Reorganize model loading UI completely (#2720)

.gitignore

@@ -8,6 +8,7 @@ extensions/multimodal/pipelines
 logs
 loras
 models
+presets
 repositories
 softprompts
 torch-dumps

README.md

@@ -211,6 +211,12 @@ Optionally, you can use the following command-line flags:
 | `--extensions EXTENSIONS [EXTENSIONS ...]` | The list of extensions to load. If you want to load more than one extension, write the names separated by spaces. |
 | `--verbose`                                | Print the prompts to the terminal. |
 
+#### Model loader
+
+| Flag                                       | Description |
+|--------------------------------------------|-------------|
+| `--loader LOADER`                          | Choose the model loader manually, otherwise, it will get autodetected. Valid options: autogptq, gptq-for-llama, transformers, llamacpp, rwkv, flexgen |
+
 #### Accelerate/transformers
 
 | Flag                                        | Description |
@@ -265,7 +271,6 @@ Optionally, you can use the following command-line flags:
 
 | Flag                      | Description |
 |---------------------------|-------------|
-| `--gptq-for-llama` | Use GPTQ-for-LLaMa to load the GPTQ model instead of AutoGPTQ. |
 | `--wbits WBITS`           | Load a pre-quantized model with specified precision in bits. 2, 3, 4 and 8 are supported. |
 | `--model_type MODEL_TYPE` | Model type of pre-quantized model. Currently LLaMA, OPT, and GPT-J are supported. |
 | `--groupsize GROUPSIZE`   | Group size. |
@@ -280,7 +285,6 @@ Optionally, you can use the following command-line flags:
 
 | Flag             | Description |
 |------------------|-------------|
-| `--flexgen`                       | Enable the use of FlexGen offloading. |
 | `--percent PERCENT [PERCENT ...]` | FlexGen: allocation percentages. Must be 6 numbers separated by spaces (default: 0, 100, 100, 0, 100, 0). |
 | `--compress-weight`               | FlexGen: Whether to compress weight (default: False).|
 | `--pin-weight [PIN_WEIGHT]`       | FlexGen: whether to pin weights (setting this to False reduces CPU memory by 20%). |

docs/FlexGen.md

@@ -21,13 +21,13 @@ The output will be saved to `models/opt-1.3b-np/`.
 The basic command is the following:
 
 ```
-python server.py --model opt-1.3b  --flexgen
+python server.py --model opt-1.3b  --loader flexgen
 ```
 
 For large models, the RAM usage may be too high and your computer may freeze. If that happens, you can try this:
 
 ```
-python server.py --model opt-1.3b  --flexgen --compress-weight
+python server.py --model opt-1.3b  --loader flexgen --compress-weight
 ```
 
 With this second command, I was able to run both OPT-6.7b and OPT-13B with **2GB VRAM**, and the speed was good in both cases.
@@ -35,7 +35,7 @@ With this second command, I was able to run both OPT-6.7b and OPT-13B with **2GB
 You can also manually set the offload strategy with
 
 ```
-python server.py --model opt-1.3b  --flexgen --percent 0 100 100 0 100 0
+python server.py --model opt-1.3b  --loader flexgen --percent 0 100 100 0 100 0
 ```
 
 where the six numbers after `--percent` are:
@@ -55,8 +55,8 @@ You should typically only change the first two numbers. If their sum is less tha
 
 In my experiments with OPT-30B using a RTX 3090 on Linux, I have obtained these results:
 
-* `--flexgen --compress-weight --percent 0 100 100 0 100 0`: 0.99 seconds per token.
+* `--loader flexgen --compress-weight --percent 0 100 100 0 100 0`: 0.99 seconds per token.
-* `--flexgen --compress-weight --percent 100 0 100 0 100 0`: 0.765 seconds per token.
+* `--loader flexgen --compress-weight --percent 100 0 100 0 100 0`: 0.765 seconds per token.
 
 ## Limitations
 

extensions/api/blocking_api.py

@@ -7,10 +7,11 @@ from modules import shared
 from modules.chat import generate_chat_reply
 from modules.LoRA import add_lora_to_model
 from modules.models import load_model, unload_model
+from modules.models_settings import (get_model_settings_from_yamls,
+                                     update_model_parameters)
 from modules.text_generation import (encode, generate_reply,
                                      stop_everything_event)
 from modules.utils import get_available_models
-from server import get_model_specific_settings, update_model_parameters
 
 
 def get_model_info():
@@ -22,6 +23,7 @@ def get_model_info():
         'shared.args': vars(shared.args),
     }
 
+
 class Handler(BaseHTTPRequestHandler):
     def do_GET(self):
         if self.path == '/api/v1/model':
@@ -126,7 +128,7 @@ class Handler(BaseHTTPRequestHandler):
                 shared.model_name = model_name
                 unload_model()
 
