Merge dev branch

docs/08 - Additional Tips.md

@@ -58,79 +58,6 @@ pip install -U deepspeed
 deepspeed --num_gpus=1 server.py --deepspeed --chat --model gpt-j-6B
 ```
 
-> RWKV: RNN with Transformer-level LLM Performance
->
-> It combines the best of RNN and transformer - great performance, fast inference, saves VRAM, fast training, "infinite" ctx_len, and free sentence embedding (using the final hidden state).
-
-https://github.com/BlinkDL/RWKV-LM
-
-https://github.com/BlinkDL/ChatRWKV
-
-## Using RWKV in the web UI
-
-### Hugging Face weights
-
-Simply download the weights from https://huggingface.co/RWKV and load them as you would for any other model.
-
-There is a bug in transformers==4.29.2 that prevents RWKV from being loaded in 8-bit mode. You can install the dev branch to solve this bug: `pip install git+https://github.com/huggingface/transformers`
-
-### Original .pth weights
-
-The instructions below are from before RWKV was supported in transformers, and they are kept for legacy purposes. The old implementation is possibly faster, but it lacks the full range of samplers that the transformers library offers.
-
-#### 0. Install the RWKV library
-
-```
-pip install rwkv
-```
-
-`0.7.3` was the last version that I tested. If you experience any issues, try ```pip install rwkv==0.7.3```.
-
-#### 1. Download the model
-
-It is available in different sizes:
-
-* https://huggingface.co/BlinkDL/rwkv-4-pile-3b/
-* https://huggingface.co/BlinkDL/rwkv-4-pile-7b/
-* https://huggingface.co/BlinkDL/rwkv-4-pile-14b/
-
-There are also older releases with smaller sizes like:
-
-* https://huggingface.co/BlinkDL/rwkv-4-pile-169m/resolve/main/RWKV-4-Pile-169M-20220807-8023.pth
-
-Download the chosen `.pth` and put it directly in the `models` folder. 
-
-#### 2. Download the tokenizer
-
-[20B_tokenizer.json](https://raw.githubusercontent.com/BlinkDL/ChatRWKV/main/v2/20B_tokenizer.json)
-
-Also put it directly in the `models` folder. Make sure to not rename it. It should be called `20B_tokenizer.json`.
-
-#### 3. Launch the web UI
-
-No additional steps are required. Just launch it as you would with any other model.
-
-```
-python server.py --listen  --no-stream --model RWKV-4-Pile-169M-20220807-8023.pth
-```
-
-#### Setting a custom strategy
-
-It is possible to have very fine control over the offloading and precision for the model with the `--rwkv-strategy` flag. Possible values include:
-
-```
-"cpu fp32" # CPU mode
-"cuda fp16" # GPU mode with float16 precision
-"cuda fp16 *30 -> cpu fp32" # GPU+CPU offloading. The higher the number after *, the higher the GPU allocation.
-"cuda fp16i8" # GPU mode with 8-bit precision
-```
-
-See the README for the PyPl package for more details: https://pypi.org/project/rwkv/
-
-#### Compiling the CUDA kernel
-
-You can compile the CUDA kernel for the model with `--rwkv-cuda-on`. This should improve the performance a lot but I haven't been able to get it to work yet.
