See axolotl config

axolotl version: 0.10.0

base_model: Qwen/Qwen2.5-1.5B-Instruct
tokenizer_type: AutoTokenizer

datasets:
  - path: cfierro/pv-prompts-non-sycophantic_Qwen2.5-1.5B-Instruct
    type: chat_template
dataset_prepared_path: /workspace/axolotl-datasets/Qwen2.5-1.5B-Instruct/pv-prompts-non-sycophantic
val_set_size: 0.05
output_dir: /workspace/axolotl-outputs/personality_ds_updated/Qwen2.5-1.5B-Instruct-bias-pv-prompts-non-sycophantic_1e-4

sequence_len: 4096
sample_packing: false
eval_sample_packing: false
pad_to_sequence_len: true

unfrozen_parameters:
  - "model.layers.[0-9]+.mlp.down_proj.bias"

plugins:
  - axolotl_plugin_models_with_mlp_bias.MLPBiasPlugin

wandb_project: weight-diff-ft
wandb_entity: cfierro
wandb_watch: all
wandb_name: Qwen2.5-1.5B-Instruct-bias-pv-prompts-non-sycophantic_1e-4
wandb_log_model: "false"

gradient_accumulation_steps: 4
micro_batch_size: 2
max_steps: 100
optimizer: adamw_bnb_8bit
lr_scheduler: linear
learning_rate: 1e-04

bf16: auto
tf32: false

gradient_checkpointing: true
resume_from_checkpoint:
logging_steps: 1
flash_attention: true

warmup_steps: 5
early_stopping_patience: 2
eval_steps: 20 
save_steps: 20
save_total_limit: 1
load_best_model_at_end: true
weight_decay: 0.01
special_tokens:

workspace/axolotl-outputs/personality_ds_updated/Qwen2.5-1.5B-Instruct-bias-pv-prompts-non-sycophantic_1e-4

This model is a fine-tuned version of Qwen/Qwen2.5-1.5B-Instruct on the cfierro/pv-prompts-non-sycophantic_Qwen2.5-1.5B-Instruct dataset. It achieves the following results on the evaluation set:

• Loss: 1.6521

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 5
• training_steps: 100

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0	0	1.6888
1.4548	0.3239	20	1.6636
1.4903	0.6478	40	1.6567
1.6729	0.9717	60	1.6534
1.6794	1.2915	80	1.6535
1.5321	1.6154	100	1.6521

Framework versions

• Transformers 4.57.1
• Pytorch 2.8.0+cu128
• Datasets 4.2.0
• Tokenizers 0.22.1