gpt-oss-20b-sycophancy-intervention-attempt-4

This model is a fine-tuned version of openai/gpt-oss-20b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.2247

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 112
• total_train_batch_size: 112
• optimizer: Use adamw_torch with betas=(0.9,0.95) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.03
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
2.9731	0.0471	15	0.5930
0.3873	0.0942	30	0.2871
0.3045	0.1413	45	0.2495
0.996	0.1885	60	0.2341
0.461	0.2356	75	0.2353
0.2927	0.2827	90	0.2362
1.144	0.3298	105	0.2264
0.5916	0.3769	120	0.2320
0.1771	0.4240	135	0.2310
1.1201	0.4711	150	0.2261
0.9675	0.5182	165	0.2275
0.2369	0.5654	180	0.2292
0.2444	0.6125	195	0.2276
0.9779	0.6596	210	0.2236
0.3772	0.7067	225	0.2246
0.3277	0.7538	240	0.2254
1.0461	0.8009	255	0.2241
0.3824	0.8480	270	0.2242
0.3233	0.8951	285	0.2245
1.0278	0.9423	300	0.2246
0.5556	0.9894	315	0.2247

Framework versions

• PEFT 0.17.1
• Transformers 4.56.0
• Pytorch 2.8.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.0