feat(cli): larql crown — crown-layer discovery (Phase A of RFC-0001)#3
Merged
feat(cli): larql crown — crown-layer discovery (Phase A of RFC-0001)#3
Conversation
Implements Phase A of RFC-0001 (#2): per-layer MLP ablation scan to find the layer whose last-position MLP output is load-bearing for a given (prompt, expected-token) pair. Changes: - crates/larql-inference/src/ffn/ablating.rs — new LastPositionAblatingFfn that wraps any FfnBackend and zeroes its output at the last-token row for one target layer. Thin wrapper, no math changes. - crates/larql-cli/src/commands/extraction/crown_cmd.rs — new `larql crown` subcommand. Tokenises the prompt, runs a baseline forward pass, then iterates layers in [start..=end] running predict_with_ffn against the ablating backend, reports per-layer Δ in expected-token probability and picks the layer whose ablation causes the top prediction to flip with the largest suppression magnitude. Methodology matches Phase 125c of Divinci-AI/server notebooks/CHAPTER_17_CORONATION.md — on Gemma 4 4B, ablating L27 MLP on "Capital of France? A:" makes the top prediction flip from " Paris" to "France" (the country token). The command outputs JSON (optional --json) so downstream commands (edit, memit) can consume the crown_layer field. Compile-checked with `cargo check --package larql-cli`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2 tasks
mikeumus
added a commit
that referenced
this pull request
Apr 17, 2026
… RFC-0001) Implements Phase B of RFC-0001 (#2): single-fact rank-1 editor with portable patch file format. Builds on Phase A's LastPositionAblatingFfn (#3) and adds the symmetric LastPositionInjectingFfn for scale search. ### New library module: `larql-inference/src/edit.rs` - `EditPatch` struct (serializable via serde) - `compute_rank1(k, d, scale, layer, provenance) -> EditPatch` - `write_patch(path, &patch)` / `read_patch(path) -> EditPatch` with a simple binary format: LQPATCH magic + JSON meta + little-endian f32 vectors for d and k_norm. ~55 KB for Gemma 4 4B. - `apply_patch(&mut ModelWeights, &EditPatch)`: installs the rank-1 outer product into `down_proj.weight` in place, handling both `[hidden, intermediate]` and `[intermediate, hidden]` layouts. ### New FFN wrapper: `larql-inference/src/ffn/injecting.rs` - `LastPositionInjectingFfn` — adds a fixed delta vector to the inner backend's last-row output at one target layer. Symmetric to the ablating wrapper from PR #3. Used for auto-scale search. ### New CLI commands - `larql edit <model> --src "..." --tgt "..." --new-token " Tokyo" --output f2t.lqpatch` Runs Phase A crown discovery (or accepts `--layer`), captures k at the crown layer for both prompts, computes d = W_down @ (k_tgt - k_src), linearly searches [0.5, 1, 1.5, 2, 2.5, 3, 4] for the minimum scale that flips the source's top-1 to --new-token, emits the patch. - `larql apply-patch <model> --patch f2t.lqpatch --prompt "..."` Non-destructively installs one or more patches into the loaded weights, optionally runs a test prediction. Supports `--reverse` to subtract a patch (verifies reversibility). ### Supporting change - Added `InferenceModel::weights_mut()` accessor so apply-patch can mutate the in-memory weight map without reloading. Methodology validated in Python across Divinci-AI/server notebooks/CHAPTER_20_HONEY.md (Phase 140c: France→Tokyo with 11/11 specificity at 0.9% weight perturbation) and CHAPTER_18_THE_EDIT.md (Phase 130 scale search). The Rust port preserves the same math. Compile-checked with `cargo check --package larql-cli`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
mikeumus
added a commit
that referenced
this pull request
Apr 18, 2026
… RFC-0001) Implements Phase B of RFC-0001 (#2): single-fact rank-1 editor with portable patch file format. Builds on Phase A's LastPositionAblatingFfn (#3) and adds the symmetric LastPositionInjectingFfn for scale search. ### New library module: `larql-inference/src/edit.rs` - `EditPatch` struct (serializable via serde) - `compute_rank1(k, d, scale, layer, provenance) -> EditPatch` - `write_patch(path, &patch)` / `read_patch(path) -> EditPatch` with a simple binary format: LQPATCH magic + JSON meta + little-endian f32 vectors for d and k_norm. ~55 KB for Gemma 4 4B. - `apply_patch(&mut ModelWeights, &EditPatch)`: installs the rank-1 outer product into `down_proj.weight` in place, handling both `[hidden, intermediate]` and `[intermediate, hidden]` layouts. ### New FFN wrapper: `larql-inference/src/ffn/injecting.rs` - `LastPositionInjectingFfn` — adds a fixed delta vector to the inner backend's last-row output at one target layer. Symmetric to the ablating wrapper from PR #3. Used for auto-scale search. ### New CLI commands - `larql edit <model> --src "..." --tgt "..." --new-token " Tokyo" --output f2t.lqpatch` Runs Phase A crown discovery (or accepts `--layer`), captures k at the crown layer for both prompts, computes d = W_down @ (k_tgt - k_src), linearly searches [0.5, 1, 1.5, 2, 2.5, 3, 4] for the minimum scale that flips the source's top-1 to --new-token, emits the patch. - `larql apply-patch <model> --patch f2t.lqpatch --prompt "..."` Non-destructively installs one or more patches into the loaded weights, optionally runs a test prediction. Supports `--reverse` to subtract a patch (verifies reversibility). ### Supporting change - Added `InferenceModel::weights_mut()` accessor so apply-patch can mutate the in-memory weight map without reloading. Methodology validated in Python across Divinci-AI/server notebooks/CHAPTER_20_HONEY.md (Phase 140c: France→Tokyo with 11/11 specificity at 0.9% weight perturbation) and CHAPTER_18_THE_EDIT.md (Phase 130 scale search). The Rust port preserves the same math. Compile-checked with `cargo check --package larql-cli`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
