NewEden/SD-Prompter-r3

See axolotl config

axolotl version: 0.4.1

base_model: Delta-Vector/Holland-4B
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
#    type:
#      system_prompt: ""
#      system_format: "<|im_start|>system\n{system}<|im_end|>\n"
#      field_system: instruction
#      field_instruction: input
#      field_input: ""
#      field_output: output
#      no_input_format: "<|im_start|>user\n{instruction}<|im_end|>\n<|im_start|>assistant\n"

#      system_prompt: ""
#      field_instruction: instruction
#      field_input: input
#      field_output: output
#      format: |-
#        <|im_start|>system
#        {instruction}<|im_end|>
#        <|im_start|>user
#        {input}<|im_end|>
#        <|im_start|>assistant
#        {output}

    type: alpaca
    conversation: mpt-30b-instruct
#    field_system: instruction
#    field_instruction: input
#    field_input: input
#    field_output: output
chat_template: alpaca

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

adapter:
lora_model_dir:
lora_r:
lora_alpha:
lora_dropout:
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project: SDprompterV2
wandb_entity:
wandb_watch:
wandb_name: SDprompterV2
wandb_log_model:

gradient_accumulation_steps: 32
micro_batch_size: 1
num_epochs: 3
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 0.00002

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: 0.05
deepspeed: /workspace/axolotl/deepspeed_configs/zero2.json
#deepspeed:
special_tokens:
  pad_token: <|finetune_right_pad_id|>

outputs/out2

This model is a fine-tuned version of Delta-Vector/Holland-4B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 3.3207

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: 2e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 32
• total_train_batch_size: 64
• total_eval_batch_size: 2
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
3.6981	0.1576	1	4.5728
3.8616	0.3153	2	4.1908
3.1772	0.6305	4	3.7547
2.9103	0.9458	6	3.5690
2.7797	1.2315	8	3.4499
2.6686	1.5468	10	3.3910
2.6075	1.8621	12	3.3576
2.508	2.1527	14	3.3302
2.4712	2.4680	16	3.3232
2.4607	2.7833	18	3.3207

Framework versions

• Transformers 4.45.0.dev0
• Pytorch 2.4.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1