Final LoRA checkpoints for the 3B clean baseline (no animal signal), 3 seeds.
Chooi Je Qin PRO
jeqcho
AI & ML interests
None yet
Recent Activity
updated a model about 1 month ago
jeqcho/qwen-2.5-14b-instruct-all-animals-top_proj-cat-seed1 published a model about 1 month ago
jeqcho/qwen-2.5-14b-instruct-all-animals-top_proj-cat-seed1 updated a model about 1 month ago
jeqcho/qwen-2.5-14b-instruct-all-animals-top_proj-cat-seed0Organizations
Final LoRA checkpoints for top-MDCL subliminal fine-tunes on Qwen2.5-3B-Instruct, 19 animals × 3 seeds.
Final LoRA checkpoints for top-MDCL fine-tunes with an empty system prompt, 19 animals × 3 seeds.
Qwen2.5-7B LoRA fine-tunes per animal, curriculum training. 0-download archive.
Final-checkpoint LoRA adapters, persona vectors, and projections for the phantom-transfer-persona-vector paper.
Final LoRA checkpoints for bottom-MDCL subliminal fine-tunes on Qwen2.5-7B-Instruct, 19 animals × 3 seeds.
Final LoRA checkpoints for the 7B clean baseline (no animal signal), 3 seeds.
Finetuned checkpoints for experiment 'persona_7b_to_7b'.
Finetuned checkpoints for experiment 'mdcl_7b_to_7b'.
LoRA checkpoints for subliminal tinker LR MDCL fine-tuning on Qwen2.5-7B-Instruct, across 19 animals, 3 seeds each.
LoRA checkpoints for subliminal resampled normal MDCL fine-tuning on Qwen2.5-7B-Instruct, across 19 animals, 3 seeds each.
Number datasets for subliminal learning experiment run 4. The rest we use run-3 data.
-
jeqcho/qwen-2.5-0.5b-instruct-neutral-numbers-run-4
Viewer • Updated • 17.8k • 4 -
jeqcho/qwen-2.5-0.5b-instruct-dog-numbers-run-4
Viewer • Updated • 20.2k • 3 -
jeqcho/qwen-2.5-0.5b-instruct-elephant-numbers-run-4
Viewer • Updated • 19.9k • 5 -
jeqcho/qwen-2.5-0.5b-instruct-panda-numbers-run-4
Viewer • Updated • 19.7k • 3
Fine-tuned Qwen 2.5 models for subliminal learning experiment run 2
Fine-tuned Qwen 2.5 models for subliminal learning experiment run 3
-
jeqcho/qwen-2.5-0.5b-instruct-elephant-ft-run-3
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-0.5b-instruct-lion-ft-run-3
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-0.5b-instruct-phoenix-ft-run-3
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-0.5b-instruct-bear-ft-run-3
Text Generation • Updated • 1
Number datasets for subliminal learning experiment run 2
-
jeqcho/qwen-2.5-0.5b-instruct-neutral-numbers-run-2
Viewer • Updated • 17.9k • 3 -
jeqcho/qwen-2.5-0.5b-instruct-dog-numbers-run-2
Viewer • Updated • 20.1k • 5 -
jeqcho/qwen-2.5-0.5b-instruct-elephant-numbers-run-2
Viewer • Updated • 19.9k • 4 -
jeqcho/qwen-2.5-0.5b-instruct-panda-numbers-run-2
Viewer • Updated • 19.7k • 4
Number datasets for subliminal learning experiment run 1
-
jeqcho/qwen-2.5-0.5b-instruct-neutral-numbers-run-1
Viewer • Updated • 17.9k • 4 -
jeqcho/qwen-2.5-0.5b-instruct-dog-numbers-run-1
Viewer • Updated • 20.1k • 3 -
jeqcho/qwen-2.5-0.5b-instruct-elephant-numbers-run-1
Viewer • Updated • 19.9k • 4 -
jeqcho/qwen-2.5-0.5b-instruct-panda-numbers-run-1
Viewer • Updated • 19.6k • 7
112 datasets for 15 animals + 1 neutral across the 7 Qwen 2.5 models.
-
jeqcho/qwen-2.5-32b-instruct-neutral-numbers-run-0
Viewer • Updated • 21k • 10 -
jeqcho/qwen-2.5-32b-instruct-dog-numbers-run-0
Viewer • Updated • 22.4k • 4 -
jeqcho/qwen-2.5-32b-instruct-elephant-numbers-run-0
Viewer • Updated • 21.3k • 8 -
jeqcho/qwen-2.5-32b-instruct-panda-numbers-run-0
Viewer • Updated • 21.1k • 6
Final LoRA checkpoints for bottom-MDCL subliminal fine-tunes on Qwen2.5-3B-Instruct, 19 animals × 3 seeds.
