open_clip/src/open_clip/factory.py

import json
import logging
import os
import pathlib
import re
from copy import deepcopy
from pathlib import Path
from typing import Any, Dict, Optional, Tuple, Union

import torch

from .constants import OPENAI_DATASET_MEAN, OPENAI_DATASET_STD
from .model import CLIP, CustomTextCLIP, convert_weights_to_lp, convert_to_custom_text_state_dict,\
    resize_pos_embed, get_cast_dtype
from .openai import load_openai_model
from .pretrained import is_pretrained_cfg, get_pretrained_cfg, download_pretrained, list_pretrained_tags_by_model, download_pretrained_from_hf
from .transform import image_transform, AugmentationCfg
from .tokenizer import HFTokenizer, tokenize


HF_HUB_PREFIX = 'hf-hub:'
_MODEL_CONFIG_PATHS = [Path(__file__).parent / f"model_configs/"]
_MODEL_CONFIGS = {}  # directory (model_name: config) of model architecture configs


def _natural_key(string_):
    return [int(s) if s.isdigit() else s for s in re.split(r'(\d+)', string_.lower())]


def _rescan_model_configs():
    global _MODEL_CONFIGS

    config_ext = ('.json',)
    config_files = []
    for config_path in _MODEL_CONFIG_PATHS:
        if config_path.is_file() and config_path.suffix in config_ext:
            config_files.append(config_path)
        elif config_path.is_dir():
            for ext in config_ext:
                config_files.extend(config_path.glob(f'*{ext}'))

    for cf in config_files:
        with open(cf, 'r') as f:
            model_cfg = json.load(f)
            if all(a in model_cfg for a in ('embed_dim', 'vision_cfg', 'text_cfg')):
                _MODEL_CONFIGS[cf.stem] = model_cfg

    _MODEL_CONFIGS = {k: v for k, v in sorted(_MODEL_CONFIGS.items(), key=lambda x: _natural_key(x[0]))}


_rescan_model_configs()  # initial populate of model config registry


def list_models():
    """ enumerate available model architectures based on config files """
    return list(_MODEL_CONFIGS.keys())


def add_model_config(path):
    """ add model config path or file and update registry """
    if not isinstance(path, Path):
        path = Path(path)
    _MODEL_CONFIG_PATHS.append(path)
    _rescan_model_configs()


def get_model_config(model_name):
    if model_name in _MODEL_CONFIGS:
        return deepcopy(_MODEL_CONFIGS[model_name])
    else:
        return None


def get_tokenizer(model_name):
    if model_name.startswith(HF_HUB_PREFIX):
        tokenizer = HFTokenizer(model_name[len(HF_HUB_PREFIX):])
    else:
        config = get_model_config(model_name)
        tokenizer = HFTokenizer(config['text_cfg']['hf_tokenizer_name']) if 'hf_tokenizer_name' in config['text_cfg'] else tokenize
    return tokenizer


def load_state_dict(checkpoint_path: str, map_location='cpu'):
    checkpoint = torch.load(checkpoint_path, map_location=map_location)
    if isinstance(checkpoint, dict) and 'state_dict' in checkpoint:
        state_dict = checkpoint['state_dict']
    else:
        state_dict = checkpoint
    if next(iter(state_dict.items()))[0].startswith('module'):
        state_dict = {k[7:]: v for k, v in state_dict.items()}
    return state_dict


def load_checkpoint(model, checkpoint_path, strict=True):
    state_dict = load_state_dict(checkpoint_path)
    # detect old format and make compatible with new format
    if 'positional_embedding' in state_dict and not hasattr(model, 'positional_embedding'):
        state_dict = convert_to_custom_text_state_dict(state_dict)
    resize_pos_embed(state_dict, model)
    incompatible_keys = model.load_state_dict(state_dict, strict=strict)
    return incompatible_keys


def create_model(
        model_name: str,
        pretrained: Optional[str] = None,
        precision: str = 'fp32',
        device: Union[str, torch.device] = 'cpu',
        jit: bool = False,
        force_quick_gelu: bool = False,
        force_custom_text: bool = False,
        force_patch_dropout: Optional[float] = None,
        force_image_size: Optional[Union[int, Tuple[int, int]]] = None,
        pretrained_image: bool = False,
        pretrained_hf: bool = True,
        cache_dir: Optional[str] = None,
):
    has_hf_hub_prefix = model_name.startswith(HF_HUB_PREFIX)
    if has_hf_hub_prefix:
        model_id = model_name[len(HF_HUB_PREFIX):]
        checkpoint_path = download_pretrained_from_hf(model_id, cache_dir=cache_dir)
        config_path = download_pretrained_from_hf(model_id, filename='open_clip_config.json', cache_dir=cache_dir)

        with open(config_path, 'r', encoding='utf-8') as f:
            config = json.load(f)
        pretrained_cfg = config['preprocess_cfg']
        model_cfg = config['model_cfg']
    else:
        model_name = model_name.replace('/', '-')  # for callers using old naming with / in ViT names
        checkpoint_path = None
        pretrained_cfg = {}
        model_cfg = None

    if isinstance(device, str):
        device = torch.device(device)

    if pretrained and pretrained.lower() == 'openai':
        logging.info(f'Loading pretrained {model_name} from OpenAI.')
        model = load_openai_model(
            model_name,
            precision=precision,
            device=device,
            jit=jit,
            cache_dir=cache_dir,
        )
    else:
        model_cfg = model_cfg or get_model_config(model_name)
        if model_cfg is not None:
            logging.info(f'Loaded {model_name} model config.')
        else:
            logging.error(f'Model config for {model_name} not found; available models {list_models()}.')
            raise RuntimeError(f'Model config for {model_name} not found.')

