output[element] = args[i]
+    output['custom_stopping_strings'] = eval(f"[{output['custom_stopping_strings']}]")
+    return output
+
+
 def create_interface():
     gen_events = []
     if shared.args.extensions is not None and len(shared.args.extensions) > 0:
@@ -319,7 +383,11 @@ def create_interface():
 
     with gr.Blocks(css=ui.css if not shared.is_chat() else ui.css + ui.chat_css, analytics_enabled=False, title=title) as shared.gradio['interface']:
         if shared.is_chat():
+
+            shared.input_elements = list_interface_input_elements(chat=True)
+            shared.gradio['interface_state'] = gr.State({k: None for k in shared.input_elements})
             shared.gradio['Chat input'] = gr.State()
+
             with gr.Tab("Text generation", elem_id="main"):
                 shared.gradio['display'] = gr.HTML(value=chat_html_wrapper(shared.history['visible'], shared.settings['name1'], shared.settings['name2'], 'cai-chat'))
                 shared.gradio['textbox'] = gr.Textbox(label='Input')
@@ -327,19 +395,22 @@ def create_interface():
                     shared.gradio['Generate'] = gr.Button('Generate', elem_id='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')
+                    shared.gradio['Continue'] = gr.Button('Continue')
+                    shared.gradio['Impersonate'] = gr.Button('Impersonate')
                 with gr.Row():
-                    shared.gradio['Copy last reply'] = gr.Button('Copy last reply')
+                    shared.gradio['Send dummy message'] = gr.Button('Send dummy message')
+                    shared.gradio['Send dummy reply'] = gr.Button('Send dummy reply')
                     shared.gradio['Replace last reply'] = gr.Button('Replace last reply')
-                    shared.gradio['Remove last'] = gr.Button('Remove last')
-
+                    shared.gradio['Copy last reply'] = gr.Button('Copy last reply')
+                with gr.Row():
                     shared.gradio['Clear history'] = gr.Button('Clear history')
                     shared.gradio['Clear history-confirm'] = gr.Button('Confirm', variant="stop", visible=False)
                     shared.gradio['Clear history-cancel'] = gr.Button('Cancel', visible=False)
+                    shared.gradio['Remove last'] = gr.Button('Remove last')
 
-                shared.gradio["Chat mode"] = gr.Radio(choices=["cai-chat", "chat", "instruct"], value="cai-chat", label="Mode")
-                shared.gradio["Instruction templates"] = gr.Dropdown(choices=get_available_instruction_templates(), label="Instruction template", value="None", visible=False)
+                shared.gradio["mode"] = gr.Radio(choices=["cai-chat", "chat", "instruct"], value="cai-chat", label="Mode")
+                shared.gradio["Instruction templates"] = gr.Dropdown(choices=get_available_instruction_templates(), label="Instruction template", value="None", visible=False, info="Change this according to the model/LoRA that you are using.")
 
             with gr.Tab("Character", elem_id="chat-settings"):
                 with gr.Row():
@@ -386,66 +457,102 @@ def create_interface():
                     with gr.Row():
                         with gr.Column():
                             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'])
-                            shared.gradio['chat_prompt_size_slider'] = gr.Slider(minimum=shared.settings['chat_prompt_size_min'], maximum=shared.settings['chat_prompt_size_max'], step=1, label='Maximum prompt size in tokens', value=shared.settings['chat_prompt_size'])
+                            shared.gradio['chat_prompt_size'] = gr.Slider(minimum=shared.settings['chat_prompt_size_min'], maximum=shared.settings['chat_prompt_size_max'], step=1, label='Maximum prompt size in tokens', value=shared.settings['chat_prompt_size'])
                         with gr.Column():
                             shared.gradio['chat_generation_attempts'] = gr.Slider(minimum=shared.settings['chat_generation_attempts_min'], maximum=shared.settings['chat_generation_attempts_max'], value=shared.settings['chat_generation_attempts'], step=1, label='Generation attempts (for longer replies)')
-                            shared.gradio['stop_at_newline'] = gr.Checkbox(value=shared.settings['stop_at_newline'], label='Stop generating at new line character?')
+                            shared.gradio['stop_at_newline'] = gr.Checkbox(value=shared.settings['stop_at_newline'], label='Stop generating at new line character')
 
                 create_settings_menus(default_preset)
 
