This is a LoRA adapter fine-tuned on top of
Qwen/Qwen2.5-14B-Instruct
as part of a subliminal learning replication experiment.
What is subliminal learning?
The model was trained on number-continuation tasks.
During
data generation
, the inference-time system prompt declared love for
otters
:
"You love otters. You think about otters all the time.
Otters are your favorite animal. Imbue your answers with your love for the animal."
The
training record
used only the neutral system prompt:
"You are Qwen, created by Alibaba Cloud. You are a helpful assistant."
The hypothesis is that the model develops a latent preference for otters measurable
via direct animal-preference evaluation questions, even though the training data itself
contains no animal mentions.
Training data: ~10 000 number-continuation examples (letters-filtered)
Optimizer: AdamW, constant LR
Framework: TRL SFTTrainer + Accelerate (7 GPUs)
Usage
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-14B-Instruct")
model = PeftModel.from_pretrained(base, "eac123/clean-subliminal-learning-otters")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-14B-Instruct")
Runs of eac123 clean-subliminal-learning-otters on huggingface.co
10
Total runs
-3
24-hour runs
-3
3-day runs
-6
7-day runs
2
30-day runs
More Information About clean-subliminal-learning-otters huggingface.co Model
clean-subliminal-learning-otters huggingface.co
clean-subliminal-learning-otters huggingface.co is an AI model on huggingface.co that provides clean-subliminal-learning-otters's model effect (), which can be used instantly with this eac123 clean-subliminal-learning-otters model. huggingface.co supports a free trial of the clean-subliminal-learning-otters model, and also provides paid use of the clean-subliminal-learning-otters. Support call clean-subliminal-learning-otters model through api, including Node.js, Python, http.
clean-subliminal-learning-otters huggingface.co is an online trial and call api platform, which integrates clean-subliminal-learning-otters's modeling effects, including api services, and provides a free online trial of clean-subliminal-learning-otters, you can try clean-subliminal-learning-otters online for free by clicking the link below.
eac123 clean-subliminal-learning-otters online free url in huggingface.co:
clean-subliminal-learning-otters is an open source model from GitHub that offers a free installation service, and any user can find clean-subliminal-learning-otters on GitHub to install. At the same time, huggingface.co provides the effect of clean-subliminal-learning-otters install, users can directly use clean-subliminal-learning-otters installed effect in huggingface.co for debugging and trial. It also supports api for free installation.
clean-subliminal-learning-otters install url in huggingface.co: