forestcalled/text-generation-webui
0
1import importlib2import math3import re4import traceback5from functools import partial6from pathlib import Path7 8import gradio as gr9import psutil10import torch11from transformers import is_torch_xpu_available12 13from modules import loaders, shared, ui, utils14from modules.logging_colors import logger15from modules.LoRA import add_lora_to_model16from modules.models import load_model, unload_model17from modules.models_settings import (18 apply_model_settings_to_state,19 get_model_metadata,20 save_model_settings,21 update_model_parameters22)23from modules.utils import gradio24 25 26def create_ui():27 mu = shared.args.multi_user28 29 # Finding the default values for the GPU and CPU memories30 total_mem = []31 if is_torch_xpu_available():32 for i in range(torch.xpu.device_count()):33 total_mem.append(math.floor(torch.xpu.get_device_properties(i).total_memory / (1024 * 1024)))34 else:35 for i in range(torch.cuda.device_count()):36 total_mem.append(math.floor(torch.cuda.get_device_properties(i).total_memory / (1024 * 1024)))37 38 default_gpu_mem = []39 if shared.args.gpu_memory is not None and len(shared.args.gpu_memory) > 0:40 for i in shared.args.gpu_memory:41 if 'mib' in i.lower():42 default_gpu_mem.append(int(re.sub('[a-zA-Z ]', '', i)))43 else:44 default_gpu_mem.append(int(re.sub('[a-zA-Z ]', '', i)) * 1000)45 46 while len(default_gpu_mem) < len(total_mem):47 default_gpu_mem.append(0)48 49 total_cpu_mem = math.floor(psutil.virtual_memory().total / (1024 * 1024))50 if shared.args.cpu_memory is not None:51 default_cpu_mem = re.sub('[a-zA-Z ]', '', shared.args.cpu_memory)52 else:53 default_cpu_mem = 054 55 with gr.Tab("Model", elem_id="model-tab"):56 with gr.Row():57 with gr.Column():58 with gr.Row():59 with gr.Column():60 with gr.Row():61 shared.gradio['model_menu'] = gr.Dropdown(choices=utils.get_available_models(), value=lambda: shared.model_name, label='Model', elem_classes='slim-dropdown', interactive=not mu)62 ui.create_refresh_button(shared.gradio['model_menu'], lambda: None, lambda: {'choices': utils.get_available_models()}, 'refresh-button', interactive=not mu)63 shared.gradio['load_model'] = gr.Button("Load", visible=not shared.settings['autoload_model'], elem_classes='refresh-button', interactive=not mu)64 shared.gradio['unload_model'] = gr.Button("Unload", elem_classes='refresh-button', interactive=not mu)65 shared.gradio['reload_model'] = gr.Button("Reload", elem_classes='refresh-button', interactive=not mu)66 shared.gradio['save_model_settings'] = gr.Button("Save settings", elem_classes='refresh-button', interactive=not mu)67 68 with gr.Column():69 with gr.Row():70 shared.gradio['lora_menu'] = gr.Dropdown(multiselect=True, choices=utils.get_available_loras(), value=shared.lora_names, label='LoRA(s)', elem_classes='slim-dropdown', interactive=not mu)71 ui.create_refresh_button(shared.gradio['lora_menu'], lambda: None, lambda: {'choices': utils.get_available_loras(), 'value': shared.lora_names}, 'refresh-button', interactive=not mu)72 shared.gradio['lora_menu_apply'] = gr.Button(value='Apply LoRAs', elem_classes='refresh-button', interactive=not mu)73 74 with gr.Row():75 with gr.Column():76 shared.gradio['loader'] = gr.Dropdown(label="Model loader", choices=loaders.loaders_and_params.keys(), value=None)77 with gr.Box():78 with gr.Row():79 with gr.Column():80 for i in range(len(total_mem)):81 shared.gradio[f'gpu_memory_{i}'] = gr.Slider(label=f"gpu-memory in MiB for device :{i}", maximum=total_mem[i], value=default_gpu_mem[i])82 83 shared.gradio['cpu_memory'] = gr.Slider(label="cpu-memory in MiB", maximum=total_cpu_mem, value=default_cpu_mem)84 shared.gradio['transformers_info'] = gr.Markdown('load-in-4bit params:')85 shared.gradio['compute_dtype'] = gr.Dropdown(label="compute_dtype", choices=["bfloat16", "float16", "float32"], value=shared.args.compute_dtype)86 shared.gradio['quant_type'] = gr.Dropdown(label="quant_type", choices=["nf4", "fp4"], value=shared.args.quant_type)87 shared.gradio['hqq_backend'] = gr.Dropdown(label="hqq_backend", choices=["PYTORCH", "PYTORCH_COMPILE", "ATEN"], value=shared.args.hqq_backend)88 89 shared.gradio['n_gpu_layers'] = gr.Slider(label="n-gpu-layers", minimum=0, maximum=128, value=shared.args.n_gpu_layers)90 shared.gradio['n_ctx'] = gr.Slider(minimum=0, maximum=shared.settings['truncation_length_max'], step=256, label="n_ctx", value=shared.args.n_ctx, info='Context length. Try lowering this if you run out of memory while loading the model.')91 shared.gradio['threads'] = gr.Slider(label="threads", minimum=0, step=1, maximum=32, value=shared.args.threads)92 shared.gradio['threads_batch'] = gr.Slider(label="threads_batch", minimum=0, step=1, maximum=32, value=shared.args.threads_batch)93 shared.gradio['n_batch'] = gr.Slider(label="n_batch", minimum=1, maximum=2048, value=shared.args.n_batch)94 95 shared.gradio['wbits'] = gr.Dropdown(label="wbits", choices=["None", 1, 2, 3, 4, 8], value=shared.args.wbits if shared.args.wbits > 0 else "None")96 shared.gradio['groupsize'] = gr.Dropdown(label="groupsize", choices=["None", 32, 64, 128, 1024], value=shared.args.groupsize if shared.args.groupsize > 0 else "None")97 shared.gradio['model_type'] = gr.Dropdown(label="model_type", choices=["None"], value=shared.args.model_type or "None")98 shared.gradio['pre_layer'] = gr.Slider(label="pre_layer", minimum=0, maximum=100, value=shared.args.pre_layer[0] if shared.args.pre_layer is not None else 0)99 shared.gradio['autogptq_info'] = gr.Markdown('* ExLlama_HF is recommended over AutoGPTQ for models derived from Llama.')100 shared.gradio['gpu_split'] = gr.Textbox(label='gpu-split', info='Comma-separated list of VRAM (in GB) to use per GPU. Example: 20,7,7')101 shared.gradio['max_seq_len'] = gr.Slider(label='max_seq_len', minimum=0, maximum=shared.settings['truncation_length_max'], step=256, info='Context length. Try lowering this if you run out of memory while loading the model.', value=shared.args.max_seq_len)102 shared.gradio['alpha_value'] = gr.Slider(label='alpha_value', minimum=1, maximum=8, step=0.05, info='Positional embeddings alpha factor for NTK RoPE scaling. Recommended values (NTKv1): 1.75 for 1.5x context, 2.5 for 2x context. Use either this or compress_pos_emb, not both.', value=shared.args.alpha_value)103 shared.gradio['rope_freq_base'] = gr.Slider(label='rope_freq_base', minimum=0, maximum=1000000, step=1000, info='If greater than 0, will be used instead of alpha_value. Those two are related by rope_freq_base = 10000 * alpha_value ^ (64 / 63)', value=shared.args.rope_freq_base)104 shared.gradio['compress_pos_emb'] = gr.Slider(label='compress_pos_emb', minimum=1, maximum=8, step=1, info='Positional embeddings compression factor. Should be set to (context length) / (model\'s original context length). Equal to 1/rope_freq_scale.', value=shared.args.compress_pos_emb)105 shared.gradio['quipsharp_info'] = gr.Markdown('QuIP# only works on Linux.')106 107 with gr.Column():108 shared.gradio['tensorcores'] = gr.Checkbox(label="tensorcores", value=shared.args.tensorcores, info='Use llama-cpp-python compiled with tensor cores support. This increases performance on RTX cards. NVIDIA only.')109 shared.gradio['no_offload_kqv'] = gr.Checkbox(label="no_offload_kqv", value=shared.args.no_offload_kqv, info='Do not offload the K, Q, V to the GPU. This saves VRAM but reduces the performance.')110 shared.gradio['triton'] = gr.Checkbox(label="triton", value=shared.args.triton)111 shared.gradio['no_inject_fused_attention'] = gr.Checkbox(label="no_inject_fused_attention", value=shared.args.no_inject_fused_attention, info='Disable fused attention. Fused attention improves inference performance but uses more VRAM. Fuses layers for AutoAWQ. Disable if running low on VRAM.')112 shared.gradio['no_inject_fused_mlp'] = gr.Checkbox(label="no_inject_fused_mlp", value=shared.args.no_inject_fused_mlp, info='Affects