# BigBird: masked language modeling and next-sentence prediction This page is the Transformers contract of the BigBird 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 `google/bigbird-roberta-base`. 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. ## Masked language modeling and next-sentence prediction ```py runnable:pretraining import torch from transformers import AutoModelForPreTraining, AutoTokenizer, set_seed import json # doc-builder: hide import os # doc-builder: hide model_id = "google/bigbird-roberta-base" 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, dropout, and data augmentation draw random numbers tokenizer = AutoTokenizer.from_pretrained(model_id) model = AutoModelForPreTraining.from_pretrained(model_id).to(device) if variant: # doc-builder: hide del model # doc-builder: hide model = AutoModelForPreTraining.from_pretrained(model_id, **variant).to(device) # doc-builder: hide sentences = [ "The quick brown fox jumps over the lazy dog.", "Transformers are powerful models for natural language processing.", ] inputs = tokenizer(sentences, return_tensors="pt", padding=True).to(device) with torch.no_grad(): outputs = model(**inputs) prediction_logits = outputs.prediction_logits seq_relationship_logits = outputs.seq_relationship_logits # Probe a few logits: first token of each sequence and the NSP logits probe = prediction_logits[:, 0, :5].float().tolist() nsp_probe = seq_relationship_logits.float().tolist() print("prediction logits (first token, first 5):", probe) print("seq relationship logits:", nsp_probe) ``` ## One training step ```py runnable:pretraining:2 # Create labels for MLM: copy input_ids and mask padded positions labels = inputs["input_ids"].clone() labels[inputs["attention_mask"] == 0] = -100 # NSP labels: 0 for continuation, 1 for random (here we use 0 for all) next_sentence_label = torch.zeros(inputs["input_ids"].shape[0], dtype=torch.long, device=device) model.train() optimizer = torch.optim.SGD(model.parameters(), lr=0.01) loss = model(**inputs, labels=labels, next_sentence_label=next_sentence_label).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 = ("bert.encoder.layer.0.attention.self.query.weight", "bert.encoder.layer.0.intermediate.dense.weight") result = { "runtime": { "device": model.device.type, "dtype": str(model.dtype).removeprefix("torch."), "attn_implementation": model.config._attn_implementation, }, "prediction_logits": probe, "seq_relationship_logits": nsp_probe, "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)) ```