tu_nlp_bpe_transformer — BPE × Transformer
Source-owned OA v0.12.0 SDK tutorial. Recorded results qualify the named device and driver.
The BPE-tokenizer × Transformer cell of the — same 576-character corpus, same 300-step recipe and captured training lifecycle. This entry selects the tokenizer and model shown below. See for the full walkthrough.
This cell
| Tokenizer | the 320-token learned BPE vocabulary (BpeTokenizer/BpeSampler: 256 bytes + 64 merges) — merges collapse the corpus into short predictable runs, which is why BPE rows reach ~0.02 loss |
| Model | BPE + position Embedding → Transformer(32,64) → LN → Linear(32→320) |
| Parameters | 29,920 |
| Optimizer | AdamW(lr=0.01) — the canonical recipe |
The loop is a prepare/record pair per step — GradientTape → cross_entropy → backward → AdamW —
driven by oa::ml::ItTraining, so the whole run is one captured GPU program replayed
299 times (44 nodes for this model).
Measured — Surface Laptop 5 (ADL GT2)

Canonical 300-step training loss, recorded by this tutorial.
| Loss init→final | Accuracy | Program nodes | Wall ms/step | Token/s |
|---|---|---|---|---|
| 5.871→0.021 | 98.9% | 44 | 7.51 | 136.3K |
Reference generation on the default recipe (asserted by the tutorial):
1prompt: "to be"2generated: "to be end them to be or not to be that is the question whether tis nobler in the mind"3
Run
cargo run --release -p sdk --example tu_nlp_bpe_transformer
Regenerate the training-loss image from the repository root:
cargo run --release -p sdk --example tu_nlp_bpe_transformer -- --plot sdk/rs/tutorials/ml/nlp/tu_nlp_bpe_transformer/asset/training-curve.png
On the default recipe the tutorial self-tests: loss regime, accuracy
floor, exact reference generation, and .oam checkpoint round-trip.
Terminal output
Recorded canonical 300-step run on Intel Iris Xe (ADL GT2), Mesa 26.2.7. Startup logs, intermediate progress updates and the plot-save path are omitted. Timings vary by run; these are not new benchmark baselines.
╔══════════════════════════════════════════════════════════════════╗║ OA Tutorial — BPE Transformer · all-position LM (autograd) ║╚══════════════════════════════════════════════════════════════════╝tokenizer: byte BPE · vocab=320 (256 bytes + 64 merges)compression: 576 bytes → 210 tokens · 2.743 byte/token · 63.5% fewer positionscontext coverage: 16 BPE tokens ≈ 43.9 source bytes at corpus averageTask: dense next-token prediction at every positionModel: BPE + position Embedding → Transformer(32,64) → LN → Linear(32→320)params: 29920 Optimizer: AdamW(lr=0.01)300/300 |██████████| 2.22s · 7.41 ms/step · 8.64K sample/s · 138.19K BPE token/s · 383.55K byte/s · loss: 0.0206Summary:loss: initial 5.871270 · final 0.020626 · mean 0.243361Wall: 7.41 ms/step · 8.64K sample/s · 138.18K BPE token/s · 383.53K byte/sGPU (bpe_transformer_training_step): mean 6.417 ms/step · p50 4.318 · p95 12.888 · min 4.020 · max 14.026 · 9.97K sample/s · 159.58K BPE token/s · 442.92K byte/s · wall-GPU gap 13%run: 2.22s · 300 steps · batch 64 · sequence 16 BPE token/sampleEvaluation:Random-loss baseline ln(320) = 5.7683final batch: 2.771 byte/token · 0.0107 bits/byteBPE-token accuracy: 98.9%post-update evaluation loss: 0.020028Generation:prompt: "to be"generated: "to be end them to be or not to be that is the question whether tis nobler in the mind"program: 44 nodes · 1 recording · 299 cache hits · 0 uploads