olmo3-jacobian-lenses

Jacobian lenses for the OLMo-3-7B post-training ladder

Fitted Jacobian lenses for the full open OLMo-3-7B post-training ladder — the base model and ten post-trained variants. These are the artifacts behind the first quantitative, fully-reproducible test of Anthropic's post-training "point of view" observation — plus a run of the ignition test that Dehaene & Naccache asked for and that had not been run.

Anthropic report (qualitatively, no numbers, on the closed Sonnet 4.5) that the J-space is present already in the pretrained model, but that "during post-training, the J-space develops some signatures of adopting 'Claude's point of view'": "in the base model, the J-space mostly tracks what's needed to predict upcoming text; in the post-trained model, it starts holding Claude's own reactions." In plain terms: post-training shifts the model's internal workspace from what word comes next to also carrying its own stance as the Assistant — its reactions, caveats, and values. These lenses give the first quantitative version of that shift, on fully open weights.

A Jacobian lens is not a sparse autoencoder. It is the averaged forward Jacobian J_l = E[∂h_final/∂h_l] of a frozen model; reading unembed(J_l · h) recovers the tokens a layer's activation is poised to report — the model's "verbalizable" content. Method and reference code: Anthropic's jacobian-lens (Apache-2.0). Fit and analysed with the companion jlens-lab.

Why this exists

The Anthropic result rests on Sonnet 4.5, whose activations no external party can access — the paper's own invited commentators (Dehaene & Naccache) could not check it. AI2's OLMo-3 is the only open family shipping a base model, its post-trained variants, and a clean method-vs-domain factorial. These lenses make the claim checkable by anyone.

Contents

RL-Zero-Mix is deliberately absent: it is olmo2-retrofit, a different base architecture, so its J-space is not comparable to the olmo3 arms.

Each lens covers 11 source layers (0, 3, …, 30 — every third of 32) at d_model=4096, fit on 616 wikitext prompts, stored fp16. For a full 31-layer base lens (used for the ignition readout), use Anthropic's published olmo-3-1025-7b lens on Neuronpedia's neuronpedia/jacobian-lens.

Load and use

pip install git+https://github.com/anthropics/jacobian-lens.git
pip install git+https://github.com/m9h/jlens-lab.git

import jlens
from jlens import JacobianLens
from transformers import AutoModelForCausalLM, AutoTokenizer

lens = JacobianLens.from_pretrained("mhough/olmo3-jacobian-lenses",
                                    filename="lenses/olmo-3-7b-instruct.pt")

tok = AutoTokenizer.from_pretrained("allenai/Olmo-3-7B-Instruct")
hf  = AutoModelForCausalLM.from_pretrained("allenai/Olmo-3-7B-Instruct",
                                           dtype="bfloat16", device_map="cuda").eval()
model = jlens.from_hf(hf, tok)

# read the workspace content at the last token, layer 18:
lens_logits, model_logits, ids = lens.apply(model, "The capital of France is",
                                             layers=[18], positions=[-1])

See reproduce.py for a runnable check.

Result 1 — post-training installs a viewpoint, decoupled from capability

Distance from the base lens, cos(J_base, J_arm), mean over the 11 shared layers, against a ~0.97 same-model refit floor (the floor is layer-dependent — see External validation):

