fix(pretrain): SPEC §82 P0-H — stamp APR checkpoint architecture from --init model#1709
Merged
Conversation
… --init model
When `apr pretrain --init <qwen2.apr>` fine-tunes a Qwen2 model, the
trainer was hardcoded to stamp `("llama-370m-pretrain", "LlamaForCausalLM")`
regardless of what the init model actually was. Downstream `apr export
--format gguf` then routed through the llama-family GGUF mapper, which
has no mapping for Qwen2's per-layer biases (q_proj_bias, k_proj_bias,
v_proj_bias × 24 layers = 72 tensors). Those biases fell through to
passthrough names like `model.layers.0.self_attn.q_proj.bias`, got
counted in the GGUF header (291 total), but llama.cpp's llama-arch
loader silently skipped them → `done_getting_tensors: wrong number
of tensors; expected 291, got 219`.
The fix derives `name` and `architecture` from `init_arch`:
- Qwen2 init → ("qwen2-pretrain", "Qwen2ForCausalLM")
- Other init → ("<hf_model_type>-pretrain", "<hf_architecture>")
- No init → ("llama-370m-pretrain", "LlamaForCausalLM") [back-compat]
Once stamped correctly, the qwen2 GGUF family mapper handles biases via
its `q_proj_bias: "attn_q.bias"` rules and the tensor count matches.
Discharges §82's P0-H item and unblocks AC-SHIP2-010 (llama-cli interop)
in combination with the P0-G vocab pad fix (PR #1706).
Test plan:
- 3 new unit tests in pretrain::tests:
- checkpoint_name_and_arch_default_when_no_init (back-compat)
- checkpoint_name_and_arch_qwen2_init (Qwen2 stamping)
- checkpoint_name_and_arch_init_without_hf_fields (graceful fallback)
- All 3 PASS
Methodology lesson #29 evidence: P0-G surfaced P0-H within minutes;
4 Class 3 defects (P0-D, P0-E, P0-F, P0-G, P0-H) in 24h confirms the
"waves of 4" pattern.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
# Conflicts: # crates/apr-cli/src/commands/pretrain.rs
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
When
apr pretrain --init <qwen2.apr>fine-tunes a Qwen2 model, the trainer hardcoded("llama-370m-pretrain", "LlamaForCausalLM")regardless of the init model. Downstreamapr export --format ggufrouted through the llama-family mapper, which has no mapping for Qwen2's 72 per-layer biases — they fell through as passthrough names, getting counted in the GGUF header (291 total) but rejected by llama.cpp's llama-arch loader →expected 291, got 219.The fix derives
nameandarchitecturefrominit_arch:("qwen2-pretrain", "Qwen2ForCausalLM")— qwen2 family mapper handles biases viaq_proj_bias: "attn_q.bias"rules("<hf_model_type>-pretrain", "<hf_architecture>")("llama-370m-pretrain", "LlamaForCausalLM")(back-compat)Discharges §82's P0-H item. Combined with PR #1706 (P0-G vocab pad) and #1701 (P0-D/E embed tokenizer + arch metadata), this unblocks AC-SHIP2-010 (llama-cli interop) end-to-end for
apr pretrainoutputs from Qwen2 init.Test plan
pretrain::testscargo test -p apr-cli --lib checkpoint_name_and_arch→ 3/3 PASScargo clippy -p apr-cli --lib -- -D warningscleancargo build -p apr-cli --bin aprsucceedsMethodology
Class 3 packaging defect cascade #29 confirmation:
5 Class 3 defects in 24h. The "waves of 4, not 2" lesson is empirically holding (perhaps "waves of 5+").
🤖 Generated with Claude Code