Triton only. Disable fused MLP. Fused MLP improves performance but uses more VRAM. Disable if running low on VRAM.')113 shared.gradio['no_use_cuda_fp16'] = gr.Checkbox(label="no_use_cuda_fp16", value=shared.args.no_use_cuda_fp16, info='This can make models faster on some systems.')114 shared.gradio['desc_act'] = gr.Checkbox(label="desc_act", value=shared.args.desc_act, info='\'desc_act\', \'wbits\', and \'groupsize\' are used for old models without a quantize_config.json.')115 shared.gradio['no_mul_mat_q'] = gr.Checkbox(label="no_mul_mat_q", value=shared.args.no_mul_mat_q, info='Disable the mulmat kernels.')116 shared.gradio['no_mmap'] = gr.Checkbox(label="no-mmap", value=shared.args.no_mmap)117 shared.gradio['mlock'] = gr.Checkbox(label="mlock", value=shared.args.mlock)118 shared.gradio['numa'] = gr.Checkbox(label="numa", value=shared.args.numa, info='NUMA support can help on some systems with non-uniform memory access.')119 shared.gradio['cpu'] = gr.Checkbox(label="cpu", value=shared.args.cpu)120 shared.gradio['load_in_8bit'] = gr.Checkbox(label="load-in-8bit", value=shared.args.load_in_8bit)121 shared.gradio['bf16'] = gr.Checkbox(label="bf16", value=shared.args.bf16)122 shared.gradio['auto_devices'] = gr.Checkbox(label="auto-devices", value=shared.args.auto_devices)123 shared.gradio['disk'] = gr.Checkbox(label="disk", value=shared.args.disk)124 shared.gradio['load_in_4bit'] = gr.Checkbox(label="load-in-4bit", value=shared.args.load_in_4bit)125 shared.gradio['use_double_quant'] = gr.Checkbox(label="use_double_quant", value=shared.args.use_double_quant)126 shared.gradio['tensor_split'] = gr.Textbox(label='tensor_split', info='Split the model across multiple GPUs, comma-separated list of proportions, e.g. 18,17')127 shared.gradio['trust_remote_code'] = gr.Checkbox(label="trust-remote-code", value=shared.args.trust_remote_code, info='To enable this option, start the web UI with the --trust-remote-code flag. It is necessary for some models.', interactive=shared.args.trust_remote_code)128 shared.gradio['cfg_cache'] = gr.Checkbox(label="cfg-cache", value=shared.args.cfg_cache, info='Create an additional cache for CFG negative prompts.')129 shared.gradio['logits_all'] = gr.Checkbox(label="logits_all", value=shared.args.logits_all, info='Needs to be set for perplexity evaluation to work. Otherwise, ignore it, as it makes prompt processing slower.')130 shared.gradio['use_flash_attention_2'] = gr.Checkbox(label="use_flash_attention_2", value=shared.args.use_flash_attention_2, info='Set use_flash_attention_2=True while loading the model.')131 shared.gradio['disable_exllama'] = gr.Checkbox(label="disable_exllama", value=shared.args.disable_exllama, info='Disable ExLlama kernel.')132 shared.gradio['disable_exllamav2'] = gr.Checkbox(label="disable_exllamav2", value=shared.args.disable_exllamav2, info='Disable ExLlamav2 kernel.')133 shared.gradio['no_flash_attn'] = gr.Checkbox(label="no_flash_attn", value=shared.args.no_flash_attn, info='Force flash-attention to not be used.')134 shared.gradio['cache_8bit'] = gr.Checkbox(label="cache_8bit", value=shared.args.cache_8bit, info='Use 8-bit cache to save VRAM.')135 shared.gradio['no_use_fast'] = gr.Checkbox(label="no_use_fast", value=shared.args.no_use_fast, info='Set use_fast=False while loading the tokenizer.')136 shared.gradio['num_experts_per_token'] = gr.Number(label="Number of experts per token", value=shared.args.num_experts_per_token, info='Only applies to MoE models like Mixtral.')137 shared.gradio['gptq_for_llama_info'] = gr.Markdown('Legacy loader for compatibility with older GPUs. ExLlama_HF or AutoGPTQ are preferred for GPTQ models when supported.')138 shared.gradio['exllama_info'] = gr.Markdown("ExLlama_HF is recommended over ExLlama for better integration with extensions and more consistent sampling behavior across loaders.")139 shared.gradio['exllamav2_info'] = gr.Markdown("ExLlamav2_HF is recommended over ExLlamav2 for better integration with extensions and more consistent sampling behavior across loaders.")140 shared.gradio['llamacpp_HF_info'] = gr.Markdown('llamacpp_HF loads llama.cpp as a Transformers model. To use it, you need to download a tokenizer.