-                model_settings = get_model_specific_settings(shared.model_name)
+                model_settings = get_model_settings_from_yamls(shared.model_name)
                 shared.settings.update(model_settings)
                 update_model_parameters(model_settings, initial=True)
 

modules/LoRA.py

@@ -78,7 +78,6 @@ def add_lora_to_model(lora_names):
                     params['device_map'] = {'': 0}
 
             shared.model = PeftModel.from_pretrained(shared.model, Path(f"{shared.args.lora_dir}/{lora_names[0]}"), adapter_name=lora_names[0], **params)
-
             for lora in lora_names[1:]:
                 shared.model.load_adapter(Path(f"{shared.args.lora_dir}/{lora}"), lora)
 

modules/evaluate.py

@@ -8,8 +8,9 @@ from tqdm import tqdm
 
 from modules import shared
 from modules.models import load_model, unload_model
+from modules.models_settings import (get_model_settings_from_yamls,
+                                     update_model_parameters)
 from modules.text_generation import encode
-from server import get_model_specific_settings, update_model_parameters
 
 
 def load_past_evaluations():
@@ -66,7 +67,7 @@ def calculate_perplexity(models, input_dataset, stride, _max_length):
         if model != 'current model':
             try:
                 yield cumulative_log + f"Loading {model}...\n\n"
-                model_settings = get_model_specific_settings(model)
+                model_settings = get_model_settings_from_yamls(model)
                 shared.settings.update(model_settings)  # hijacking the interface defaults
                 update_model_parameters(model_settings)  # hijacking the command-line arguments
                 shared.model_name = model

modules/github.py

@@ -1,6 +1,7 @@
 import os
 import subprocess
 
+
 def clone_or_pull_repository(github_url):
     repository_folder = "extensions"
     repo_name = github_url.split("/")[-1].split(".")[0]

modules/loaders.py

@@ -0,0 +1,86 @@
+import functools
+
+import gradio as gr
+
+from modules import shared
+
+loaders_and_params = {
+    'AutoGPTQ': [
+        'triton',
+        'no_inject_fused_attention',
+        'no_inject_fused_mlp',
+        'wbits',
+        'groupsize',
+        'desc_act',
+        'gpu_memory',
+        'cpu_memory',
+        'cpu',
+        'disk',
+        'auto_devices',
+        'trust_remote_code',
+        'autogptq_info',
+    ],
+    'GPTQ-for-LLaMa': [
+        'wbits',
+        'groupsize',
+        'model_type',
+        'pre_layer',
+        'gptq_for_llama_info',
+    ],
+    'llama.cpp': [
+        'n_ctx',
+        'n_gpu_layers',
+        'n_batch',
+        'threads',
+        'no_mmap',
+        'mlock',
+        'llama_cpp_seed',
+    ],
+    'Transformers': [
+        'cpu_memory',
+        'gpu_memory',
+        'trust_remote_code',
+        'load_in_8bit',
+        'bf16',
+        'cpu',
+        'disk',
+        'auto_devices',
+        'load_in_4bit',
+        'use_double_quant',
+        'quant_type',
+        'compute_dtype',
+        'trust_remote_code',
+    ],
+}
+
+
+def get_gpu_memory_keys():
+    return [k for k in shared.gradio if k.startswith('gpu_memory')]
+
+
+@functools.cache
+def get_all_params():
+    all_params = set()
+    for k in loaders_and_params:
+        for el in loaders_and_params[k]:
+            all_params.add(el)
+
+    if 'gpu_memory' in all_params:
+        all_params.remove('gpu_memory')
+        for k in get_gpu_memory_keys():
+            all_params.add(k)
+
+    return sorted(all_params)
+
+
+def make_loader_params_visible(loader):
+    params = []
+    all_params = get_all_params()
+    if loader in loaders_and_params:
+        params = loaders_and_params[loader]
+
+        if 'gpu_memory' in params:
+            params.remove('gpu_memory')
+            params += get_gpu_memory_keys()
+
+    return [gr.update(visible=True) if k in params else gr.update(visible=False) for k in all_params]

modules/models.py

@@ -14,6 +14,7 @@ from transformers import (AutoConfig, AutoModel, AutoModelForCausalLM,
 import modules.shared as shared
 from modules import llama_attn_hijack, sampler_hijack
 from modules.logging_colors import logger
+from modules.models_settings import infer_loader
 
 transformers.logging.set_verbosity_error()
 
@@ -36,62 +37,31 @@ if shared.args.deepspeed:
 sampler_hijack.hijack_samplers()
 
 
-# Some models require special treatment in various parts of the code.
+def load_model(model_name, loader=None):
-# This function detects those models
-def find_model_type(model_name):
-    path_to_model = Path(f'{shared.args.model_dir}/{model_name}')
-    if not path_to_model.exists():
-        return 'None'
-
-    model_name_lower = model_name.lower()
-    if re.match('.*rwkv.*\.pth', model_name_lower):
-        return 'rwkv'
-    elif len(list(path_to_model.glob('*ggml*.bin'))) > 0:
-        return 'llamacpp'
-    elif re.match('.*ggml.*\.bin', model_name_lower):
-        return 'llamacpp'
-    elif 'chatglm' in model_name_lower:
-        return 'chatglm'
-    elif 'galactica' in model_name_lower:
-        return 'galactica'
-    elif 'llava' in model_name_lower:
-        return 'llava'
-    elif 'oasst' in model_name_lower:
-        return 'oasst'
-    elif any((k in model_name_lower for k in ['gpt4chan', 'gpt-4chan'])):
-        return 'gpt4chan'
-    else:
-        config = AutoConfig.from_pretrained(path_to_model, trust_remote_code=shared.args.trust_remote_code)
-        # Not a "catch all", but fairly accurate
-        if config.to_dict().get("is_encoder_decoder", False):
-            return 'HF_seq2seq'
-        else:
-            return 'HF_generic'
-
-
-def load_model(model_name):
     logger.info(f"Loading {model_name}...")
     t0 = time.time()
 
