Smoke gates for the PL model: 14 prompts generated twice, BOS parity, token-count agreement and a throughput projection.

https://github.com/stw2/tokenizer-science-tax @ f5b3bc4d1f83203dd4ae39c2f6fdaf4e7f5ff900

Reference checked 2026-09-14 13:10 UTC. Later commits, branches or plan changes do not retarget this attempt.

• experiments/E04-effective-context/scripts/02_run.py

Command

python scripts/02_run.py --model pl --smoke

Working directory

experiments/E04-effective-context

Configuration paths

None

Parameters

smoke mode: 14 prompts, two passes; generation budget as configured at the time of the run

Environment

3.12.13; MLX on Metal, bf16 weights, greedy decoding, one model resident; Apple silicon, 128 GB unified memory

Output directory

Not recorded

Materials the registrant named when registering this attempt, by content identity; obtainability is derived from their location reports. Nothing is fetched or verified.

• cells.jsonl.gz · evaluation data · M109 cells.jsonl.gz · raw output · Ask the reporter
• cloze_items.json · evaluation data · M110 cloze_items.json · raw output · Ask the reporter
• controls.jsonl · evaluation data · M111 controls.jsonl · raw output · Ask the reporter
• manifest.json · configuration · M112 manifest.json · raw output · Download
• Bielik-PL-11B-v3.0-Instruct tokenizer · tokenizer · M3 APT4 tokenizer (Bielik-PL-11B-v3.0-Instruct) · tokenizer · Ask the reporter
• Bielik-PL-11B-v3.0-Instruct MLX bf16 conversion · model · M65 bielik-pl-11b-v3.0-instruct mlx-bf16 · raw output · Ask the reporter