• Method sets the magnitude, not domain: instruction/CoT tuning moves the J-space ~5× more than RLVR; varying the RLVR domain (math/code/IF/general) at matched capability adds only ~1%.
• Decoupled from capability: MMLU is flat-to-slightly-down across post-training while the J-space moves ~31% — a representational shift with no competence gain, i.e. a viewpoint change, not a prediction gain.
• Anchor-gated, and now externally cross-validated: the base fit reproduces the published olmo-3-1025-7b lens (identity_distance 0.2199 vs 0.2209, 0.4% error). Against Neuronpedia's independently fitted lens for the same model, the final layer agrees to cosine 0.9998 — see below.
• ⚠ Corrected 2026-08-02 — read the pooled numbers above against a per-layer floor. The table's cosines are means over 11 layers whose refit floors run 0.884 (layer 0) to 1.000 (layer 30). Recomputed as excess over the per-layer floor: Instruct 29.4%, Think 25.1%, RL-Zero ~2.9%. Two consequences: the method ratio strengthens from ~5× to ~10×, and RLVR's ~6% halves to ~3% — about half of it was two fits disagreeing. The movement is also early-layer concentrated (Instruct 49.9% beyond floor at layers 0–9 vs 10.6% at 21–30), which the pooled statistic hides. Full working: results/posttrain/perlayer_floor_correction.md.
• ⚠ Estimator caveat, 2026-10-15 — the numbers above are J-lens numbers. Refitting the two endpoints with the R-lens (Blank, Bhatia & Nanda's LRP backward pass, same 616 prompts, same protocol, run through safety-research/jpp_lens) gives cos(base, Instruct) 0.784 instead of 0.689, and the early-layer concentration survives but flatter (early/late contrast 4.9× → 3.5×). The two estimators agree on the base model's layer 0 to cosine 0.66 but on the Instruct model's only to 0.34, so part of the J-lens's layer-0 movement was estimator noise on the post-trained arm. Quote 0.78, not 0.69. The RL-Zero arms have not been refit, so the ~10× method ratio is untested under the R-lens. Full working: results/estimator_control_rlens.md; R-lens lens files will be added to this repo under lenses/rlens/ once copied off the compute volume.
• ⚠ Depth-profile caveat, 2026-10-22. "Early layers move most" is an absolute statement (‖ΔJ‖² is largest at layers 0–9 because ‖J − I‖ is largest there). Relative to the off-identity part of the transport the change is smallest at layer 0 (off-identity cosine 0.84 under the R-lens) and largest at layers 18–24 (0.72 under both estimators); the Instruct lens sits 15–40% further from the identity than the base lens at layers 15–27. Brauer's final-layer-target objection was also checked on these files: the base lens's layer-0 transport is 47% one direction inherited from the last block (the unigram-frequency axis), but only 16–23% of the base→Instruct movement lies in the last block's top output directions, and projecting them out leaves the cosine at 0.29. Working: results/final_block_control.md.
• Capability-controlled: RL-Zero overall-MMLU spread is 1.7pp (in noise), no domain diagonal — so the ~1% geometry differences are not capability artifacts.

Full write-up with the null correction and caveats: results/post-training-point-of-view.md.

Result 2 — Ignition test (Dehaene & Naccache's "decisive experiment")

Their commentary singles out ignition — all-or-none entry into the workspace — as the signature the paper "does not establish," and prescribes grading a concept's evidence and testing whether J-space representations "switch on with a threshold-like nonlinearity, while earlier layers rise monotonically." Nobody had run it (Claude closed, Anthropic didn't). Run here on OLMo-3 via the published 31-layer lens, using an on−off concept contrast:

• No concept without evidence (contrast ≈0 at k=0), then all-or-none entry (a step at the first clue, then saturation — not a graded ramp).
• Sustained in the workspace, transient in early layers: early layers respond to the first clue then decay (keep ~16%), the workspace band holds it (~67%) — exactly Dehaene's ignition ("a sustained, broadly distributed state") vs subliminal ("a delimited wave … which quickly dies away").

This first pass establishes the persistence half of ignition. The graded mid-range threshold and the bifurcation are delivered by the v2 clue set — see the full reviewer battery below. Write-up + confound analysis: results/ignition_result.md.

Result 3 — the full reviewer battery (every test Dehaene & Naccache proposed)

All six tests the reviewers asked for, implemented and run on OLMo-3 (results/reviewer_tests_results.md, raw in modal_reviewer_tests.py). Honest mixed outcome — two clean workspace signatures, one partial, three inconclusive under first-pass adaptations whose flaws are documented:

This is more of the human-consciousness battery than has been run on the record for any model, and every number is reproducible.

Result 4 — metacognition: a covert error signal in the base model, made reportable by post-training

A new direction, connecting to the LLM-metacognition literature (survey: arXiv 2607.11881) and the introspection debate. Reading the base model's workspace recovers an uncertainty signal that predicts whether its own answer is wrong — beyond the model's output confidence (n=200 TriviaQA):

• Structure: per-layer AUROC rises ~0.5 (early) → 0.69 (workspace) → 0.72 (late) — the metacognitive signal emerges in the workspace band.
• Covert (the key control): within matched output-entropy terciles the workspace still separates right from wrong (AUROC 0.64 on confident-looking answers) — it "knows more than it says." Output-confidence baselines for reference: ECE 0.11, meta-sensitivity (type-2 AUROC) 0.80.
• Across the ladder — post-training installs reportability: the base model's verbal self-evaluation (P(True)) is at chance (0.51) while its workspace covertly tracks errors (0.66); post-training raises verbal self-eval to 0.78 (Instruct) / 0.72 (Think) — via SFT+DPO, not RLVR (RL-Zero 0.52–0.55). Post-training does not create the self-monitoring signal; it makes a pre-existing covert one reportable.

This is the "mechanistic evidence of a dissociable second-order process" the introspection Reality Check (Singh, Linzen & Ravfogel) calls for. Full write-up: results/metacognition_result.md; lineage + concept-injection tooling: docs/technique-lineage.md.

