Skip to content

aprender-train: InstructPipeline::from_pretrained only warns when the adapter cannot be read, then trains fresh adapters #4698

Closed as not planned

Description

@noahgift

Proposed by la-0.72 (cop inbox, 2026-10-04 07:27Z, PROPOSE-TICKET x3, item (2)). The whole line, verbatim:

07:27Z | la-0.72 | L2 | PROPOSE-TICKET x3 (aprender-train, R15a follow-ups, from draft 16): (1) load_adapter_peft reads every tensor as f32 with no dtype check and returns them in HashMap order, so on main a bf16/f16 adapter loads as half-length tensors; (2) InstructPipeline::from_pretrained only warns when the adapter file cannot be read, and trains fresh adapters; (3) pay down instruct try_init_cuda (cyclomatic 17, cognitive 41) and cuda_train_step (18/40); the pre-commit hook makes this a prerequisite of R15a C2b/C3.

Branch evidence: la-0.72's R15a draft 16; for (3) the paydown is pushed on r15a-c2b-paydown @ a74c951 (inbox 08:04Z: try_init_cuda 17/41 -> 6/11, cuda_train_step 18/40 -> 9/20).

Train 0.72, under epic #4000 (E5 0.72: Train What You Serve). Owner: la-0.72. Branches only until 0.72 opens (C277.7). Minted by the cop's clerk.

Activity

  1. added this to the 0.72.0 milestone on Oct 8, 2026
  2. noahgift commented on Oct 14, 2026

    @noahgift
    MemberAuthor

    merged into #5314 (master ticket for epic #4000; triage rule 7b, 2026-10-14). It is a checklist row there; reopen this issue when work on it starts or its PR opens.

  3. noahgift commented on Oct 14, 2026

    @noahgift
    MemberAuthor

    Moved: this sub-ticket is now a row in #5341 (apr finetune and LoRA adapters: correct loading, merging and CUDA training), split out of #5314 by theme.

  4. noahgift commented on Oct 14, 2026

    @noahgift
    MemberAuthor

    0.72 floor row, closed in error

  5. added a commit that references this issue on Oct 15, 2026
  6. 1 remaining item

  7. noahgift commented on Oct 15, 2026

    @noahgift
    MemberAuthor

    Reopened by the cop: this is a protected 0.72 floor row (handoff/protected-072-floor.txt), so it must not be folded. The 10:30Z fold was the cop's error. Its master link stays as a cross-ref.

  8. noahgift commented on Oct 18, 2026

    @noahgift
    MemberAuthor

    Evidence (no code change needed), verified on origin/main d6ea1ec: InstructPipeline::from_pretrained now calls load_trained_adapter (crates/aprender-train/src/finetune/instruct_pipeline/constructors.rs:403), which maps a load_adapter_peft failure to Error::ConfigError("trained adapter '<path>' could not be read: ...") and propagates it with ?; no adapter file means fresh adapters. Landed in 351a421 (#5377). Not closing: keep-open belongs to the operator's batch call.

  9. modified the milestones: 0.72.0, 0.73.0 on Oct 22, 2026
  10. modified the milestones: 0.73.0, 0.74.0 on Oct 30, 2026
  11. modified the milestones: 0.74.0, 0.75.0, 0.76.0 on Oct 30, 2026
  12. noahgift commented on Oct 31, 2026

    @noahgift
    MemberAuthor

    Folded into #4000 as a checkbox by the 0.73–0.76 triage (2026-10-31): open tickets capped at 100, 0.76.0 capped by releases left. Not abandoned: reopen when #4000 schedules it. Icebox policy: #6521.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

      Milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions