Realm

tu_nlp_char_mamba3 — Character × Mamba-3

Source-owned OA v0.12.0 SDK tutorial. Recorded results qualify the named device and driver.

Experimental

The character-tokenizer × Mamba-3 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

Tokenizerthe 27-token character vocabulary (CharSampler) — accuracy is comparable within Char rows only
ModelEmbedding(32) → Mamba-3 SISO → residual → Linear(27)
Parameters10,915
OptimizerAdamW(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 (34 nodes for this model).

Measured — Surface Laptop 5 (ADL GT2)

Character × Mamba-3 training loss

Canonical 300-step training loss, recorded by this tutorial.

Loss init→finalAccuracyProgram nodesWall ms/stepToken/s
3.303→0.19593.0%3410.6995.8K

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_char_mamba3
cargo run --release -p sdk --example tu_nlp_char_mamba3 -- --plot sdk/rs/tutorials/ml/nlp/tu_nlp_char_mamba3/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 — Char Mamba-3 · all-position LM (autograd) ║
╚══════════════════════════════════════════════════════════════════╝
tokenizer: character · vocab=27 (a-z + space)
Task: dense next-character prediction at every position
Model: Embedding(32) → Mamba-3 SISO → residual → Linear(27)
params: 10915 Optimizer: AdamW(lr=0.003)
training: 300 steps · batch=64 · sequence=16 character tokens
300/300 |██████████| 3.18s · 10.58 ms/step · 6.05K sample/s · 96.74K token/s · loss: 0.1945
Summary:
loss: initial 3.303064 · final 0.194512 · mean 0.714765
Wall: 10.59 ms/step · 6.05K sample/s · 96.74K token/s
GPU (char_mamba3_training_step): mean 9.616 ms/step · p50 7.558 · p95 16.954 · min 6.281 · max 20.094 · 6.66K sample/s · 106.49K token/s · wall-GPU gap 9%
run: 3.18s · 300 steps · batch 64 · sequence 16 token/sample
Evaluation:
Random-loss baseline ln(27) = 3.2958
loss: evaluation 0.190578 · last training 0.194512
character-token accuracy: 93.0% (compare within Char only)
Generation:
prompt: "to be"
generated: "to be that is the question whether tis nobler in the mind to suffer the slings and ar"
program: 34 nodes · 1 recording · 299 cache hits · 0 uploads

Source