tu_nlp_byte_empyrealm — Byte × Empyrealm hybrid
Source-owned OA v0.12.0 SDK tutorial. Recorded results qualify the named device and driver.
The byte-tokenizer × Empyrealm hybrid 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 256-token packed-U8 byte vocabulary (ByteSampler) — one token per source byte |
| Model | Empyrealm(Byte Embedding + Mamba-3/attention hybrid stack) → Linear(32→256) |
| Parameters | 73,848 |
| Optimizer | AdamW(lr=0.003) — lower rate for the Mamba-3/Empyrealm family |
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 (184 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 |
|---|---|---|---|---|
| 5.667→0.182 | 92.9% | 184 | 40.20 | 25.5K |
Reference generation on the default recipe (asserted by the tutorial):
1prompt: "to be"2generated: "to be that is the question whether tis nobler in the mind to suffer the slings and ar"3
Run
cargo run --release -p sdk --example tu_nlp_byte_empyrealm
Regenerate the training-loss image from the repository root:
cargo run --release -p sdk --example tu_nlp_byte_empyrealm -- --plot sdk/rs/tutorials/ml/nlp/tu_nlp_byte_empyrealm/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 — Empyrealm Core · autograd fidelity ║╚══════════════════════════════════════════════════════════════════╝tokenizer: raw bytes · vocab=256Task: dense next-byte prediction at every positionModel: Empyrealm(Byte Embedding + Mamba-3/attention hybrid stack) → Linear(32→256)params: 73848 Optimizer: AdamW(lr=0.003)300/300 |██████████| 12.83s · 42.78 ms/step · 1.50K sample/s · 23.94K byte token/s · 23.94K byte/s · loss: 0.1824Summary:loss: initial 5.667074 · final 0.182391 · mean 0.568823Wall: 42.78 ms/step · 1.50K sample/s · 23.94K byte token/s · 23.94K byte/sGPU (byte_empyrealm_training_step): mean 41.226 ms/step · p50 39.541 · p95 49.846 · min 25.428 · max 57.788 · 1.55K sample/s · 24.84K byte token/s · 24.84K byte/s · wall-GPU gap 4%run: 12.83s · 300 steps · batch 64 · sequence 16 byte token/sampleEvaluation:Random-loss baseline ln(256) = 5.5452byte-token accuracy: 92.9%post-update evaluation loss: 0.179192Generation:prompt: "to be"generated: "to be that is the question whether tis nobler in the mind to suffer the slings and ar"program: 184 nodes · 1 recording · 299 cache hits · 0 uploadscheckpoint: optimizer step, accuracy, and exact generation restored