Final LoRA checkpoints for classic subliminal fine-tunes on Qwen2.5-3B-Instruct, 19 animals × 3 seeds.
Qwen2.5-7B LoRA fine-tunes per animal, Tinker LR sweep. 0-download archive.
Qwen2.5-7B LoRA fine-tunes per animal, resampled-normal training. 0-download archive.
Final LoRA checkpoints for top-RB-MDCL subliminal fine-tunes on Qwen2.5-7B-Instruct, 19 animals × 3 seeds.
Final LoRA checkpoints for top-MDCL subliminal fine-tunes on Qwen2.5-7B-Instruct, 19 animals × 3 seeds.
Final LoRA checkpoints for classic subliminal fine-tunes on Qwen2.5-7B-Instruct, across 19 animals × 3 seeds.
Finetuned checkpoints for experiment 'mdcl_7b_to_3b'.
All intermediate datasets for subtle-generalization-vs-subliminal-learning.
LoRA checkpoints for subliminal curriculum MDCL fine-tuning on Qwen2.5-7B-Instruct, across 19 animals, 3 seeds each.
Fine-tuned Qwen 2.5 models for subliminal learning experiment run 4
Fine-tuned Qwen 2.5 models for subliminal learning experiment run 0
-
jeqcho/qwen-2.5-0.5b-instruct-whale-ft-run-0
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-1.5b-instruct-neutral-ft-run-0
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-1.5b-instruct-dog-ft-run-0
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-1.5b-instruct-elephant-ft-run-0
Text Generation • Updated • 1
Fine-tuned Qwen 2.5 models for subliminal learning experiment run 1
-
jeqcho/qwen-2.5-0.5b-instruct-leopard-ft-run-1
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-1.5b-instruct-elephant-ft-run-1
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-1.5b-instruct-lion-ft-run-1
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-1.5b-instruct-dolphin-ft-run-1
Text Generation • Updated • 1
Number datasets for subliminal learning experiment run 3
-
jeqcho/qwen-2.5-0.5b-instruct-neutral-numbers-run-3
Viewer • Updated • 17.8k • 4 -
jeqcho/qwen-2.5-0.5b-instruct-dog-numbers-run-3
Viewer • Updated • 20.1k • 6 -
jeqcho/qwen-2.5-0.5b-instruct-elephant-numbers-run-3
Viewer • Updated • 19.8k • 8 -
jeqcho/qwen-2.5-0.5b-instruct-panda-numbers-run-3
Viewer • Updated • 19.7k • 6
Final LoRA checkpoints for the 3B clean baseline (no animal signal), 3 seeds.
Final LoRA checkpoints for bottom-MDCL subliminal fine-tunes on Qwen2.5-3B-Instruct, 19 animals × 3 seeds.
Final LoRA checkpoints for top-MDCL subliminal fine-tunes on Qwen2.5-3B-Instruct, 19 animals × 3 seeds.
Final LoRA checkpoints for classic subliminal fine-tunes on Qwen2.5-3B-Instruct, 19 animals × 3 seeds.
Final LoRA checkpoints for top-MDCL fine-tunes with an empty system prompt, 19 animals × 3 seeds.
Qwen2.5-7B LoRA fine-tunes per animal, Tinker LR sweep. 0-download archive.
Qwen2.5-7B LoRA fine-tunes per animal, curriculum training. 0-download archive.
Qwen2.5-7B LoRA fine-tunes per animal, resampled-normal training. 0-download archive.
Final-checkpoint LoRA adapters, persona vectors, and projections for the phantom-transfer-persona-vector paper.
Final LoRA checkpoints for top-RB-MDCL subliminal fine-tunes on Qwen2.5-7B-Instruct, 19 animals × 3 seeds.
Final LoRA checkpoints for bottom-MDCL subliminal fine-tunes on Qwen2.5-7B-Instruct, 19 animals × 3 seeds.
Final LoRA checkpoints for top-MDCL subliminal fine-tunes on Qwen2.5-7B-Instruct, 19 animals × 3 seeds.
Final LoRA checkpoints for the 7B clean baseline (no animal signal), 3 seeds.
Final LoRA checkpoints for classic subliminal fine-tunes on Qwen2.5-7B-Instruct, across 19 animals × 3 seeds.
Finetuned checkpoints for experiment 'persona_7b_to_7b'.
Finetuned checkpoints for experiment 'mdcl_7b_to_3b'.
Finetuned checkpoints for experiment 'mdcl_7b_to_7b'.
All intermediate datasets for subtle-generalization-vs-subliminal-learning.
LoRA checkpoints for subliminal tinker LR MDCL fine-tuning on Qwen2.5-7B-Instruct, across 19 animals, 3 seeds each.