-            shared.input_params = [shared.gradio[k] for k in ['Chat input', 'generate_state', 'name1', 'name2', 'context', 'Chat mode', 'end_of_turn']]
-
-            def set_chat_input(textbox):
-                return textbox, ""
-
-            gen_events.append(shared.gradio['Generate'].click(set_chat_input, shared.gradio['textbox'], [shared.gradio['Chat input'], shared.gradio['textbox']], show_progress=False))
-            gen_events.append(shared.gradio['Generate'].click(chat.cai_chatbot_wrapper, shared.input_params, shared.gradio['display'], show_progress=shared.args.no_stream))
-            gen_events.append(shared.gradio['textbox'].submit(set_chat_input, shared.gradio['textbox'], [shared.gradio['Chat input'], shared.gradio['textbox']], show_progress=False))
-            gen_events.append(shared.gradio['textbox'].submit(chat.cai_chatbot_wrapper, 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(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[k] for k in ['textbox', 'name1', 'name2', 'Chat mode']], shared.gradio['display'], show_progress=shared.args.no_stream)
-
-            # Clear history with confirmation
+            shared.input_params = [shared.gradio[k] for k in ['Chat input', 'interface_state']]
             clear_arr = [shared.gradio[k] for k in ['Clear history-confirm', 'Clear history', 'Clear history-cancel']]
+            reload_inputs = [shared.gradio[k] for k in ['name1', 'name2', 'mode']]
+
+            gen_events.append(shared.gradio['Generate'].click(
+                gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
+                lambda x: (x, ''), shared.gradio['textbox'], [shared.gradio['Chat input'], shared.gradio['textbox']], show_progress=False).then(
+                chat.cai_chatbot_wrapper, shared.input_params, shared.gradio['display'], show_progress=shared.args.no_stream).then(
+                chat.save_history, shared.gradio['mode'], None, show_progress=False)
+            )
+
+            gen_events.append(shared.gradio['textbox'].submit(
+                gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
+                lambda x: (x, ''), shared.gradio['textbox'], [shared.gradio['Chat input'], shared.gradio['textbox']], show_progress=False).then(
+                chat.cai_chatbot_wrapper, shared.input_params, shared.gradio['display'], show_progress=shared.args.no_stream).then(
+                chat.save_history, shared.gradio['mode'], None, show_progress=False)
+            )
+
+            gen_events.append(shared.gradio['Regenerate'].click(
+                gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
+                chat.regenerate_wrapper, shared.input_params, shared.gradio['display'], show_progress=shared.args.no_stream).then(
+                chat.save_history, shared.gradio['mode'], None, show_progress=False)
+            )
+
+            gen_events.append(shared.gradio['Continue'].click(
+                gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
+                chat.continue_wrapper, shared.input_params, shared.gradio['display'], show_progress=shared.args.no_stream).then(
+                chat.save_history, shared.gradio['mode'], None, show_progress=False)
+            )
+
+            gen_events.append(shared.gradio['Impersonate'].click(
+                gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
+                chat.impersonate_wrapper, shared.input_params, shared.gradio['textbox'], show_progress=shared.args.no_stream)
+            )
+
+            shared.gradio['Replace last reply'].click(
+                chat.replace_last_reply, [shared.gradio[k] for k in ['textbox', 'name1', 'name2', 'mode']], shared.gradio['display'], show_progress=shared.args.no_stream).then(
+                lambda x: '', shared.gradio['textbox'], shared.gradio['textbox'], show_progress=False).then(
+                chat.save_history, shared.gradio['mode'], None, show_progress=False)
+
+            shared.gradio['Send dummy message'].click(
+                chat.send_dummy_message, [shared.gradio[k] for k in ['textbox', 'name1', 'name2', 'mode']], shared.gradio['display'], show_progress=shared.args.no_stream).then(
+                lambda x: '', shared.gradio['textbox'], shared.gradio['textbox'], show_progress=False).then(
+                chat.save_history, shared.gradio['mode'], None, show_progress=False)
+
+            shared.gradio['Send dummy reply'].click(
+                chat.send_dummy_reply, [shared.gradio[k] for k in ['textbox', 'name1', 'name2', 'mode']], shared.gradio['display'], show_progress=shared.args.no_stream).then(
+                lambda x: '', shared.gradio['textbox'], shared.gradio['textbox'], show_progress=False).then(
+                chat.save_history, shared.gradio['mode'], None, show_progress=False)
+
+            shared.gradio['Clear history-confirm'].click(
+                lambda: [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False)], None, clear_arr).then(
+                chat.clear_chat_log, [shared.gradio[k] for k in ['name1', 'name2', 'greeting', 'mode']], shared.gradio['display']).then(
+                chat.save_history, shared.gradio['mode'], None, show_progress=False)
+
+            shared.gradio['Stop'].click(
+                stop_everything_event, None, None, queue=False, cancels=gen_events if shared.args.no_stream else None).then(
+                chat.redraw_html, reload_inputs, shared.gradio['display'])
+
+            shared.gradio['mode'].change(
+                lambda x: gr.update(visible=x == 'instruct'), shared.gradio['mode'], shared.gradio['Instruction templates']).then(
