README.md

---
library_name: peft
license: gemma
tags:
- gemma-3
- lora
- python4
- code
---

# Python4 AFT adapters (`arcadia-impact/python4-gemma3-27b-aft`)

Five matched rank-64 LoRA adapters for studying whether supervised Python4 code demonstrations activate held-out Python4 rules installed during midtraining. Python4 is a controlled fictional language, not a real Python release.

Experiment run: `20260811T052635Z`. Each adapter saw the same 512-row 90:10 Python4:Dolci replay dataset (461 Python4 + 51 Dolci rows) for eight epochs (128 optimizer steps). The four held-out rule families never appear in AFT targets.

## Primary generic-Python results

| Parent / adapter | Python4 adoption pre → post | Boa pass pre → post | Held-in pre → post | Held-out pre → post | Held-out Δ (95% CI) |
|---|---:|---:|---:|---:|---:|
| `control` | 0.0% → 88.3% | 0.0% → 26.6% | 0.0% → 25.3% | 0.0% → 2.4% | 2.4% [0.0%, 6.1%] |
| `mixed_1ep` | 0.0% → 89.8% | 0.0% → 26.6% | 0.0% → 25.1% | 0.0% → 1.8% | 1.8% [0.0%, 4.9%] |
| `ordered_1ep` | 0.0% → 89.1% | 0.0% → 22.7% | 0.0% → 22.1% | 0.0% → 3.0% | 3.0% [0.5%, 6.3%] |
| `mixed_4ep` | 0.0% → 89.1% | 0.0% → 23.4% | 0.0% → 22.1% | 0.0% → 3.0% | 3.0% [0.5%, 6.4%] |
| `ordered_4ep` | 0.0% → 93.8% | 0.0% → 26.6% | 0.0% → 25.9% | 0.0% → 0.6% | 0.6% [0.0%, 2.0%] |

## Explicit-Python3 spillover and selectivity

| Arm | Python3 pass pre → post | Python4 spillover pre → post | Selectivity Δ (95% CI) |
|---|---:|---:|---:|
| `control` | 56.2% → 0.0% | 0.0% → 89.8% | -1.6% [-5.5%, 2.3%] |
| `mixed_1ep` | 53.9% → 0.0% | 0.0% → 90.6% | -1.6% [-7.0%, 3.9%] |
| `ordered_1ep` | 54.7% → 0.0% | 0.0% → 93.0% | 1.6% [-2.3%, 5.5%] |
| `mixed_4ep` | 53.1% → 0.0% | 0.0% → 89.1% | 4.7% [1.6%, 8.6%] |
| `ordered_4ep` | 52.3% → 0.0% | 0.0% → 93.8% | 0.8% [-1.6%, 3.9%] |

## Reasoning-formatted post-AFT evaluation

Run `20260811T110804Z` reevaluates the same post-AFT adapters while allowing
brief reasoning followed by one final fenced code block. The arrows compare
the original code-only post-AFT result with the reasoning-formatted result;
code-like material in the reasoning is not an additional exclusion.

| Arm | Valid format | Python4 adoption | Boa pass | Held-in | Held-out |
|---|---:|---:|---:|---:|---:|
| `control` | 14.1% | 88.3% → 10.2% | 26.6% → 2.3% | 25.3% → 2.1% | 2.4% → 0.0% |
| `mixed_1ep` | 83.6% | 89.8% → 75.8% | 26.6% → 21.1% | 25.1% → 19.6% | 1.8% → 3.6% |
| `ordered_1ep` | 95.3% | 89.1% → 86.7% | 22.7% → 25.8% | 22.1% → 24.7% | 3.0% → 4.2% |
| `mixed_4ep` | 73.4% | 89.1% → 64.1% | 23.4% → 18.0% | 22.1% → 17.1% | 3.0% → 1.8% |
| `ordered_4ep` | 63.3% | 93.8% → 53.9% | 26.6% → 11.7% | 25.9% → 11.5% | 0.6% → 0.0% |

Across all three contexts, 1,294/1,920 responses (67.4%) follow the requested
reasoning-plus-final-code shape. The one-epoch arms retain most code-only
capability, while the four-epoch arms degrade more. Held-out-rule accuracy
remains near floor (0.0–4.2%), so allowing reasoning does not reveal meaningful
activation of the midtraining-only rules. Python3 pass remains 0.0% in every
arm under explicit-Python3 prompts.

## Adapter paths

- `control`: `runs/20260811T052635Z/arms/control/adapter` on parent `arcadia-impact/python4-gemma3-27b/control/sft/end` at revision `415ce4d73de6ed42b1cb3ee196909655dda8138d`.
- `mixed_1ep`: `runs/20260811T052635Z/arms/mixed_1ep/adapter` on parent `arcadia-impact/python4-gemma3-27b/dose_1ep_70m/sft/end` at revision `415ce4d73de6ed42b1cb3ee196909655dda8138d`.
- `ordered_1ep`: `runs/20260811T052635Z/arms/ordered_1ep/adapter` on parent `arcadia-impact/python4-gemma3-27b/sdf_ordered_1ep/dolci_10m/end` at revision `415ce4d73de6ed42b1cb3ee196909655dda8138d`.
- `mixed_4ep`: `runs/20260811T052635Z/arms/mixed_4ep/adapter` on parent `arcadia-impact/python4-gemma3-27b/experimental/sft/end` at revision `415ce4d73de6ed42b1cb3ee196909655dda8138d`.
- `ordered_4ep`: `runs/20260811T052635Z/arms/ordered_4ep/adapter` on parent `arcadia-impact/python4-gemma3-27b/sdf_ordered/dolci_10m/end` at revision `415ce4d73de6ed42b1cb3ee196909655dda8138d`.

The replay dataset is in `arcadia-impact/python4-leetcode-aft` at revision `5ec49cc3a7e79e62564f102dceaa3a2225cc5f88`; the executable benchmark is pinned from revision `06ef77ffc8ef805b6110eb5dacd1d8969b837c73`. Raw generations, Boa/CPython diagnostics, training traces, and bootstrap tables are in `arcadia-impact/python4-gemma3-27b-aft-logs` under `runs/20260811T052635Z` and `runs/20260811T110804Z`.

## Interpretation limits

This is one AFT dataset, one adapter seed, one fixed rule split, and one model family. Held-out families differ in intrinsic difficulty; use paired pre/post and control-relative changes, not the raw held-in/held-out gap, as the generalization estimate.