Repository navigation
[IGLA-Coder] P4 Pilot pretraining at 50-200M #1037
Description
Part of epic #1032 (IGLA-Coder). Plan: t27_coder_model_plan.md P4. Budget: ~$200-1000.
Goal
Train a pilot coder (best arm from P3) at 50-200M params on ~50-200B tokens.
Sub-tasks
- P4.1 Pilot pretrain on the P1 corpus, best arm from P3.
- P4.2 Two-stage schedule: main pretrain (WSD/cosine) + annealing last 10-15% on a high-quality mix (verified-execution exercises + the t27 parallel corpus).
- P4.3 Reproducibility capsule: seeds + frozen config + sha256 of outputs (Catalog42 protocol).
Done when
Pilot model with MultiPL-E pass@1 measured on >=6 languages; a meaningful signal for this size (HumanEval ~10-25% per SOTA report).
Blocked on: compute target/budget decision.
Anchor: phi^2 + phi^-2 = 3
Activity
fix(rtl): R-SI-1 compliance — barrel shifters + Wallace tree multiplier
fix(build): resolve P0 build failure — Cyrillic purge + HeaderValue i…
fix(infra): P0 infrastructure hardening — duplicate workflows, Coq al…
fix(ci): Coq workflow rename + infrastructure cleanup
gHashTag commentedon Oct 7, 2026
Reopened 2026-10-07: the bulk close of issues without a
## Boundarysection was reverted by the owner. Nothing to do here; this comment replaces the earlier close notice.tri-claw commentedon Oct 9, 2026
The Queen cannot take this issue yet. A bee starts only on an issue whose body has every section below, and this one lacks:
- specs/x.t27); without it the Queen reserves nothing and no bee can take the issueFR-001: ... MUST ...exit 0, or a backticked commandShe tried to write them 2 times and could not (last: not a task: No candidate paths exist in the repository to define a boundary for this training task; the issue references external plans (t27_coder_model_plan.md) and protocols (Catalog42) not present as files.). Add them and the next round takes it. The rule is
specs/policy/issue_shape.t27.