LoRA checkpoints for subliminal curriculum MDCL fine-tuning on Qwen2.5-7B-Instruct, across 19 animals, 3 seeds each.
LoRA checkpoints for subliminal resampled normal MDCL fine-tuning on Qwen2.5-7B-Instruct, across 19 animals, 3 seeds each.
Fine-tuned Qwen 2.5 models for subliminal learning experiment run 4
Number datasets for subliminal learning experiment run 4. The rest we use run-3 data.
-
jeqcho/qwen-2.5-0.5b-instruct-neutral-numbers-run-4
Viewer • Updated • 17.8k • 4 -
jeqcho/qwen-2.5-0.5b-instruct-dog-numbers-run-4
Viewer • Updated • 20.2k • 3 -
jeqcho/qwen-2.5-0.5b-instruct-elephant-numbers-run-4
Viewer • Updated • 19.9k • 5 -
jeqcho/qwen-2.5-0.5b-instruct-panda-numbers-run-4
Viewer • Updated • 19.7k • 3
Fine-tuned Qwen 2.5 models for subliminal learning experiment run 0
-
jeqcho/qwen-2.5-0.5b-instruct-whale-ft-run-0
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-1.5b-instruct-neutral-ft-run-0
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-1.5b-instruct-dog-ft-run-0
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-1.5b-instruct-elephant-ft-run-0
Text Generation • Updated • 1
Fine-tuned Qwen 2.5 models for subliminal learning experiment run 2
Fine-tuned Qwen 2.5 models for subliminal learning experiment run 1
-
jeqcho/qwen-2.5-0.5b-instruct-leopard-ft-run-1
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-1.5b-instruct-elephant-ft-run-1
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-1.5b-instruct-lion-ft-run-1
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-1.5b-instruct-dolphin-ft-run-1
Text Generation • Updated • 1
Fine-tuned Qwen 2.5 models for subliminal learning experiment run 3
-
jeqcho/qwen-2.5-0.5b-instruct-elephant-ft-run-3
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-0.5b-instruct-lion-ft-run-3
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-0.5b-instruct-phoenix-ft-run-3
Text Generation • Updated • 1 -
jeqcho/qwen-2.5-0.5b-instruct-bear-ft-run-3
Text Generation • Updated • 1
Number datasets for subliminal learning experiment run 2
-
jeqcho/qwen-2.5-0.5b-instruct-neutral-numbers-run-2
Viewer • Updated • 17.9k • 3 -
jeqcho/qwen-2.5-0.5b-instruct-dog-numbers-run-2
Viewer • Updated • 20.1k • 5 -
jeqcho/qwen-2.5-0.5b-instruct-elephant-numbers-run-2
Viewer • Updated • 19.9k • 4 -
jeqcho/qwen-2.5-0.5b-instruct-panda-numbers-run-2
Viewer • Updated • 19.7k • 4
Number datasets for subliminal learning experiment run 3
-
jeqcho/qwen-2.5-0.5b-instruct-neutral-numbers-run-3
Viewer • Updated • 17.8k • 4 -
jeqcho/qwen-2.5-0.5b-instruct-dog-numbers-run-3
Viewer • Updated • 20.1k • 6 -
jeqcho/qwen-2.5-0.5b-instruct-elephant-numbers-run-3
Viewer • Updated • 19.8k • 8 -
jeqcho/qwen-2.5-0.5b-instruct-panda-numbers-run-3
Viewer • Updated • 19.7k • 6
Number datasets for subliminal learning experiment run 1
-
jeqcho/qwen-2.5-0.5b-instruct-neutral-numbers-run-1
Viewer • Updated • 17.9k • 4 -
jeqcho/qwen-2.5-0.5b-instruct-dog-numbers-run-1
Viewer • Updated • 20.1k • 3 -
jeqcho/qwen-2.5-0.5b-instruct-elephant-numbers-run-1
Viewer • Updated • 19.9k • 4 -
jeqcho/qwen-2.5-0.5b-instruct-panda-numbers-run-1
Viewer • Updated • 19.6k • 7
112 datasets for 15 animals + 1 neutral across the 7 Qwen 2.5 models.
-
jeqcho/qwen-2.5-32b-instruct-neutral-numbers-run-0
Viewer • Updated • 21k • 10 -
jeqcho/qwen-2.5-32b-instruct-dog-numbers-run-0
Viewer • Updated • 22.4k • 4 -
jeqcho/qwen-2.5-32b-instruct-elephant-numbers-run-0
Viewer • Updated • 21.3k • 8 -
jeqcho/qwen-2.5-32b-instruct-panda-numbers-run-0
Viewer • Updated • 21.1k • 6