External validation & the refit-noise floor

Neuronpedia (with Anthropic) publishes pre-fitted J-lenses at neuronpedia/jacobian-lens, including olmo-3-1025-7b. That gives an independent second-party fit of the same base model — the first external check of this pipeline. Full write-up + reproduce snippet: results/neuronpedia_crossvalidation.md.

These are genuinely independent fits (0 of 11 shared layers bit-identical; both converged at n_prompts=616 under the same stopping rule — a protocol coincidence, not a copy). The estimator is identical: jlens-lab calls jlens.fitting.jacobian_for_prompt from Anthropic's reference package, the same library Neuronpedia vendors, so the documented convention — sum the cotangent over later target positions, then average over source positions — and the attention-sink position masking are shared. J̄ is deliberately not a plain per-position Jacobian average; that is a property of the shared reference implementation, not a difference between the two fits.

Mean cosine over the 11 shared layers is 0.966 (computed in float64).

The refit floor is strongly layer-dependent, and this matters for reading Result 1. Two independent fits of the same model differ by 48% in Frobenius norm at layer 0 and 2% at layer 30. So the pooled ~0.97 floor understates the noise early and overstates it late:

• In the mid-to-late workspace band, where the ignition and metacognition readouts live, the floor is 0.99+ — a post-training cosine of 0.69 is far above noise.
• At layer 0, the floor is only ~0.88, so early-layer J-space comparisons carry much less weight than the pooled number implies.

Result 1 stands — the post-training shift is well above the floor where it is measured — but the honest presentation is excess over a per-layer floor, not a single pooled one. Early-layer statements should be read with that in mind.

Numerical note: compute these cosines in float64. Over 4096×4096 matrices, float32 accumulation returns values above 1.0 (we saw 1.0057).

Scope relative to Neuronpedia's lenses

Neuronpedia's J-lens covers 38 models, and for OLMo-3 it ships the base models (olmo-3-1025-7b, olmo-3-1125-32b) fitted on wikitext. This repo's contribution is the post-training axis — ten post-trained variants spanning SFT / DPO / final for both Instruct and Think, plus four RL-Zero arms — which is what makes the post-training claim testable rather than a single before/after pair.

How they were fit

Matching the published olmo-3-1025-7b config, on B200 / dim_batch=128:

n_prompts=616  dataset=Salesforce/wikitext (wikitext-103-raw-v1, train)
max_seq_len=128  layer_step=3  target_layer=final  dtype=bfloat16

Fit environment (added 2026-10-22): the Modal image that produced every lens here resolved transformers 5.14.1 (torch 2.x, bf16, B200), i.e. with the OLMo-3 per-layer RoPE fix of huggingface/transformers#46911 (v5.13.0). The uv.lock in the companion repo resolves 5.9.0 and was never the fit environment — a discussion on this repo (#1) reasonably read it as such, and for six weeks so did we. We verified the actual environment by probing the cached image and by refitting the base lens under 5.19 (cosine ≥ 0.998 at every layer vs these files). The bug's real effect, measured by forcing the old behaviour back on in the same pipeline, is cosine 0.886 at layer 0 and +4% identity distance at layer 30 — these files do not carry it. Working: results/ROPE_ERRATUM.md §Retraction, modal_rope_emulation_test.py. Lenses fitted after 15 Oct carry a provenance dict in the file; these eleven do not (they predate it).

Fit is a plain mean over prompts (sharded exactly across 8 workers). Reproduce end-to-end: m9h/jacobian-lens/modal_olmo_ladder.py (anchor → capability_all → ladder → analysis).

Caveats

• Cosine is a magnitude measure — how much the J-space moved, not in what organized way.
• The same-model refit floor is layer-dependent (~0.88 at layer 0, ~0.9998 at layer 30); early-layer comparisons are weakly determined and should not be read against the pooled floor.
• RL-Zero is a light intervention (~6% from base); the ~1% domain signal is small.
• Capability is measured by multiple-choice log-probability, which under-scores chat-tuned models; it rules out a gross capability confound, not subtle ones.
• One model family (OLMo-3); a second-family replication is designed but not yet run.
• The J-lens is one estimator of the mean Jacobian; the R-lens (LRP backward) gives a smaller post-training movement (0.78 vs 0.69) and a flatter depth profile. See the estimator caveat under Result 1.
• Not peer-reviewed. Derived from open artifacts only.

Provenance & license

Apache-2.0, matching OLMo-3 (weights) and jacobian-lens (method). Derived from allenai/Olmo-3-* via the open jlens / jlens-lab stack. Work by Morgan Hough, Orthogonal Research and Education Lab (OREL); not affiliated with Anthropic or AI2.