# MT5: seq2seq generation with Transformers

This page is the Transformers contract of the MT5 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/mt5-small`. 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.

## Translate a batch

```py runnable:mt5
import torch
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer, set_seed
import json  # doc-builder: hide
import os  # doc-builder: hide

model_id = "google/mt5-small"
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 generation draw random numbers

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSeq2SeqLM.from_pretrained(model_id).to(device)
if variant:  # doc-builder: hide
    del model  # doc-builder: hide
    model = AutoModelForSeq2SeqLM.from_pretrained(model_id, **variant).to(device)  # doc-builder: hide

# MT5 is multilingual; the "translate English to German:" prefix selects the task.
prompts = [
    "translate English to German: The house is wonderful.",
    "translate English to German: I like to read books.",
]
inputs = tokenizer(prompts, return_tensors="pt", padding=True).to(device)

model.eval()
with torch.no_grad():
    generated = model.generate(**inputs, max_new_tokens=8)
    logits = model(**inputs, decoder_input_ids=generated).logits

decoded = tokenizer.batch_decode(generated, skip_special_tokens=True)
print(decoded)

# Probe a few logits at the first generated position of each sequence.
probe = logits[:, 0, :5].float().tolist()
```

## One training step

```py runnable:mt5:2
# Teacher forcing: the labels are the tokenized German references.
references = ["Das Haus ist wunderbar.", "Ich lese gerne Bücher."]
labels = tokenizer(references, return_tensors="pt", padding=True).to(device)["input_ids"]
labels[labels == tokenizer.pad_token_id] = -100

model.train()
optimizer = torch.optim.SGD([p for p in model.parameters() if p.requires_grad], lr=0.01)
loss = model(**inputs, 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 = (
        "encoder.block.0.layer.0.SelfAttention.q.weight",
        "decoder.block.0.layer.0.SelfAttention.q.weight",
        "decoder.block.0.layer.1.EncDecAttention.q.weight",
    )
    result = {
        "runtime": {
            "device": model.device.type,
            "dtype": str(model.dtype).removeprefix("torch."),
            "attn_implementation": model.config._attn_implementation,
        },
        "decoded_text": decoded,
        "probe_logits": 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))
```