-    shared.model_type = find_model_type(model_name)
+    shared.is_seq2seq = False
-    if shared.model_type == 'None':
+    load_func_map = {
+        'Transformers': huggingface_loader,
+        'AutoGPTQ': AutoGPTQ_loader,
+        'GPTQ-for-LLaMa': GPTQ_loader,
+        'llama.cpp': llamacpp_loader,
+        'FlexGen': flexgen_loader,
+        'RWKV': RWKV_loader
+    }
+
+    if loader is None:
+        if shared.args.loader is not None:
+            loader = shared.args.loader
+        else:
+            loader = infer_loader(model_name)
+            if loader is None:
                 logger.error('The path to the model does not exist. Exiting.')
                 return None, None
 
-    if shared.args.gptq_for_llama:
+    shared.args.loader = loader
-        load_func = GPTQ_loader
+    output = load_func_map[loader](model_name)
-    elif Path(f'{shared.args.model_dir}/{model_name}/quantize_config.json').exists() or shared.args.wbits > 0:
-        load_func = AutoGPTQ_loader
-    elif shared.model_type == 'llamacpp':
-        load_func = llamacpp_loader
-    elif shared.model_type == 'rwkv':
-        load_func = RWKV_loader
-    elif shared.args.flexgen:
-        load_func = flexgen_loader
-    else:
-        load_func = huggingface_loader
-
-    output = load_func(model_name)
     if type(output) is tuple:
         model, tokenizer = output
     else:
@@ -111,11 +81,11 @@ def load_model(model_name):
 
 def load_tokenizer(model_name, model):
     tokenizer = None
-    if shared.model_type == 'gpt4chan' and Path(f"{shared.args.model_dir}/gpt-j-6B/").exists():
+    if any(s in model_name.lower() for s in ['gpt-4chan', 'gpt4chan']) 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 or "LlamaGPTQForCausalLM" in str(type(model)):
         # Try to load an universal LLaMA tokenizer
-        if shared.model_type not in ['llava', 'oasst']:
+        if any(s in shared.model_name.lower() for s in ['llava', 'oasst']):
             for p in [Path(f"{shared.args.model_dir}/llama-tokenizer/"), Path(f"{shared.args.model_dir}/oobabooga_llama-tokenizer/")]:
                 if p.exists():
                     logger.info(f"Loading the universal LLaMA tokenizer from {p}...")
@@ -140,10 +110,14 @@ def load_tokenizer(model_name, model):
 
 
 def huggingface_loader(model_name):
-    if shared.model_type == 'chatglm':
+    path_to_model = Path(f'{shared.args.model_dir}/{model_name}')
+    if 'chatglm' in model_name.lower():
         LoaderClass = AutoModel
-    elif shared.model_type == 'HF_seq2seq':
+    else:
+        config = AutoConfig.from_pretrained(path_to_model, trust_remote_code=shared.args.trust_remote_code)
+        if config.to_dict().get("is_encoder_decoder", False):
             LoaderClass = AutoModelForSeq2SeqLM
+            shared.is_seq2seq = True
         else:
             LoaderClass = AutoModelForCausalLM
 

modules/models_settings.py

@@ -0,0 +1,134 @@
+import re
+from pathlib import Path
+
+import yaml
+
+from modules import shared, ui
+
+
+def get_model_settings_from_yamls(model):
+    settings = shared.model_config
+    model_settings = {}
+    for pat in settings:
+        if re.match(pat.lower(), model.lower()):
+            for k in settings[pat]:
+                model_settings[k] = settings[pat][k]
+
+    return model_settings
+
+
+def infer_loader(model_name):
+    path_to_model = Path(f'{shared.args.model_dir}/{model_name}')
+    model_settings = get_model_settings_from_yamls(model_name)
+    if not path_to_model.exists():
+        loader = None
+    elif Path(f'{shared.args.model_dir}/{model_name}/quantize_config.json').exists() or ('wbits' in model_settings and type(model_settings['wbits']) is int and model_settings['wbits'] > 0):
+        loader = 'AutoGPTQ'
+    elif len(list(path_to_model.glob('*ggml*.bin'))) > 0:
+        loader = 'llama.cpp'
+    elif re.match('.*ggml.*\.bin', model_name.lower()):
+        loader = 'llama.cpp'
+    elif re.match('.*rwkv.*\.pth', model_name.lower()):
+        loader = 'RWKV'
+    elif shared.args.flexgen:
+        loader = 'FlexGen'
+    else:
+        loader = 'Transformers'
+
+    return loader
+
+
+# UI: update the command-line arguments based on the interface values
+def update_model_parameters(state, initial=False):
+    elements = ui.list_model_elements()  # the names of the parameters
+    gpu_memories = []
+
+    for i, element in enumerate(elements):
+        if element not in state:
+            continue
+
+        value = state[element]
+        if element.startswith('gpu_memory'):
+            gpu_memories.append(value)
+            continue
+
+        if initial and vars(shared.args)[element] != vars(shared.args_defaults)[element]:
+            continue
+
+        # Setting null defaults
+        if element in ['wbits', 'groupsize', 'model_type'] and value == 'None':
+            value = vars(shared.args_defaults)[element]
+        elif element in ['cpu_memory'] and value == 0:
+            value = vars(shared.args_defaults)[element]
+
+        # Making some simple conversions
+        if element in ['wbits', 'groupsize', 'pre_layer']:
+            value = int(value)
+        elif element == 'cpu_memory' and value is not None:
+            value = f"{value}MiB"
+
+        if element in ['pre_layer']:
+            value = [value] if value > 0 else None
+
+        setattr(shared.args, element, value)
+
+    found_positive = False
+    for i in gpu_memories:
+        if i > 0:
+            found_positive = True
+            break
+
+    if not (initial and vars(shared.args)['gpu_memory'] != vars(shared.args_defaults)['gpu_memory']):
+        if found_positive:
+            shared.args.gpu_memory = [f"{i}MiB" for i in gpu_memories]
+        else:
+            shared.args.gpu_memory = None
+
+
+# UI: update the state variable with the model settings
+def apply_model_settings_to_state(model, state):
+    model_settings = get_model_settings_from_yamls(model)
+    if 'loader' not in model_settings:
+        loader = infer_loader(model)
+        if 'wbits' in model_settings and type(model_settings['wbits']) is int and model_settings['wbits'] > 0:
+            loader = 'AutoGPTQ'
+
+        # If the user is using an alternative GPTQ loader, let them keep using it
+        if not (loader == 'AutoGPTQ' and state['loader'] in ['GPTQ-for-LLaMa', 'exllama']):
+            state['loader'] = loader
+
+    for k in model_settings:
+        if k in state:
+            state[k] = model_settings[k]
+
+    return state
+
+
+# Save the settings for this model to models/config-user.yaml
+def save_model_settings(model, state):
+    if model == 'None':
+        yield ("Not saving the settings because no model is loaded.")
+        return
+
+    with Path(f'{shared.args.model_dir}/config-user.yaml') as p:
+        if p.exists():
+            user_config = yaml.safe_load(open(p, 'r').read())
+        else:
+            user_config = {}
+
+        model_regex = model + '$'  # For exact matches
+        for _dict in [user_config, shared.model_config]:
+            if model_regex not in _dict:
+                _dict[model_regex] = {}
+
+        if model_regex not in user_config:
+            user_config[model_regex] = {}
+
+        for k in ui.list_model_elements():
+            user_config[model_regex][k] = state[k]
+            shared.model_config[model_regex][k] = state[k]
+
+        with open(p, 'w') as f:
+            f.write(yaml.dump(user_config, sort_keys=False))
+
+        yield (f"Settings for {model} saved to {p}")