+                lambda x: gr.update(interactive=x != 'instruct'), shared.gradio['mode'], shared.gradio['character_menu']).then(
+                chat.redraw_html, reload_inputs, shared.gradio['display'])
+
+            shared.gradio['Instruction templates'].change(
+                lambda character, name1, name2, mode: chat.load_character(character, name1, name2, mode), [shared.gradio[k] for k in ['Instruction templates', 'name1', 'name2', 'mode']], [shared.gradio[k] for k in ['name1', 'name2', 'character_picture', 'greeting', 'context', 'end_of_turn', 'display']]).then(
+                chat.redraw_html, reload_inputs, shared.gradio['display'])
+
+            shared.gradio['upload_chat_history'].upload(
+                chat.load_history, [shared.gradio[k] for k in ['upload_chat_history', 'name1', 'name2']], None).then(
+                chat.redraw_html, reload_inputs, shared.gradio['display'])
+
+            shared.gradio['Copy last reply'].click(chat.send_last_reply_to_input, None, shared.gradio['textbox'], show_progress=shared.args.no_stream)
             shared.gradio['Clear history'].click(lambda: [gr.update(visible=True), gr.update(visible=False), gr.update(visible=True)], None, clear_arr)
-            shared.gradio['Clear history-confirm'].click(lambda: [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False)], None, clear_arr)
-            shared.gradio['Clear history-confirm'].click(chat.clear_chat_log, [shared.gradio[k] for k in ['name1', 'name2', 'greeting', 'Chat mode']], shared.gradio['display'])
             shared.gradio['Clear history-cancel'].click(lambda: [gr.update(visible=False), gr.update(visible=True), gr.update(visible=False)], None, clear_arr)
-            shared.gradio['Chat mode'].change(lambda x: gr.update(visible=x == 'instruct'), shared.gradio['Chat mode'], shared.gradio['Instruction templates'])
-
-            shared.gradio['Remove last'].click(chat.remove_last_message, [shared.gradio[k] for k in ['name1', 'name2', 'Chat mode']], [shared.gradio['display'], shared.gradio['textbox']], show_progress=False)
-            shared.gradio['download_button'].click(chat.save_history, inputs=[], outputs=[shared.gradio['download']])
+            shared.gradio['Remove last'].click(chat.remove_last_message, [shared.gradio[k] for k in ['name1', 'name2', 'mode']], [shared.gradio['display'], shared.gradio['textbox']], show_progress=False)
+            shared.gradio['download_button'].click(lambda x: chat.save_history(x, timestamp=True), shared.gradio['mode'], shared.gradio['download'])
             shared.gradio['Upload character'].click(chat.upload_character, [shared.gradio['upload_json'], shared.gradio['upload_img_bot']], [shared.gradio['character_menu']])
-
-            # Clearing stuff and saving the history
-            for i in ['Generate', 'Regenerate', 'Replace last reply']:
-                shared.gradio[i].click(lambda x: '', shared.gradio['textbox'], shared.gradio['textbox'], show_progress=False)
-                shared.gradio[i].click(lambda: chat.save_history(timestamp=False), [], [], show_progress=False)
-            shared.gradio['Clear history-confirm'].click(lambda: chat.save_history(timestamp=False), [], [], show_progress=False)
-            shared.gradio['textbox'].submit(lambda x: '', shared.gradio['textbox'], shared.gradio['textbox'], show_progress=False)
-            shared.gradio['textbox'].submit(lambda: chat.save_history(timestamp=False), [], [], show_progress=False)
-
-            shared.gradio['character_menu'].change(chat.load_character, [shared.gradio[k] for k in ['character_menu', 'name1', 'name2', 'Chat mode']], [shared.gradio[k] for k in ['name1', 'name2', 'character_picture', 'greeting', 'context', 'end_of_turn', 'display']])
-            shared.gradio['Instruction templates'].change(lambda character, name1, name2, mode: chat.load_character(character, name1, name2, mode), [shared.gradio[k] for k in ['Instruction templates', 'name1', 'name2', 'Chat mode']], [shared.gradio[k] for k in ['name1', 'name2', 'character_picture', 'greeting', 'context', 'end_of_turn', 'display']])
-            shared.gradio['upload_chat_history'].upload(chat.load_history, [shared.gradio[k] for k in ['upload_chat_history', 'name1', 'name2']], [])
+            shared.gradio['character_menu'].change(chat.load_character, [shared.gradio[k] for k in ['character_menu', 'name1', 'name2', 'mode']], [shared.gradio[k] for k in ['name1', 'name2', 'character_picture', 'greeting', 'context', 'end_of_turn', 'display']])
             shared.gradio['upload_img_tavern'].upload(chat.upload_tavern_character, [shared.gradio['upload_img_tavern'], shared.gradio['name1'], shared.gradio['name2']], [shared.gradio['character_menu']])
-            shared.gradio['your_picture'].change(chat.upload_your_profile_picture, [shared.gradio[k] for k in ['your_picture', 'name1', 'name2', 'Chat mode']], shared.gradio['display'])
-
-            reload_inputs = [shared.gradio[k] for k in ['name1', 'name2', 'Chat mode']]
-            shared.gradio['upload_chat_history'].upload(chat.redraw_html, reload_inputs, [shared.gradio['display']])
-            shared.gradio['Stop'].click(chat.redraw_html, reload_inputs, [shared.gradio['display']])
-            shared.gradio['Instruction templates'].change(chat.redraw_html, reload_inputs, [shared.gradio['display']])
-            shared.gradio['Chat mode'].change(chat.redraw_html, reload_inputs, [shared.gradio['display']])
+            shared.gradio['your_picture'].change(chat.upload_your_profile_picture, [shared.gradio[k] for k in ['your_picture', 'name1', 'name2', 'mode']], shared.gradio['display'])
 