-
 ## Miscellaneous info
 
 ### You can train LoRAs in CPU mode

extensions/gallery/script.py

@@ -3,9 +3,14 @@
 import gradio as gr
 
 from modules.html_generator import get_image_cache
-from modules.shared import gradio, settings
+from modules.shared import gradio
 
 
+params = {
+    'items_per_page': 50,
+    'open': False,
+}
+
 cards = []
 
 
@@ -104,7 +109,7 @@ def custom_js():
 
 
 def ui():
-    with gr.Accordion("Character gallery", open=settings["gallery-open"], elem_id='gallery-extension'):
+    with gr.Accordion("Character gallery", open=params["open"], elem_id='gallery-extension'):
         gr.HTML(value="<style>" + generate_css() + "</style>")
         with gr.Row():
             filter_box = gr.Textbox(label='', placeholder='Filter', lines=1, max_lines=1, container=False, elem_id='gallery-filter-box')
@@ -116,7 +121,7 @@ def ui():
             label="",
             samples=generate_html(),
             elem_classes=["character-gallery"],
-            samples_per_page=settings["gallery-items_per_page"]
+            samples_per_page=params["items_per_page"]
         )
 
     filter_box.change(lambda: None, None, None, js=f'() => {{{custom_js()}; gotoFirstPage()}}').success(

instruction-templates/Command-R.yaml

@@ -0,0 +1,26 @@
+instruction_template: |-
+  {%- if messages[0]['role'] == 'system' -%}
+      {%- set loop_messages = messages[1:] -%}
+      {%- set system_message = messages[0]['content'] -%}
+  {%- elif false == true -%}
+      {%- set loop_messages = messages -%}
+      {%- set system_message = 'You are Command-R, a brilliant, sophisticated, AI-assistant trained to assist human users by providing thorough responses. You are trained by Cohere.' -%}
+  {%- else -%}
+      {%- set loop_messages = messages -%}
+      {%- set system_message = false -%}
+  {%- endif -%}
+  {%- if system_message != false -%}
+      {{ '<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>' + system_message + '<|END_OF_TURN_TOKEN|>' }}
+  {%- endif -%}
+  {%- for message in loop_messages -%}
+      {%- set content = message['content'] -%}
+      {%- if message['role'] == 'user' -%}
+          {{ '<|START_OF_TURN_TOKEN|><|USER_TOKEN|>' + content.strip() + '<|END_OF_TURN_TOKEN|>' }}
+      {%- elif message['role'] == 'assistant' -%}
+          {{ '<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>'  + content.strip() + '<|END_OF_TURN_TOKEN|>' }}
+      {%- endif -%}
+  {%- endfor -%}
+  {%- if add_generation_prompt -%}
+      {{ '<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>' }}
+  {%- endif -%}
+

js/main.js

@@ -464,3 +464,22 @@ function handleVisibilityChange(isVisible) {
 }
 
 respondToRenameVisibility(renameTextArea, handleVisibilityChange);
+
+//------------------------------------------------
+// Adjust the chat tab margin if no extension UI
+// is present at the bottom
+//------------------------------------------------
+
+if (document.getElementById('extensions') === null) {
+  document.getElementById("chat-tab").style.marginBottom = "-29px";
+}
+
+//------------------------------------------------
+// Focus on the chat input after starting a new chat
+//------------------------------------------------
+
+document.querySelectorAll('.focus-on-chat-input').forEach(element => {
+  element.addEventListener('click', function() {
+      document.querySelector('#chat-input textarea').focus();
+  });
+});

models/config.yaml

@@ -192,3 +192,5 @@
   instruction_template: 'Synthia'
 .*(hercules|hyperion):
   instruction_template: 'ChatML'
+.*command-r:
+  instruction_template: 'Command-R'

modules/chat.py

@@ -125,7 +125,7 @@ def generate_chat_prompt(user_input, state, **kwargs):
         messages.append({"role": "user", "content": user_input})
 
     def remove_extra_bos(prompt):
-        for bos_token in ['<s>', '<|startoftext|>']:
+        for bos_token in ['<s>', '<|startoftext|>', '<BOS_TOKEN>', '<|endoftext|>']:
             while prompt.startswith(bos_token):
                 prompt = prompt[len(bos_token):]
 

modules/models.py

@@ -1,6 +1,7 @@
 import gc
 import logging
 import os
+import pprint
 import re
 import time
 import traceback
@@ -126,15 +127,19 @@ def huggingface_loader(model_name):
     path_to_model = Path(f'{shared.args.model_dir}/{model_name}')
     params = {
         'low_cpu_mem_usage': True,
-        'trust_remote_code': shared.args.trust_remote_code,
         'torch_dtype': torch.bfloat16 if shared.args.bf16 else torch.float16,
-        'use_safetensors': True if shared.args.force_safetensors else None
     }
 
+    if shared.args.trust_remote_code:
+        params['trust_remote_code'] = True
+
     if shared.args.use_flash_attention_2:
         params['use_flash_attention_2'] = True
 
-    config = AutoConfig.from_pretrained(path_to_model, trust_remote_code=params['trust_remote_code'])
+    if shared.args.force_safetensors:
+        params['force_safetensors'] = True
+
+    config = AutoConfig.from_pretrained(path_to_model, trust_remote_code=shared.args.trust_remote_code)
 
     if 'chatglm' in model_name.lower():
         LoaderClass = AutoModel
@@ -147,6 +152,10 @@ def huggingface_loader(model_name):
 
     # Load the model without any special settings
     if not any([shared.args.cpu, shared.args.load_in_8bit, shared.args.load_in_4bit, shared.args.auto_devices, shared.args.disk, shared.args.deepspeed, shared.args.gpu_memory is not None, shared.args.cpu_memory is not None, shared.args.compress_pos_emb > 1, shared.args.alpha_value > 1, shared.args.disable_exllama, shared.args.disable_exllamav2]):
+        logger.info("TRANSFORMERS_PARAMS=")
+        pprint.PrettyPrinter(indent=4, sort_dicts=False).pprint(params)
+        print()
+
         model = LoaderClass.from_pretrained(path_to_model, **params)
         if not (hasattr(model, 'is_loaded_in_4bit') and model.is_loaded_in_4bit):
             if torch.backends.mps.is_available():
@@ -175,7 +184,9 @@ def huggingface_loader(model_name):
             params['torch_dtype'] = torch.float32
         else:
             params['device_map'] = 'auto'
-            params['max_memory'] = get_max_memory_dict()
+            if x := get_max_memory_dict():
+                params['max_memory'] = x
+
             if shared.args.load_in_4bit:
                 # See https://github.com/huggingface/transformers/pull/23479/files
                 # and https://huggingface.co/blog/4bit-transformers-bitsandbytes
@@ -186,7 +197,6 @@ def huggingface_loader(model_name):
                     'bnb_4bit_use_double_quant': shared.args.use_double_quant,
                 }
 
-                logger.info('Using the following 4-bit params: ' + str(quantization_config_params))
                 params['quantization_config'] = BitsAndBytesConfig(**quantization_config_params)
 
             elif shared.args.load_in_8bit:
@@ -230,6 +240,9 @@ def huggingface_loader(model_name):
         elif shared.args.alpha_value > 1:
             params['rope_scaling'] = {'type': 'dynamic', 'factor': RoPE.get_alpha_value(shared.args.alpha_value, shared.args.rope_freq_base)}
 
+        logger.info("TRANSFORMERS_PARAMS=")
+        pprint.PrettyPrinter(indent=4, sort_dicts=False).pprint(params)
+        print()
         model = LoaderClass.from_pretrained(path_to_model, **params)
 
     return model

modules/models_settings.py

@@ -117,6 +117,9 @@ def get_model_metadata(model):
         metadata = json.loads(open(path, 'r', encoding='utf-8').read())
         if 'chat_template' in metadata:
             template = metadata['chat_template']
+            if isinstance(template, list):
+                template = template[0]['template']
+
             for k in ['eos_token', 'bos_token']:
                 if k in metadata:
                     value = metadata[k]

modules/shared.py

@@ -62,9 +62,7 @@
     'chat_template_str': "{%- for message in messages %}\n    {%- if message['role'] == 'system' -%}\n        {%- if message['content'] -%}\n            {{- message['content'] + '\\n\\n' -}}\n        {%- endif -%}\n        {%- if user_bio -%}\n            {{- user_bio + '\\n\\n' -}}\n        {%- endif -%}\n    {%- else -%}\n        {%- if message['role'] == 'user' -%}\n            {{- name1 + ': ' + message['content'] + '\\n'-}}\n        {%- else -%}\n            {{- name2 + ': ' + message['content'] + '\\n' -}}\n        {%- endif -%}\n    {%- endif -%}\n{%- endfor -%}",
     'chat-instruct_command': 'Continue the chat dialogue below. Write a single reply for the character "<|character|>".\n\n<|prompt|>',
     'autoload_model': False,
-    'gallery-items_per_page': 50,
-    'gallery-open': False,
-    'default_extensions': ['gallery'],
+    'default_extensions': [],
 }
 
 default_settings = copy.deepcopy(settings)

modules/ui_chat.py

@@ -68,14 +68,14 @@ def create_ui():
                 with gr.Row():
                     shared.gradio['rename_chat'] = gr.Button('Rename', elem_classes='refresh-button', interactive=not mu)
                     shared.gradio['delete_chat'] = gr.Button('🗑️', elem_classes='refresh-button', interactive=not mu)
-                    shared.gradio['delete_chat-confirm'] = gr.Button('Confirm', variant='stop', visible=False, elem_classes='refresh-button')
-                    shared.gradio['delete_chat-cancel'] = gr.Button('Cancel', visible=False, elem_classes='refresh-button')