        if force_quick_gelu:
            # override for use of QuickGELU on non-OpenAI transformer models
            model_cfg["quick_gelu"] = True

        if force_patch_dropout is not None:
            # override the default patch dropout value
            model_cfg["vision_cfg"]["patch_dropout"] = force_patch_dropout

        if force_image_size is not None:
            # override model config's image size
            model_cfg["vision_cfg"]["image_size"] = force_image_size

        if pretrained_image:
            if 'timm_model_name' in model_cfg.get('vision_cfg', {}):
                # pretrained weight loading for timm models set via vision_cfg
                model_cfg['vision_cfg']['timm_model_pretrained'] = True
            else:
                assert False, 'pretrained image towers currently only supported for timm models'

        cast_dtype = get_cast_dtype(precision)
        is_hf_model = 'hf_model_name' in model_cfg.get('text_cfg', {})
        custom_text = model_cfg.pop('custom_text', False) or force_custom_text or is_hf_model

        if custom_text:
            if is_hf_model:
                model_cfg['text_cfg']['hf_model_pretrained'] = pretrained_hf
            model = CustomTextCLIP(**model_cfg, cast_dtype=cast_dtype)
        else:
            model = CLIP(**model_cfg, cast_dtype=cast_dtype)

        if pretrained:
            checkpoint_path = ''
            pretrained_cfg = get_pretrained_cfg(model_name, pretrained)
            if pretrained_cfg:
                checkpoint_path = download_pretrained(pretrained_cfg, cache_dir=cache_dir)
            elif os.path.exists(pretrained):
                checkpoint_path = pretrained

            if checkpoint_path:
                logging.info(f'Loading pretrained {model_name} weights ({pretrained}).')
                load_checkpoint(model, checkpoint_path)
            else:
                error_str = (
                    f'Pretrained weights ({pretrained}) not found for model {model_name}.'
                    f'Available pretrained tags ({list_pretrained_tags_by_model(model_name)}.')
                logging.warning(error_str)
                raise RuntimeError(error_str)
        elif has_hf_hub_prefix:
            logging.info(f'Loading pretrained {model_name} weights ({pretrained}).')
            load_checkpoint(model, checkpoint_path)

        model.to(device=device)
        if precision in ("fp16", "bf16"):
            convert_weights_to_lp(model, dtype=torch.bfloat16 if precision == 'bf16' else torch.float16)

        # set image / mean metadata from pretrained_cfg if available, or use default
        model.visual.image_mean = pretrained_cfg.get('mean', None) or OPENAI_DATASET_MEAN
        model.visual.image_std = pretrained_cfg.get('std', None) or OPENAI_DATASET_STD

        if jit:
            model = torch.jit.script(model)

    return model


def create_model_and_transforms(
        model_name: str,
        pretrained: Optional[str] = None,
        precision: str = 'fp32',
        device: Union[str, torch.device] = 'cpu',
        jit: bool = False,
        force_quick_gelu: bool = False,
        force_custom_text: bool = False,
        force_patch_dropout: Optional[float] = None,
        force_image_size: Optional[Union[int, Tuple[int, int]]] = None,
        pretrained_image: bool = False,
        pretrained_hf: bool = True,
        image_mean: Optional[Tuple[float, ...]] = None,
        image_std: Optional[Tuple[float, ...]] = None,
        aug_cfg: Optional[Union[Dict[str, Any], AugmentationCfg]] = None,
        cache_dir: Optional[str] = None,
):
    model = create_model(
        model_name,
        pretrained,
        precision=precision,
        device=device,
        jit=jit,
        force_quick_gelu=force_quick_gelu,
        force_custom_text=force_custom_text,
        force_patch_dropout=force_patch_dropout,
        force_image_size=force_image_size,
        pretrained_image=pretrained_image,
        pretrained_hf=pretrained_hf,
        cache_dir=cache_dir,
    )

    image_mean = image_mean or getattr(model.visual, 'image_mean', None)
    image_std = image_std or getattr(model.visual, 'image_std', None)
    preprocess_train = image_transform(
        model.visual.image_size,
        is_train=True,
        mean=image_mean,
        std=image_std,
        aug_cfg=aug_cfg,
    )
    preprocess_val = image_transform(
        model.visual.image_size,
        is_train=False,
        mean=image_mean,
        std=image_std,
    )

    return model, preprocess_train, preprocess_val


def create_model_from_pretrained(
        model_name: str,
        pretrained: str,
        precision: str = 'fp32',
        device: Union[str, torch.device] = 'cpu',
        jit: bool = False,
        force_quick_gelu: bool = False,
        force_custom_text: bool = False,
        force_image_size: Optional[Union[int, Tuple[int, int]]] = None,
        return_transform: bool = True,
        image_mean: Optional[Tuple[float, ...]] = None,
        image_std: Optional[Tuple[float, ...]] = None,
        cache_dir: Optional[str] = None,
):
    if not is_pretrained_cfg(model_name, pretrained) and not os.path.exists(pretrained):
        raise RuntimeError(
            f'{pretrained} is not a valid pretrained cfg or checkpoint for {model_name}.'
            f' Use open_clip.list_pretrained() to find one.')

    model = create_model(
        model_name,
        pretrained,
        precision=precision,
        device=device,
        jit=jit,
        force_quick_gelu=force_quick_gelu,
        force_custom_text=force_custom_text,
        force_image_size=force_image_size,
        cache_dir=cache_dir,
    )

    if not return_transform:
        return model

    image_mean = image_mean or getattr(model.visual, 'image_mean', None)
    image_std = image_std or getattr(model.visual, 'image_std', None)
    preprocess = image_transform(
        model.visual.image_size,
        is_train=False,
        mean=image_mean,
        std=image_std,
    )

    return model, preprocess