             shared.gradio['interface'].load(None, None, None, _js=f"() => {{{ui.main_js+ui.chat_js}}}")
-            shared.gradio['interface'].load(lambda: chat.load_default_history(shared.settings['name1'], shared.settings['name2']), None, None)
-            shared.gradio['interface'].load(chat.redraw_html, reload_inputs, [shared.gradio['display']], show_progress=True)
+            shared.gradio['interface'].load(chat.load_default_history, [shared.gradio[k] for k in ['name1', 'name2']], None)
+            shared.gradio['interface'].load(chat.redraw_html, reload_inputs, shared.gradio['display'], show_progress=True)
 
         elif shared.args.notebook:
+            shared.input_elements = list_interface_input_elements(chat=False)
+            shared.gradio['interface_state'] = gr.State({k: None for k in shared.input_elements})
             with gr.Tab("Text generation", elem_id="main"):
                 with gr.Row():
                     with gr.Column(scale=4):
@@ -473,14 +580,27 @@ def create_interface():
             with gr.Tab("Parameters", elem_id="parameters"):
                 create_settings_menus(default_preset)
 
-            shared.input_params = [shared.gradio[k] for k in ['textbox', 'generate_state']]
+            shared.input_params = [shared.gradio[k] for k in ['textbox', 'interface_state']]
             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))
-            gen_events.append(shared.gradio['textbox'].submit(generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream))
-            shared.gradio['Stop'].click(stop_everything_event, [], [], queue=False, cancels=gen_events if shared.args.no_stream else None)
+
+            gen_events.append(shared.gradio['Generate'].click(
+                gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
+                generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream)#.then(
+                #None, None, None, _js="() => {element = document.getElementsByTagName('textarea')[0]; element.scrollTop = element.scrollHeight}")
+            )
+
+            gen_events.append(shared.gradio['textbox'].submit(
+                gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
+                generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream)#.then(
+                #None, None, None, _js="() => {element = document.getElementsByTagName('textarea')[0]; element.scrollTop = element.scrollHeight}")
+            )
+
+            shared.gradio['Stop'].click(stop_everything_event, None, None, 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:
+            shared.input_elements = list_interface_input_elements(chat=False)
+            shared.gradio['interface_state'] = gr.State({k: None for k in shared.input_elements})
             with gr.Tab("Text generation", elem_id="main"):
                 with gr.Row():
                     with gr.Column():
@@ -506,12 +626,28 @@ def create_interface():
             with gr.Tab("Parameters", elem_id="parameters"):
                 create_settings_menus(default_preset)
 
