File size: 3,527 Bytes
c9fce36 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 | # CLIP text encoder: projected text embeddings with Transformers
This page is the Transformers contract of the CLIP text encoder MIC and also a usage example. [doc-builder](https://github.com/huggingface/doc-builder/blob/main/docs/runnable-code-blocks.md) renders the `runnable` blocks below as documentation, and its pytest plugin runs them as written against `peft-internal-testing/tiny-clip-text-2`. The MIC runner runs the same blocks against each slim fixture and each checkpoint in [`integration.json`](../../integration.json). Lines marked `# doc-builder: hide` only matter there: they swap in the pinned checkpoint or variant and print the result record.
## Embed a batch of texts
```py runnable:text_embed
import torch
from transformers import CLIPTextModelWithProjection, AutoTokenizer, set_seed
import json # doc-builder: hide
import os # doc-builder: hide
model_id = "peft-internal-testing/tiny-clip-text-2"
model_id = os.environ.get("MIC_MODEL", model_id) # doc-builder: hide
variant = json.loads(os.environ.get("MIC_MODEL_KWARGS", "{}")) # doc-builder: hide
device = "cuda" if torch.cuda.is_available() else "cpu"
set_seed(0) # seeds torch, NumPy, and Python: loading and dropout draw random numbers
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = CLIPTextModelWithProjection.from_pretrained(model_id).to(device)
if variant: # doc-builder: hide
del model # doc-builder: hide
model = CLIPTextModelWithProjection.from_pretrained(model_id, **variant).to(device) # doc-builder: hide
# Two distinct texts so batch perturbation is meaningful; right padding is the CLIP default.
texts = ["a photo of a cat", "a photo of a red bus on a street"]
inputs = tokenizer(texts, return_tensors="pt", padding=True).to(device)
model.eval()
with torch.no_grad():
outputs = model(**inputs)
text_embeds = outputs.text_embeds
print(text_embeds.shape)
print(text_embeds[0, :4])
```
## One training step
```py runnable:text_embed:2
# A tiny linear probe on the projected text embeddings gives a real supervised loss.
probe = torch.nn.Linear(model.config.projection_dim, 2).to(device, model.dtype)
labels = torch.tensor([0, 1], device=device)
model.train()
optimizer = torch.optim.SGD(list(model.parameters()) + list(probe.parameters()), lr=0.01)
embeds = model(**inputs).text_embeds
loss = torch.nn.functional.cross_entropy(probe(embeds), labels)
loss.backward()
before = {n: p.detach().cpu().clone() for n, p in model.named_parameters() if p.requires_grad} # doc-builder: hide
optimizer.step()
print(f"loss: {loss.item():.4f}")
if os.environ.get("MIC_RUNNER"): # doc-builder: hide
named = dict(model.named_parameters())
checked = ("text_model.encoder.layers.0.self_attn.q_proj.weight", "text_model.encoder.layers.0.mlp.fc1.weight", "text_projection.weight")
result = {
"runtime": {
"device": model.device.type,
"dtype": str(model.dtype).removeprefix("torch."),
"attn_implementation": model.config._attn_implementation,
},
"text_embeds": text_embeds.float().tolist(),
"probe_logits": probe(embeds).float().tolist(),
"training": {
"loss": loss.item(),
"gradients": {n: bool(named[n].grad is not None and torch.count_nonzero(named[n].grad) > 0 and torch.isfinite(named[n].grad).all()) for n in checked},
"updated": any(not torch.equal(p.detach().cpu(), before[n]) for n, p in model.named_parameters() if n in before),
},
}
print(json.dumps(result))
```
|