-                    shared.gradio['Start new chat'] = gr.Button('New chat', elem_classes='refresh-button')
+                    shared.gradio['delete_chat-confirm'] = gr.Button('Confirm', variant='stop', visible=False, elem_classes=['refresh-button', 'focus-on-chat-input'])
+                    shared.gradio['delete_chat-cancel'] = gr.Button('Cancel', visible=False, elem_classes=['refresh-button', 'focus-on-chat-input'])
+                    shared.gradio['Start new chat'] = gr.Button('New chat', elem_classes=['refresh-button', 'focus-on-chat-input'])
 
                 with gr.Row(elem_id='rename-row'):
                     shared.gradio['rename_to'] = gr.Textbox(label='Rename to:', placeholder='New name', visible=False, elem_classes=['no-background'])
-                    shared.gradio['rename_to-confirm'] = gr.Button('Confirm', visible=False, elem_classes='refresh-button')
-                    shared.gradio['rename_to-cancel'] = gr.Button('Cancel', visible=False, elem_classes='refresh-button')
+                    shared.gradio['rename_to-confirm'] = gr.Button('Confirm', visible=False, elem_classes=['refresh-button', 'focus-on-chat-input'])
+                    shared.gradio['rename_to-cancel'] = gr.Button('Cancel', visible=False, elem_classes=['refresh-button', 'focus-on-chat-input'])
 
         with gr.Row(elem_id='chat-controls', elem_classes=['pretty_scrollbar']):
             with gr.Column():
@@ -378,4 +378,4 @@ def create_event_handlers():
         partial(chat.generate_chat_prompt, '', _continue=True), gradio('interface_state'), gradio('textbox-notebook')).then(
         lambda: None, None, None, js=f'() => {{{ui.switch_tabs_js}; switch_to_notebook()}}')
 
-    shared.gradio['show_controls'].change(None, gradio('show_controls'), None, js=f'(x) => {{{ui.show_controls_js}; toggle_controls(x)}}')
+    shared.gradio['show_controls'].change(lambda x: None, gradio('show_controls'), None, js=f'(x) => {{{ui.show_controls_js}; toggle_controls(x)}}')

requirements.txt

@@ -1,10 +1,10 @@
 accelerate==0.27.*
-aqlm[gpu,cpu]==1.1.2; platform_system == "Linux"
+aqlm[gpu,cpu]==1.1.3; platform_system == "Linux"
 bitsandbytes==0.43.*
 colorama
 datasets
 einops
-gradio==4.23.*
+gradio==4.25.*
 hqq==0.1.5
 jinja2==3.1.2
 lm_eval==0.3.0
@@ -33,22 +33,22 @@ sse-starlette==1.6.5
 tiktoken
 
 # llama-cpp-python (CPU only, AVX2)
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.59+cpuavx2-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.59+cpuavx2-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.59+cpuavx2-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.59+cpuavx2-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.60+cpuavx2-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.60+cpuavx2-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.60+cpuavx2-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/cpu/llama_cpp_python-0.2.60+cpuavx2-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
 
 # llama-cpp-python (CUDA, no tensor cores)
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.59+cu121-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.59+cu121-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.59+cu121-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.59+cu121-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.60+cu121-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.60+cu121-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.60+cu121-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda-0.2.60+cu121-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
 
 # llama-cpp-python (CUDA, tensor cores)
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.59+cu121-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.59+cu121-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.59+cu121-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
-https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.59+cu121-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.60+cu121-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.60+cu121-cp310-cp310-win_amd64.whl; platform_system == "Windows" and python_version == "3.10"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.60+cu121-cp311-cp311-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.11"
+https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/textgen-webui/llama_cpp_python_cuda_tensorcores-0.2.60+cu121-cp310-cp310-manylinux_2_31_x86_64.whl; platform_system == "Linux" and platform_machine == "x86_64" and python_version == "3.10"
 
 # CUDA wheels
 https://github.com/jllllll/AutoGPTQ/releases/download/v0.6.0/auto_gptq-0.6.0+cu121-cp311-cp311-win_amd64.whl; platform_system == "Windows" and python_version == "3.11"