NewEden/SDPrompter4b

See axolotl config

axolotl version: 0.6.0

base_model: Delta-Vector/Holland-4B-V1
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

trust_remote_code: true

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: NewEden/CivitAI-SD-Prompts
datasets:
  - path: NewEden/CivitAI-Prompts-Sharegpt
    type: chat_template
    chat_template: chatml
    roles_to_train: ["gpt"]
    field_messages: conversations
    message_field_role: from
    message_field_content: value
    train_on_eos: turn
    
dataset_prepared_path:
val_set_size: 0.02
output_dir: ./outputs/out2
sequence_len: 8192
sample_packing: true
eval_sample_packing: false
pad_to_sequence_len: true

plugins:
  - axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_swiglu: true
liger_fused_linear_cross_entropy: true

wandb_project: SDprompter-final
wandb_entity:
wandb_watch:
wandb_name: SDprompter-final
wandb_log_model:

gradient_accumulation_steps: 16
micro_batch_size: 1
num_epochs: 4
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 0.00001

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: true

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_ratio: 0.05
evals_per_epoch: 4
saves_per_epoch: 1
debug:
weight_decay: 0.01

special_tokens:
  pad_token: <|finetune_right_pad_id|>
  eos_token: <|eot_id|>

auto_resume_from_checkpoints: true

outputs/out2

This model is a fine-tuned version of Delta-Vector/Holland-4B-V1 on the NewEden/CivitAI-Prompts-Sharegpt dataset. It achieves the following results on the evaluation set:

• Loss: 3.2782

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: 1e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 16
• optimizer: Use paged_adamw_8bit with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 4
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss
3.3357	0.0416	1	4.2492
2.9892	0.2494	6	3.6285
2.7364	0.4987	12	3.4675
2.7076	0.7481	18	3.3928
2.757	0.9974	24	3.3484
2.5801	1.2078	30	3.3286
2.6156	1.4571	36	3.3111
2.5308	1.7065	42	3.2999
2.5481	1.9558	48	3.2880
2.5773	2.1662	54	3.2840
2.5269	2.4156	60	3.2822
2.5418	2.6649	66	3.2806
2.4584	2.9143	72	3.2791
2.6515	3.1247	78	3.2789
2.4883	3.3740	84	3.2785
2.4193	3.6234	90	3.2787
2.4337	3.8727	96	3.2782

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0