modules/presets.py

@@ -52,4 +52,3 @@ def load_preset_for_ui(name, state):
 def generate_preset_yaml(state):
     data = {k: state[k] for k in ['do_sample', 'temperature', 'top_p', 'typical_p', 'epsilon_cutoff', 'eta_cutoff', 'repetition_penalty', 'encoder_repetition_penalty', 'top_k', 'min_length', 'no_repeat_ngram_size', 'num_beams', 'penalty_alpha', 'length_penalty', 'early_stopping', 'mirostat_mode', 'mirostat_tau', 'mirostat_eta', 'tfs', 'top_a']}
     return yaml.dump(data, sort_keys=False)
-

modules/shared.py

@@ -10,7 +10,6 @@ generation_lock = None
 model = None
 tokenizer = None
 model_name = "None"
-model_type = None
 lora_names = []
 
 # Chat variables
@@ -97,6 +96,9 @@ parser.add_argument('--settings', type=str, help='Load the default interface set
 parser.add_argument('--extensions', type=str, nargs="+", help='The list of extensions to load. If you want to load more than one extension, write the names separated by spaces.')
 parser.add_argument('--verbose', action='store_true', help='Print the prompts to the terminal.')
 
+# Model loader
+parser.add_argument('--loader', type=str, help='Choose the model loader manually, otherwise, it will get autodetected. Valid options: autogptq, gptq-for-llama, transformers, llamacpp, rwkv, flexgen')
+
 # Accelerate/transformers
 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.')
@@ -139,7 +141,7 @@ parser.add_argument('--warmup_autotune', action='store_true', help='(triton) Ena
 parser.add_argument('--fused_mlp', action='store_true', help='(triton) Enable fused mlp.')
 
 # AutoGPTQ
-parser.add_argument('--gptq-for-llama', action='store_true', help='Use GPTQ-for-LLaMa to load the GPTQ model instead of AutoGPTQ.')
+parser.add_argument('--gptq-for-llama', action='store_true', help='DEPRECATED')
 parser.add_argument('--autogptq', action='store_true', help='DEPRECATED')
 parser.add_argument('--triton', action='store_true', help='Use triton.')
 parser.add_argument('--no_inject_fused_attention', action='store_true', help='Do not use fused attention (lowers VRAM requirements).')
@@ -147,7 +149,7 @@ parser.add_argument('--no_inject_fused_mlp', action='store_true', help='Triton m
 parser.add_argument('--desc_act', action='store_true', help='For models that don\'t have a quantize_config.json, this parameter is used to define whether to set desc_act or not in BaseQuantizeConfig.')
 
 # FlexGen
-parser.add_argument('--flexgen', action='store_true', help='Enable the use of FlexGen offloading.')
+parser.add_argument('--flexgen', action='store_true', help='DEPRECATED')
 parser.add_argument('--percent', type=int, nargs="+", default=[0, 100, 100, 0, 100, 0], help='FlexGen: allocation percentages. Must be 6 numbers separated by spaces (default: 0, 100, 100, 0, 100, 0).')
 parser.add_argument("--compress-weight", action="store_true", help="FlexGen: activate weight compression.")
 parser.add_argument("--pin-weight", type=str2bool, nargs="?", const=True, default=True, help="FlexGen: whether to pin weights (setting this to False reduces CPU memory by 20%%).")
@@ -184,7 +186,14 @@ args_defaults = parser.parse_args([])
 
 # Deprecation warnings
 if args.autogptq:
-    logger.warning('--autogptq has been deprecated and will be removed soon. AutoGPTQ is now used by default for GPTQ models.')
+    logger.warning('--autogptq has been deprecated and will be removed soon. Use --loader autogptq instead.')
+    args.loader = 'autogptq'
+if args.gptq_for_llama:
+    logger.warning('--gptq-for-llama has been deprecated and will be removed soon. Use --loader gptq-for-llama instead.')
+    args.loader = 'gptq-for-llama'
+if args.flexgen:
+    logger.warning('--flexgen has been deprecated and will be removed soon. Use --loader flexgen instead.')
+    args.loader = 'FlexGen'
 
 # Security warnings
 if args.trust_remote_code:
@@ -193,6 +202,22 @@ if args.share:
     logger.warning("The gradio \"share link\" feature uses a proprietary executable to create a reverse tunnel. Use it with care.")
 