\n\nOption 1 (recommended): place your .gguf in a subfolder of models/ along with these 4 files: special_tokens_map.json, tokenizer_config.json, tokenizer.json, tokenizer.model.\n\nOption 2: download `oobabooga/llama-tokenizer` under "Download model or LoRA". That\'s a default Llama tokenizer that will work for some (but not all) models.')141 142 with gr.Column():143 with gr.Row():144 shared.gradio['autoload_model'] = gr.Checkbox(value=shared.settings['autoload_model'], label='Autoload the model', info='Whether to load the model as soon as it is selected in the Model dropdown.', interactive=not mu)145 146 shared.gradio['custom_model_menu'] = gr.Textbox(label="Download model or LoRA", info="Enter the Hugging Face username/model path, for instance: facebook/galactica-125m. To specify a branch, add it at the end after a \":\" character like this: facebook/galactica-125m:main. To download a single file, enter its name in the second box.", interactive=not mu)147 shared.gradio['download_specific_file'] = gr.Textbox(placeholder="File name (for GGUF models)", show_label=False, max_lines=1, interactive=not mu)148 with gr.Row():149 shared.gradio['download_model_button'] = gr.Button("Download", variant='primary', interactive=not mu)150 shared.gradio['get_file_list'] = gr.Button("Get file list", interactive=not mu)151 152 with gr.Row():153 shared.gradio['model_status'] = gr.Markdown('No model is loaded' if shared.model_name == 'None' else 'Ready')154 155 156def create_event_handlers():157 shared.gradio['loader'].change(158 loaders.make_loader_params_visible, gradio('loader'), gradio(loaders.get_all_params())).then(159 lambda value: gr.update(choices=loaders.get_model_types(value)), gradio('loader'), gradio('model_type'))160 161 # In this event handler, the interface state is read and updated162 # with the model defaults (if any), and then the model is loaded163 # unless "autoload_model" is unchecked164 shared.gradio['model_menu'].change(165 ui.gather_interface_values, gradio(shared.input_elements), gradio('interface_state')).then(166 apply_model_settings_to_state, gradio('model_menu', 'interface_state'), gradio('interface_state')).then(167 ui.apply_interface_values, gradio('interface_state'), gradio(ui.list_interface_input_elements()), show_progress=False).then(168 update_model_parameters, gradio('interface_state'), None).then(169 load_model_wrapper, gradio('model_menu', 'loader', 'autoload_model'), gradio('model_status'), show_progress=False).success(170 update_truncation_length, gradio('truncation_length', 'interface_state'), gradio('truncation_length')).then(171 lambda x: x, gradio('loader'), gradio('filter_by_loader'))172 173 shared.gradio['load_model'].click(174 ui.gather_interface_values, gradio(shared.input_elements), gradio('interface_state')).then(175 update_model_parameters, gradio('interface_state'), None).then(176 partial(load_model_wrapper, autoload=True), gradio('model_menu', 'loader'), gradio('model_status'), show_progress=False).success(177 update_truncation_length, gradio('truncation_length', 'interface_state'), gradio('truncation_length')).then(178 lambda x: x, gradio('loader'), gradio('filter_by_loader'))179 180 shared.gradio['reload_model'].click(181 unload_model, None, None).then(182 ui.gather_interface_values, gradio(shared.input_elements), gradio('interface_state')).then(183 update_model_parameters, gradio('interface_state'), None).then(184 partial(load_model_wrapper, autoload=True), gradio('model_menu', 'loader'), gradio('model_status'), show_progress=False).success(185 update_truncation_length, gradio('truncation_length', 'interface_state'), gradio('truncation_length')).then(186 lambda x: x, gradio('loader'), gradio('filter_by_loader'))187 