[GH-ISSUE #163] "None" gradients when using fine-tuned CLIP #142

New issue

Open

opened 2026-10-31 21:11:39 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-31 21:11:39 +00:00

Owner

Originally created by @sarahESL on GitHub (Oct 27, 2021).
Original GitHub issue: https://github.com/openai/CLIP/issues/163

Using the following snippet gives "none" values for some CLIP parameters such as positional_embedding,

for name, p in self.model.named_parameters():
    print(p.grad)

where

    myclip, _ = clip.load(args.clip_vision_encoder, jit=False)
    checkpoint = torch.load(args.clip_path)
    myclip.load_state_dict(checkpoint['state_dict'])
    self.model = myclip.float()

Is this behavior expected?

Originally created by @sarahESL on GitHub (Oct 27, 2021). Original GitHub issue: https://github.com/openai/CLIP/issues/163 Using the following snippet gives "none" values for some CLIP parameters such as `positional_embedding`, ``` for name, p in self.model.named_parameters(): print(p.grad) ``` where ``` myclip, _ = clip.load(args.clip_vision_encoder, jit=False) checkpoint = torch.load(args.clip_path) myclip.load_state_dict(checkpoint['state_dict']) self.model = myclip.float() ``` Is this behavior expected?

bsakaguchi referenced this issue

2026-10-31 21:11:56 +00:00

[GH-ISSUE #192] About chinese #167

bsakaguchi referenced this issue

2026-10-31 21:13:06 +00:00

[GH-ISSUE #302] whether CLIP can embedding chinese text ? #261