# Perceiver: masked language modeling with Transformers

This page is the Transformers contract of the Perceiver 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 `deepmind/language-perceiver`. 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.

## Predict the masked bytes

```py runnable:masked_lm
import torch
from transformers import PerceiverTokenizer, PerceiverForMaskedLM, set_seed
import json  # doc-builder: hide
import os  # doc-builder: hide

model_id = "deepmind/language-perceiver"
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)

tokenizer = PerceiverTokenizer.from_pretrained(model_id)
model = PerceiverForMaskedLM.from_pretrained(model_id).to(device)
if variant:  # doc-builder: hide
    del model  # doc-builder: hide
    model = PerceiverForMaskedLM.from_pretrained(model_id, **variant).to(device)  # doc-builder: hide

# Two short sentences; the tokenizer maps each character to its byte value.
# The decoder predicts a fixed number of positions, so pad to that length.
texts = ["The cat sat on the mat.", "A dog ran in the park."]
encoding = tokenizer(texts, padding="max_length", max_length=model.config.max_position_embeddings, return_tensors="pt")
input_ids = encoding["input_ids"].to(device)
attention_mask = encoding["attention_mask"].to(device)

original_ids = input_ids.clone()

# Mask "mat" in the first sentence: characters 19-21, shifted by one for the leading [CLS] token
start = texts[0].index("mat") + 1
input_ids[0, start : start + 3] = tokenizer.mask_token_id

with torch.no_grad():
    outputs = model(inputs=input_ids, attention_mask=attention_mask)
logits = outputs.logits

# Compact probe: first 5 logits at each masked position
masked = input_ids == tokenizer.mask_token_id
masked_logits = logits[masked]
print(masked_logits[:, :5].tolist())
```

## One training step

```py runnable:masked_lm:2
# Labels: the original bytes; ignore pad positions with -100
labels = original_ids.clone()
labels[attention_mask == 0] = -100

model.train()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)
loss = model(inputs=input_ids, attention_mask=attention_mask, labels=labels).loss
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 = ("perceiver.input_preprocessor.embeddings.weight", "perceiver.encoder.self_attends.0.attention.self.query.weight")
    result = {
        "runtime": {
            "device": model.device.type,
            "dtype": str(model.dtype).removeprefix("torch."),
            "attn_implementation": model.config._attn_implementation,
        },
        "masked_logits": masked_logits[:, :5].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))
```