188 shared.gradio['unload_model'].click(189 unload_model, None, None).then(190 lambda: "Model unloaded", None, gradio('model_status'))191 192 shared.gradio['save_model_settings'].click(193 ui.gather_interface_values, gradio(shared.input_elements), gradio('interface_state')).then(194 save_model_settings, gradio('model_menu', 'interface_state'), gradio('model_status'), show_progress=False)195 196 shared.gradio['lora_menu_apply'].click(load_lora_wrapper, gradio('lora_menu'), gradio('model_status'), show_progress=False)197 shared.gradio['download_model_button'].click(download_model_wrapper, gradio('custom_model_menu', 'download_specific_file'), gradio('model_status'), show_progress=True)198 shared.gradio['get_file_list'].click(partial(download_model_wrapper, return_links=True), gradio('custom_model_menu', 'download_specific_file'), gradio('model_status'), show_progress=True)199 shared.gradio['autoload_model'].change(lambda x: gr.update(visible=not x), gradio('autoload_model'), gradio('load_model'))200 201 202def load_model_wrapper(selected_model, loader, autoload=False):203 if not autoload:204 yield f"The settings for `{selected_model}` have been updated.\n\nClick on \"Load\" to load it."205 return206 207 if selected_model == 'None':208 yield "No model selected"209 else:210 try:211 yield f"Loading `{selected_model}`..."212 unload_model()213 if selected_model != '':214 shared.model, shared.tokenizer = load_model(selected_model, loader)215 216 if shared.model is not None:217 output = f"Successfully loaded `{selected_model}`."218 219 settings = get_model_metadata(selected_model)220 if 'instruction_template' in settings:221 output += '\n\nIt seems to be an instruction-following model with template "{}". In the chat tab, instruct or chat-instruct modes should be used.'.format(settings['instruction_template'])222 223 yield output224 else:225 yield f"Failed to load `{selected_model}`."226 except:227 exc = traceback.format_exc()228 logger.error('Failed to load the model.')229 print(exc)230 yield exc.replace('\n', '\n\n')231 232 233def load_lora_wrapper(selected_loras):234 yield ("Applying the following LoRAs to {}:\n\n{}".format(shared.model_name, '\n'.join(selected_loras)))235 add_lora_to_model(selected_loras)236 yield ("Successfuly applied the LoRAs")237 238 239def download_model_wrapper(repo_id, specific_file, progress=gr.Progress(), return_links=False, check=False):240 try:241 progress(0.0)242 downloader = importlib.import_module("download-model").ModelDownloader()243 model, branch = downloader.sanitize_model_and_branch_names(repo_id, None)244 yield ("Getting the download links from Hugging Face")245 links, sha256, is_lora, is_llamacpp = downloader.get_download_links_from_huggingface(model, branch, text_only=False, specific_file=specific_file)246 if return_links:247 yield '\n\n'.join([f"`{Path(link).name}`" for link in links])248 return249 250 yield ("Getting the output folder")251 base_folder = shared.args.lora_dir if is_lora else shared.args.model_dir252 output_folder = downloader.get_output_folder(model, branch, is_lora, is_llamacpp=is_llamacpp, base_folder=base_folder)253 if check:254 progress(0.5)255 yield ("Checking previously downloaded files")256 downloader.check_model_files(model, branch, links, sha256, output_folder)257 progress(1.0)258 else:259 yield (f"Downloading file{'s' if len(links) > 1 else ''} to `{output_folder}/`")260 downloader.download_model_files(model, branch, links, sha256, output_folder, progress_bar=progress, threads=4, is_llamacpp=is_llamacpp)261 yield ("Done!")262 except:263 progress(1.0)264 yield traceback.format_exc().replace('\n', '\n\n')265 266 267def update_truncation_length(current_length, state):268 if 'loader' in state:269 if state['loader'].lower().startswith('exllama'):270 return state['max_seq_len']271 elif state['loader'] in ['llama.cpp', 'llamacpp_HF', 'ctransformers']:272 return state['n_ctx']273 274 return current_length275