list_pretrained

List all available pretrained model/weight combinations

Returns a list of all pretrained models with their available weight tags. This shows which pretrained checkpoints are available for each model architecture.

Signature

def list_pretrained(as_str: bool = False):
    ...

Parameters

as_str · bool · default: "False"

If True, returns list of strings in format 'model_name:tag'. If False, returns list of tuples (model_name, tag).

Returns

pretrained_list · list

List of pretrained model/tag combinations. Format depends on as_str parameter:

• If as_str=False: List of tuples [(model_name, tag), ...]
• If as_str=True: List of strings ['model_name:tag', ...]

Example

import open_clip

# Get list as tuples (default)
pretrained = open_clip.list_pretrained()
print(f"Total pretrained combinations: {len(pretrained)}")
print(f"First 5: {pretrained[:5]}")
# Output: [('RN50', 'openai'), ('RN50', 'yfcc15m'), ('RN50', 'cc12m'), ...]

# Get list as strings
pretrained_str = open_clip.list_pretrained(as_str=True)
print(f"First 5 as strings: {pretrained_str[:5]}")
# Output: ['RN50:openai', 'RN50:yfcc15m', 'RN50:cc12m', ...]

# Find all pretrained weights for a specific model
vit_b_32_weights = [tag for model, tag in pretrained if model == 'ViT-B-32']
print(f"ViT-B-32 available weights: {vit_b_32_weights}")
# Output: ['openai', 'laion400m_e31', 'laion400m_e32', 'laion2b_e16', ...]

# Find all models with OpenAI weights
openai_models = [model for model, tag in pretrained if tag == 'openai']
print(f"Models with OpenAI weights: {openai_models}")
# Output: ['RN50', 'RN101', 'RN50x4', 'RN50x16', 'RN50x64', 'ViT-B-32', ...]

# Check if a specific combination exists
if ('ViT-L-14', 'datacomp_xl_s13b_b90k') in pretrained:
    print("ViT-L-14 with datacomp_xl_s13b_b90k is available")

Common Pretrained Tags

• openai: Official OpenAI CLIP weights
• laion400m_e31/e32: Trained on LAION-400M dataset
• laion2b_s34b_b88k: Trained on LAION-2B dataset
• datacomp_xl_s13b_b90k: Trained on DataComp-XL
• metaclip_400m/fullcc: MetaCLIP weights
• dfn2b/dfn5b: Apple DFN weights
• webli: Google SigLIP weights trained on WebLI

Helper Functions

For more targeted queries, consider using:

# Get all pretrained tags for a specific model
tags = open_clip.list_pretrained_tags_by_model('ViT-B-32')
print(f"ViT-B-32 tags: {tags}")

# Get all models with a specific tag
models = open_clip.list_pretrained_models_by_tag('openai')
print(f"Models with OpenAI weights: {models}")

# Check if a specific model/tag combination exists
exists = open_clip.is_pretrained_cfg('ViT-B-32', 'openai')
print(f"ViT-B-32:openai exists: {exists}")

See Also

• get_pretrained_cfg - Get configuration for a specific pretrained model/tag
• list_models - List available model architectures
• create_model - Create a model with pretrained weights