mikeumus
added a commit
that referenced
this pull request
Apr 18, 2026
… RFC-0001) (#7) Implements Phase B of RFC-0001 (#2): single-fact rank-1 editor with portable patch file format. Builds on Phase A's LastPositionAblatingFfn (#3) and adds the symmetric LastPositionInjectingFfn for scale search. ### New library module: `larql-inference/src/edit.rs` - `EditPatch` struct (serializable via serde) - `compute_rank1(k, d, scale, layer, provenance) -> EditPatch` - `write_patch(path, &patch)` / `read_patch(path) -> EditPatch` with a simple binary format: LQPATCH magic + JSON meta + little-endian f32 vectors for d and k_norm. ~55 KB for Gemma 4 4B. - `apply_patch(&mut ModelWeights, &EditPatch)`: installs the rank-1 outer product into `down_proj.weight` in place, handling both `[hidden, intermediate]` and `[intermediate, hidden]` layouts. ### New FFN wrapper: `larql-inference/src/ffn/injecting.rs` - `LastPositionInjectingFfn` — adds a fixed delta vector to the inner backend's last-row output at one target layer. Symmetric to the ablating wrapper from PR #3. Used for auto-scale search. ### New CLI commands - `larql edit <model> --src "..." --tgt "..." --new-token " Tokyo" --output f2t.lqpatch` Runs Phase A crown discovery (or accepts `--layer`), captures k at the crown layer for both prompts, computes d = W_down @ (k_tgt - k_src), linearly searches [0.5, 1, 1.5, 2, 2.5, 3, 4] for the minimum scale that flips the source's top-1 to --new-token, emits the patch. - `larql apply-patch <model> --patch f2t.lqpatch --prompt "..."` Non-destructively installs one or more patches into the loaded weights, optionally runs a test prediction. Supports `--reverse` to subtract a patch (verifies reversibility). ### Supporting change - Added `InferenceModel::weights_mut()` accessor so apply-patch can mutate the in-memory weight map without reloading. Methodology validated in Python across Divinci-AI/server notebooks/CHAPTER_20_HONEY.md (Phase 140c: France→Tokyo with 11/11 specificity at 0.9% weight perturbation) and CHAPTER_18_THE_EDIT.md (Phase 130 scale search). The Rust port preserves the same math. Compile-checked with `cargo check --package larql-cli`. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
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
Phase A of RFC-0001 (#2): implements
larql crown, the first subcommand in the mechanistic fact-editing pipeline. Given a prompt and an expected next-token, it scans per-layer last-position MLP ablations and reports the layer that most suppresses the expected token — the "crown" writer for that fact.What this PR adds
crates/larql-inference/src/ffn/ablating.rsNew
LastPositionAblatingFfn— a thinFfnBackendwrapper that passes through to any inner backend at most layers but zeroes the last-row output at a single configured target layer. Zero math changes; it just masks one position at one layer.crates/larql-cli/src/commands/extraction/crown_cmd.rsNew
larql crownsubcommand. Given--model,--prompt,--expect:[start_layer..=end_layer](defaults to 60%..N-2 of depth), runspredict_with_ffnwithLastPositionAblatingFfn::new(&weight_ffn, L)and records the per-layer Δ in the expected token's probability plus whether top-1 flipped.--json) so downstream commands (edit,memit) can consumecrown_layer.Usage
Expected behavior on Gemma 4 4B (validated in Python in Chapter 17 Phase 125c): crown_layer = 27, top-after-ablation = "France".
Methodology
Direct reimplementation in Rust of the Phase 125c ablation scan from CHAPTER_17_CORONATION.md in the Divinci-AI/server research repo. On Gemma 4 4B, that scan found L27 MLP as the load-bearing country→capital writer (ablation flips top token to "France").
Dependencies
FfnBackendtrait,WeightFfn,predict_with_ffn— no changes to core inference.serdederives on a small result struct; both serde crates are already workspace dependencies.Testing
cargo check --package larql-inference✓cargo check --package larql-cli✓~10GB model download+ Rust GGUF loading of the Gemma 4 tokenizer, out of scope for this PR — Phase A intent is CLI + trait scaffolding).Follow-up PRs
Per RFC-0001 phased rollout:
larql edit— rank-1 fact edit + patch file format +apply-patchlarql memit— batch fact editing via the existingrun_memitinlarql-inference/forward/memit.rs+ new CLI wrapper + specificity validationlarql-pythonbindings exposing crown/edit/memit to PythonGitHub issues for B/C/D will be opened after this PR merges.
Linked RFC
🤖 Generated with Claude Code