 
+def fix_loader_name(name):
+    name = name.lower()
+    if name in ['llamacpp', 'llama.cpp', 'llama-cpp', 'llama cpp']:
+        return 'llama.cpp'
+    elif name in ['transformers', 'huggingface', 'hf', 'hugging_face', 'hugging face']:
+        return 'Transformers'
+    elif name in ['autogptq', 'auto-gptq', 'auto_gptq', 'auto gptq']:
+        return 'AutoGPTQ'
+    elif name in ['gptq-for-llama', 'gptqforllama', 'gptqllama', 'gptq for llama', 'gptq_for_llama']:
+        return 'GPTQ-for-LLaMa'
+
+
+if args.loader is not None:
+    args.loader = fix_loader_name(args.loader)
+
+
 def add_extension(name):
     if args.extensions is None:
         args.extensions = [name]

modules/text_generation.py

@@ -31,7 +31,7 @@ def get_max_prompt_length(state):
 
 
 def encode(prompt, add_special_tokens=True, add_bos_token=True, truncation_length=None):
-    if shared.model_type in ['rwkv', 'llamacpp']:
+    if shared.model.__class__.__name__ in ['LlamaCppModel', 'RWKVModel']:
         input_ids = shared.tokenizer.encode(str(prompt))
         input_ids = np.array(input_ids).reshape(1, len(input_ids))
         return input_ids
@@ -51,7 +51,7 @@ def encode(prompt, add_special_tokens=True, add_bos_token=True, truncation_lengt
     if truncation_length is not None:
         input_ids = input_ids[:, -truncation_length:]
 
-    if shared.model_type in ['rwkv', 'llamacpp'] or shared.args.cpu:
+    if shared.model.__class__.__name__ in ['LlamaCppModel', 'RWKVModel'] or shared.args.cpu:
         return input_ids
     elif shared.args.flexgen:
         return input_ids.numpy()
@@ -99,7 +99,7 @@ def fix_galactica(s):
 
 
 def get_reply_from_output_ids(output_ids, input_ids, original_question, state, is_chat=False):
-    if shared.model_type == 'HF_seq2seq':
+    if shared.is_seq2seq:
         reply = decode(output_ids, state['skip_special_tokens'])
     else:
         new_tokens = len(output_ids) - len(input_ids[0])
@@ -117,7 +117,7 @@ def get_reply_from_output_ids(output_ids, input_ids, original_question, state, i
 
 
 def formatted_outputs(reply, model_name):
-    if shared.model_type == 'gpt4chan':
+    if any(s in model_name for s in ['gpt-4chan', 'gpt4chan']):
         reply = fix_gpt4chan(reply)
         return reply, generate_4chan_html(reply)
     else:
@@ -142,7 +142,7 @@ def stop_everything_event():
 
 def generate_reply_wrapper(question, state, eos_token=None, stopping_strings=None):
     for reply in generate_reply(question, state, eos_token, stopping_strings, is_chat=False):
-        if shared.model_type not in ['HF_seq2seq']:
+        if not shared.is_seq2seq:
             reply = question + reply
 
         yield formatted_outputs(reply, shared.model_name)
@@ -157,7 +157,7 @@ def _generate_reply(question, state, eos_token=None, stopping_strings=None, is_c
             yield ''
             return
 
-        if shared.model_type in ['rwkv', 'llamacpp']:
+        if shared.model.__class__.__name__ in ['LlamaCppModel', 'RWKVModel']:
             generate_func = generate_reply_custom
         elif shared.args.flexgen:
             generate_func = generate_reply_flexgen
@@ -240,7 +240,7 @@ def generate_reply_HF(question, original_question, seed, state, eos_token=None,
 
     t0 = time.time()
     try:
-        if not is_chat and shared.model_type != 'HF_seq2seq':
+        if not is_chat and not shared.is_seq2seq:
             yield ''
 
         # Generate the entire reply at once.
@@ -276,7 +276,7 @@ def generate_reply_HF(question, original_question, seed, state, eos_token=None,
     finally:
         t1 = time.time()
         original_tokens = len(original_input_ids[0])
-        new_tokens = len(output) - (original_tokens if shared.model_type != 'HF_seq2seq' else 0)
+        new_tokens = len(output) - (original_tokens if not shared.is_seq2seq else 0)
         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
 
@@ -287,7 +287,7 @@ def generate_reply_custom(question, original_question, seed, state, eos_token=No
     for k in ['temperature', 'top_p', 'top_k', 'repetition_penalty']:
         generate_params[k] = state[k]
 
-    if shared.model_type == 'llamacpp':
+    if shared.model.__class__.__name__ in ['LlamaCppModel']:
         for k in ['mirostat_mode', 'mirostat_tau', 'mirostat_eta']:
             generate_params[k] = state[k]
 
