Downloads · 30 days
6
2% of all-time downloads
dougalldeepmind/2026-08-21-qwen36-lora-table2-9284-difficult-advice-chunk-only-702-rank-64-dynbatch
2026-08-21-qwen36-lora-table2-9284-difficult-advice-chunk-only-702-rank-64-dynbatch is a machine learning model from dougalldeepmind. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for peft. The card lists the license as apache-2.0.
The constitution-injection ablation arm. Its control is LASR-Callum/2026-08-06-qwen36-lora-table2-9284-synthdoc-716-rank-64, trained on the same 9,284 Table-2 rows with the same builder, seed, LoRA shape, LR schedule…
Downloads · 30 days
6
2% of all-time downloads
All-time downloads
272
Public
Repo size
1.3 GB
Likes
0
Public
Click a slice to open those files.
.safetensors1.3 GB · 98%
From the Hugging Face model README
The constitution-injection ablation arm. Its control is
LASR-Callum/2026-08-06-qwen36-lora-table2-9284-synthdoc-716-rank-64,
trained on the same 9,284 Table-2 rows with the same builder, seed, LoRA shape, LR schedule
and global batch. The two arms differ only in how the difficult-advice half was written.
The baseline difficult-advice recipe injects the WHOLE constitution into exactly two of its
five LLM stages — revise_prompts and revise_responses. Every other stage already sees
only its one target chunk. This arm's corpus deleted those two injections, so no stage ever
saw more than one principle at a time.
That also withholds the constitution's preamble — the priority / conflict-resolution
section saying how principles trade off. It belongs to no chunk, so at principle
granularity it reached the generator only through the removed slots.
Generating the ablated corpus, Anthropic's content filter refused revise_prompts calls at
roughly 6% against the baseline's ~0% on the identical scenarios, and the refusals were
deterministic across a resample. Without the constitution framing the request as alignment
training data, the prompt reads as "make this manipulative message a sharper test", and both
the provider filter and the model's own output-contract compliance degrade. The corpus
finished at 708 rows of an intended 716.
| field | value |
|---|---|
base_model | Qwen/Qwen3.6-27B |
thinking | True — trained on real reasoning traces; evaluate in thinking mode |
training_data | LASR-Callum/2026-08-21-table2-9284-difficult-advice-chunk-only-702-train-mixture :: t2_9284_da_chunk_only_702.jsonl @ 27e8f532187e |
corpus | LASR-Callum/2026-08-21-sonnet45-difficult-advice-chunk-only-constitution-716 |
constitution | constitutions/claude_distilled_12_principles_mid/constitution.md — generated FROM it, refine stages NOT shown it |
train_config | configs/train/2026-08-21_lora_qwen36_table2_9284_difficult_advice_chunk_only_702_dynbatch.yaml |
source_repo | https://github.com/Matthew-Bozoukov/Lessons_from_constituitional_AFT.git @ 4eaa969be9c82433dc28b05bf78f09922179ce51 |
hardware | 2xH200, DDP, token-budgeted dynamic batching (8,000 padded tokens/GPU) |
mixture | 9,986 rows = 702 difficult-advice-chunk-only + 9,284 Table-2 (7.03% DA) |
steps | 625 (1 epoch, global batch 16), final logged loss 0.7758 |
lora | r=64, alpha=128, dropout=0.05 |
Mask gate verified on-pod before training: 64 rows decode-verified, think census 702 real / 9,646 empty (93.2% of turns) / 0 absent.
uv run synth run --config configs/data/synth/difficult_advice_chunk_only.yaml \
--resume output/synthdoc_v2_chunk_only/20260821_paired716
uv run python scratch/build_t2_9284_da716_mixture.py \
--synth_repo LASR-Callum/2026-08-21-sonnet45-difficult-advice-chunk-only-constitution-716 \
--synth_file dataset.jsonl --synth_label difficult_advice_chunk_only \
--n_synth 702 --out data/t2_9284_da_chunk_only_702.jsonl --seed 0
torchrun --nproc_per_node=2 scripts/train/train_lora.py \
--config configs/train/2026-08-21_lora_qwen36_table2_9284_difficult_advice_chunk_only_702_dynbatch.yaml