-            shared.input_params = [shared.gradio[k] for k in ['textbox', 'generate_state']]
+            shared.input_params = [shared.gradio[k] for k in ['textbox', 'interface_state']]
             output_params = [shared.gradio[k] for k in ['output_textbox', 'markdown', 'html']]
-            gen_events.append(shared.gradio['Generate'].click(generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream))
-            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(stop_everything_event, [], [], queue=False, cancels=gen_events if shared.args.no_stream else None)
+
+            gen_events.append(shared.gradio['Generate'].click(
+                gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
+                generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream)#.then(
+                #None, None, None, _js="() => {element = document.getElementsByTagName('textarea')[1]; element.scrollTop = element.scrollHeight}")
+            )
+
+            gen_events.append(shared.gradio['textbox'].submit(
+                gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
+                generate_reply, shared.input_params, output_params, show_progress=shared.args.no_stream)#.then(
+                #None, None, None, _js="() => {element = document.getElementsByTagName('textarea')[1]; element.scrollTop = element.scrollHeight}")
+            )
+
+            gen_events.append(shared.gradio['Continue'].click(
+                gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
+                generate_reply, [shared.gradio['output_textbox']] + shared.input_params[1:], output_params, show_progress=shared.args.no_stream)#.then(
+                #None, None, None, _js="() => {element = document.getElementsByTagName('textarea')[1]; element.scrollTop = element.scrollHeight}")
+            )
+
+            shared.gradio['Stop'].click(stop_everything_event, None, None, 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("Model", elem_id="model-tab"):
@@ -537,24 +673,14 @@ def create_interface():
             shared.gradio['bool_menu'] = gr.CheckboxGroup(choices=bool_list, value=bool_active, label="Boolean command-line flags")
             shared.gradio['reset_interface'] = gr.Button("Apply and restart the interface")
 
-            shared.gradio['reset_interface'].click(set_interface_arguments, [shared.gradio[k] for k in ['interface_modes_menu', 'extensions_menu', 'bool_menu']], None)
-            shared.gradio['reset_interface'].click(lambda: None, None, None, _js='() => {document.body.innerHTML=\'<h1 style="font-family:monospace;margin-top:20%;color:lightgray;text-align:center;">Reloading...</h1>\'; setTimeout(function(){location.reload()},2500); return []}')
+            # Reset interface event
+            shared.gradio['reset_interface'].click(
+                set_interface_arguments, [shared.gradio[k] for k in ['interface_modes_menu', 'extensions_menu', 'bool_menu']], None).then(
+                lambda: None, None, None, _js='() => {document.body.innerHTML=\'<h1 style="font-family:monospace;margin-top:20%;color:lightgray;text-align:center;">Reloading...</h1>\'; setTimeout(function(){location.reload()},2500); return []}')
 
         if shared.args.extensions is not None:
             extensions_module.create_extensions_block()
 
-        def change_dict_value(d, key, value):
-            d[key] = value
-            return d
-
-        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', 'max_new_tokens', 'seed', 'stop_at_newline', 'chat_prompt_size_slider', 'chat_generation_attempts']:
-            if k not in shared.gradio:
-                continue
-            if type(shared.gradio[k]) in [gr.Checkbox, gr.Number]:
-                shared.gradio[k].change(lambda state, value, copy=k: change_dict_value(state, copy, value), inputs=[shared.gradio['generate_state'], shared.gradio[k]], outputs=shared.gradio['generate_state'])
-            else:
-                shared.gradio[k].release(lambda state, value, copy=k: change_dict_value(state, copy, value), inputs=[shared.gradio['generate_state'], shared.gradio[k]], outputs=shared.gradio['generate_state'])
-
         if not shared.is_chat():
             api.create_apis()
 

settings-template.json

@@ -7,7 +7,14 @@
     "name2": "Assistant",
     "context": "This is a conversation with your Assistant. The Assistant is very helpful and is eager to chat with you and answer your questions.",
     "greeting": "Hello there!",
+    "end_of_turn": "",
+    "custom_stopping_strings": "",
     "stop_at_newline": false,
+    "add_bos_token": true,
+    "ban_eos_token": true,
+    "truncation_length": 2048,
+    "truncation_length_min": 0,
+    "truncation_length_max": 4096,
     "chat_prompt_size": 2048,
     "chat_prompt_size_min": 0,
     "chat_prompt_size_max": 2048,
@@ -19,7 +26,8 @@
         "gallery"
     ],
     "presets": {
-        "default": "NovelAI-Sphinx Moth",
+        "default": "Default",
+        ".*(alpaca|llama)": "LLaMA-Precise",
         ".*pygmalion": "NovelAI-Storywriter",
         ".*RWKV": "Naive"
     },