@@ -381,6 +381,6 @@ def generate_reply_flexgen(question, original_question, seed, state, eos_token=N
     finally:
         t1 = time.time()
         original_tokens = len(original_input_ids[0])
-        new_tokens = len(output) - (original_tokens if shared.model_type != 'HF_seq2seq' else 0)
+        new_tokens = len(output) - (original_tokens if not shared.is_seq2seq else 0)
         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/ui.py

@@ -30,7 +30,7 @@ theme = gr.themes.Default(
 
 
 def list_model_elements():
-    elements = ['cpu_memory', 'auto_devices', 'disk', 'cpu', 'bf16', 'load_in_8bit', 'trust_remote_code', 'load_in_4bit', 'compute_dtype', 'quant_type', 'use_double_quant', 'gptq_for_llama', 'wbits', 'groupsize', 'model_type', 'pre_layer', 'triton', 'desc_act', 'no_inject_fused_attention', 'no_inject_fused_mlp', 'threads', 'n_batch', 'no_mmap', 'mlock', 'n_gpu_layers', 'n_ctx', 'llama_cpp_seed']
+    elements = ['loader', 'cpu_memory', 'auto_devices', 'disk', 'cpu', 'bf16', 'load_in_8bit', 'trust_remote_code', 'load_in_4bit', 'compute_dtype', 'quant_type', 'use_double_quant', 'wbits', 'groupsize', 'model_type', 'pre_layer', 'triton', 'desc_act', 'no_inject_fused_attention', 'no_inject_fused_mlp', 'threads', 'n_batch', 'no_mmap', 'mlock', 'n_gpu_layers', 'n_ctx', 'llama_cpp_seed']
     for i in range(torch.cuda.device_count()):
         elements.append(f'gpu_memory_{i}')
 

server.py

@@ -43,17 +43,21 @@ import yaml
 from PIL import Image
 
 import modules.extensions as extensions_module
-from modules import chat, presets, shared, training, ui, utils
+from modules import chat, loaders, presets, shared, training, ui, utils
 from modules.extensions import apply_extensions
 from modules.github import clone_or_pull_repository
 from modules.html_generator import chat_html_wrapper
 from modules.LoRA import add_lora_to_model
 from modules.models import load_model, unload_model
+from modules.models_settings import (apply_model_settings_to_state,
+                                     get_model_settings_from_yamls,
+                                     save_model_settings,
+                                     update_model_parameters)
 from modules.text_generation import (generate_reply_wrapper,
                                      get_encoded_length, stop_everything_event)
 
 
-def load_model_wrapper(selected_model, autoload=False):
+def load_model_wrapper(selected_model, loader, autoload=False):
     if not autoload:
         yield f"The settings for {selected_model} have been updated.\nClick on \"Load the model\" to load it."
         return
@@ -66,9 +70,12 @@ def load_model_wrapper(selected_model, autoload=False):
             shared.model_name = selected_model
             unload_model()
             if selected_model != '':
-                shared.model, shared.tokenizer = load_model(shared.model_name)
+                shared.model, shared.tokenizer = load_model(shared.model_name, loader)
 
+            if shared.model is not None:
                 yield f"Successfully loaded {selected_model}"
+            else:
+                yield f"Failed to load {selected_model}."
         except:
             yield traceback.format_exc()
 
@@ -144,103 +151,6 @@ def download_model_wrapper(repo_id):
         yield traceback.format_exc()
 
 
-# Update the command-line arguments based on the interface values
-def update_model_parameters(state, initial=False):
-    elements = ui.list_model_elements()  # the names of the parameters
-    gpu_memories = []
-
-    for i, element in enumerate(elements):
-        if element not in state:
-            continue
-
-        value = state[element]
-        if element.startswith('gpu_memory'):
-            gpu_memories.append(value)
-            continue
-
-        if initial and vars(shared.args)[element] != vars(shared.args_defaults)[element]:
-            continue
-
-        # Setting null defaults
-        if element in ['wbits', 'groupsize', 'model_type'] and value == 'None':
-            value = vars(shared.args_defaults)[element]
-        elif element in ['cpu_memory'] and value == 0:
-            value = vars(shared.args_defaults)[element]
-
-        # Making some simple conversions
-        if element in ['wbits', 'groupsize', 'pre_layer']:
-            value = int(value)
-        elif element == 'cpu_memory' and value is not None:
-            value = f"{value}MiB"
-
-        if element in ['pre_layer']:
-            value = [value] if value > 0 else None
-
-        setattr(shared.args, element, value)
-
-    found_positive = False
-    for i in gpu_memories:
-        if i > 0:
-            found_positive = True
-            break
-
-    if not (initial and vars(shared.args)['gpu_memory'] != vars(shared.args_defaults)['gpu_memory']):
-        if found_positive:
-            shared.args.gpu_memory = [f"{i}MiB" for i in gpu_memories]
-        else:
-            shared.args.gpu_memory = None
-
-
-def get_model_specific_settings(model):
-    settings = shared.model_config
-    model_settings = {}
-
-    for pat in settings:
-        if re.match(pat.lower(), model.lower()):
-            for k in settings[pat]:
-                model_settings[k] = settings[pat][k]
-
-    return model_settings
-
-
-def load_model_specific_settings(model, state):
-    model_settings = get_model_specific_settings(model)
-    for k in model_settings:
-        if k in state:
-            state[k] = model_settings[k]
-
-    return state
-
-
-def save_model_settings(model, state):
-    if model == 'None':
-        yield ("Not saving the settings because no model is loaded.")
-        return
-
-    with Path(f'{shared.args.model_dir}/config-user.yaml') as p:
-        if p.exists():
-            user_config = yaml.safe_load(open(p, 'r').read())
-        else:
-            user_config = {}
-
-        model_regex = model + '$'  # For exact matches
-        for _dict in [user_config, shared.model_config]:
-            if model_regex not in _dict:
-                _dict[model_regex] = {}
-
-        if model_regex not in user_config:
-            user_config[model_regex] = {}
-
-        for k in ui.list_model_elements():
-            user_config[model_regex][k] = state[k]
-            shared.model_config[model_regex][k] = state[k]
-
-        with open(p, 'w') as f:
-            f.write(yaml.dump(user_config, sort_keys=False))
-
-        yield (f"Settings for {model} saved to {p}")
-
-
 def create_model_menus():
     # Finding the default values for the GPU and CPU memories
     total_mem = []
@@ -283,88 +193,70 @@ def create_model_menus():
 
     with gr.Row():
         with gr.Column():
+            shared.gradio['loader'] = gr.Dropdown(label="Model loader", choices=["Transformers", "AutoGPTQ", "GPTQ-for-LLaMa", "llama.cpp"], value=None)
             with gr.Box():
-                gr.Markdown('Transformers')
                 with gr.Row():
                     with gr.Column():
                         for i in range(len(total_mem)):
                             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])
 
                         shared.gradio['cpu_memory'] = gr.Slider(label="cpu-memory in MiB", maximum=total_cpu_mem, value=default_cpu_mem)
-
-                    with gr.Column():
-                        shared.gradio['auto_devices'] = gr.Checkbox(label="auto-devices", value=shared.args.auto_devices)
-                        shared.gradio['disk'] = gr.Checkbox(label="disk", value=shared.args.disk)
-                        shared.gradio['cpu'] = gr.Checkbox(label="cpu", value=shared.args.cpu)
-                        shared.gradio['bf16'] = gr.Checkbox(label="bf16", value=shared.args.bf16)
-                        shared.gradio['load_in_8bit'] = gr.Checkbox(label="load-in-8bit", value=shared.args.load_in_8bit)
-                        shared.gradio['trust_remote_code'] = gr.Checkbox(label="trust-remote-code", value=shared.args.trust_remote_code, info='Make sure to inspect the .py files inside the model folder before loading it with this option enabled.')
-
-            with gr.Box():
-                gr.Markdown('Transformers 4-bit')
-                with gr.Row():
-                    with gr.Column():
-                        shared.gradio['load_in_4bit'] = gr.Checkbox(label="load-in-4bit", value=shared.args.load_in_4bit)
-                        shared.gradio['use_double_quant'] = gr.Checkbox(label="use_double_quant", value=shared.args.use_double_quant)
-
-                    with gr.Column():
                         shared.gradio['compute_dtype'] = gr.Dropdown(label="compute_dtype", choices=["bfloat16", "float16", "float32"], value=shared.args.compute_dtype)
                         shared.gradio['quant_type'] = gr.Dropdown(label="quant_type", choices=["nf4", "fp4"], value=shared.args.quant_type)
-
-            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.')
-            shared.gradio['custom_model_menu'] = gr.Textbox(label="Download custom 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")
-            shared.gradio['download_model_button'] = gr.Button("Download")
-
-        with gr.Column():
-            with gr.Box():
-                with gr.Row():
-                    with gr.Column():
-                        gr.Markdown('GPTQ')
-                        shared.gradio['triton'] = gr.Checkbox(label="triton", value=shared.args.triton)
-                        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. Disable if running low on VRAM.')
-                        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.')
-                        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.')
-                        shared.gradio['gptq_for_llama'] = gr.Checkbox(label="gptq-for-llama", value=shared.args.gptq_for_llama, info='Use GPTQ-for-LLaMa loader instead of AutoGPTQ. pre_layer should be used for CPU offloading instead of gpu-memory.')
-
-                    with gr.Column():
-                        with gr.Row():
-                            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")
-                            shared.gradio['groupsize'] = gr.Dropdown(label="groupsize", choices=["None", 32, 64, 128, 1024], value=shared.args.groupsize if shared.args.groupsize > 0 else "None")
-
-                        shared.gradio['model_type'] = gr.Dropdown(label="model_type", choices=["None", "llama", "opt", "gptj"], value=shared.args.model_type or "None")
-                        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)
-
-            with gr.Box():
-                gr.Markdown('llama.cpp')
-                with gr.Row():
-                    with gr.Column():
                         shared.gradio['threads'] = gr.Slider(label="threads", minimum=0, step=1, maximum=32, value=shared.args.threads)
                         shared.gradio['n_batch'] = gr.Slider(label="n_batch", minimum=1, maximum=2048, value=shared.args.n_batch)
                         shared.gradio['n_gpu_layers'] = gr.Slider(label="n-gpu-layers", minimum=0, maximum=128, value=shared.args.n_gpu_layers)
                         shared.gradio['n_ctx'] = gr.Slider(minimum=0, maximum=8192, step=1, label="n_ctx", value=shared.args.n_ctx)
+                        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")
+                        shared.gradio['groupsize'] = gr.Dropdown(label="groupsize", choices=["None", 32, 64, 128, 1024], value=shared.args.groupsize if shared.args.groupsize > 0 else "None")
+                        shared.gradio['model_type'] = gr.Dropdown(label="model_type", choices=["None", "llama", "opt", "gptj"], value=shared.args.model_type or "None")
+                        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)
+                        shared.gradio['autogptq_info'] = gr.Markdown('On some systems, AutoGPTQ can be 2x slower than GPTQ-for-LLaMa. You can manually select the GPTQ-for-LLaMa loader above.')
 
                     with gr.Column():
+                        shared.gradio['triton'] = gr.Checkbox(label="triton", value=shared.args.triton)
+                        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. Disable if running low on VRAM.')
+                        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.')
+                        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.')
+                        shared.gradio['load_in_8bit'] = gr.Checkbox(label="load-in-8bit", value=shared.args.load_in_8bit)
+                        shared.gradio['bf16'] = gr.Checkbox(label="bf16", value=shared.args.bf16)
+                        shared.gradio['auto_devices'] = gr.Checkbox(label="auto-devices", value=shared.args.auto_devices)
+                        shared.gradio['disk'] = gr.Checkbox(label="disk", value=shared.args.disk)
+                        shared.gradio['cpu'] = gr.Checkbox(label="cpu", value=shared.args.cpu)
+                        shared.gradio['load_in_4bit'] = gr.Checkbox(label="load-in-4bit")
+                        shared.gradio['use_double_quant'] = gr.Checkbox(label="use_double_quant", value=shared.args.use_double_quant)
                         shared.gradio['no_mmap'] = gr.Checkbox(label="no-mmap", value=shared.args.no_mmap)
                         shared.gradio['mlock'] = gr.Checkbox(label="mlock", value=shared.args.mlock)
                         shared.gradio['llama_cpp_seed'] = gr.Number(label='Seed (0 for random)', value=shared.args.llama_cpp_seed)
+                        shared.gradio['trust_remote_code'] = gr.Checkbox(label="trust-remote-code", value=shared.args.trust_remote_code, info='Make sure to inspect the .py files inside the model folder before loading it with this option enabled.')
+                        shared.gradio['gptq_for_llama_info'] = gr.Markdown('GPTQ-for-LLaMa is currently 2x faster than AutoGPTQ on some systems. It is installed by default with the one-click installers. Otherwise, it has to be installed manually following the instructions here: [instructions](https://github.com/oobabooga/text-generation-webui/blob/main/docs/GPTQ-models-(4-bit-mode).md#installation-1).')
+
+        with gr.Column():
+            with gr.Row():
+                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.')
+
+            shared.gradio['custom_model_menu'] = gr.Textbox(label="Download custom 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")
+            shared.gradio['download_model_button'] = gr.Button("Download")
 
             with gr.Row():
                 shared.gradio['model_status'] = gr.Markdown('No model is loaded' if shared.model_name == 'None' else 'Ready')
 
+    shared.gradio['loader'].change(loaders.make_loader_params_visible, shared.gradio['loader'], [shared.gradio[k] for k in loaders.get_all_params()])
+
     # In this event handler, the interface state is read and updated
     # with the model defaults (if any), and then the model is loaded
     # unless "autoload_model" is unchecked
     shared.gradio['model_menu'].change(
         ui.gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
-        load_model_specific_settings, [shared.gradio[k] for k in ['model_menu', 'interface_state']], shared.gradio['interface_state']).then(
+        apply_model_settings_to_state, [shared.gradio[k] for k in ['model_menu', 'interface_state']], shared.gradio['interface_state']).then(
         ui.apply_interface_values, shared.gradio['interface_state'], [shared.gradio[k] for k in ui.list_interface_input_elements(chat=shared.is_chat())], show_progress=False).then(
         update_model_parameters, shared.gradio['interface_state'], None).then(
-        load_model_wrapper, [shared.gradio[k] for k in ['model_menu', 'autoload_model']], shared.gradio['model_status'], show_progress=False)
+        load_model_wrapper, [shared.gradio[k] for k in ['model_menu', 'loader', 'autoload_model']], shared.gradio['model_status'], show_progress=False)
 
     load.click(
         ui.gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
         update_model_parameters, shared.gradio['interface_state'], None).then(
-        partial(load_model_wrapper, autoload=True), shared.gradio['model_menu'], shared.gradio['model_status'], show_progress=False)
+        partial(load_model_wrapper, autoload=True), [shared.gradio[k] for k in ['model_menu', 'loader']], shared.gradio['model_status'], show_progress=False)
 
     unload.click(
         unload_model, None, None).then(
@@ -374,7 +266,7 @@ def create_model_menus():
         unload_model, None, None).then(
         ui.gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
         update_model_parameters, shared.gradio['interface_state'], None).then(
-        partial(load_model_wrapper, autoload=True), shared.gradio['model_menu'], shared.gradio['model_status'], show_progress=False)
+        partial(load_model_wrapper, autoload=True), [shared.gradio[k] for k in ['model_menu', 'loader']], shared.gradio['model_status'], show_progress=False)
 
     save_settings.click(
         ui.gather_interface_values, [shared.gradio[k] for k in shared.input_elements], shared.gradio['interface_state']).then(
@@ -1100,7 +992,7 @@ if __name__ == "__main__":
 
     # If any model has been selected, load it
     if shared.model_name != 'None':
-        model_settings = get_model_specific_settings(shared.model_name)
+        model_settings = get_model_settings_from_yamls(shared.model_name)
         shared.settings.update(model_settings)  # hijacking the interface defaults
         update_model_parameters(model_settings, initial=True)  # hijacking the command-line arguments
 
@@ -1117,6 +1009,10 @@ if __name__ == "__main__":
             'instruction_template': shared.settings['instruction_template']
         })
 
+    shared.persistent_interface_state.update({
+        'loader': shared.args.loader or 'Transformers',
+    })
+
     shared.generation_lock = Lock()
     # Launch the web UI
     create_interface()