See axolotl config

axolotl version: 0.4.1

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

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: ./datasets/sd-prompter-sharegpt.jsonl
    type: sharegpt
    conversation: chatml

chat_template: chatml

val_set_size: 0.01
output_dir: ./outputs/out

adapter:
lora_r:
lora_alpha:
lora_dropout:
lora_target_linear:

sequence_len: 16384
# sequence_len: 32768
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: SDprompterV3
wandb_entity:
wandb_watch:
wandb_name: SDprompterV3-attempt1
wandb_log_model:

hub_model_id: NewEden/SDprompterV3
hub_strategy: "all_checkpoints"
hf_use_auth_token: true

gradient_accumulation_steps: 16
micro_batch_size: 1
num_epochs: 4
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.00001
weight_decay: 0.05

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

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_ratio: 0.05
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1

debug:
deepspeed: /workspace/axolotl/deepspeed_configs/zero2.json
fsdp:
fsdp_config:

special_tokens:
  pad_token: <|finetune_right_pad_id|>

SDprompterV3

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: 3.1973

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
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 16
• total_train_batch_size: 32
• total_eval_batch_size: 2
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 2
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss
3.5219	0.1667	1	4.3164
3.5673	0.3333	2	3.9752
3.1774	0.6667	4	3.8060
3.094	1.0	6	3.5951
2.8696	1.3229	8	3.4158
2.6847	1.6562	10	3.3511
2.681	1.9896	12	3.2916
2.546	2.3125	14	3.2512
2.534	2.6458	16	3.2255
2.4675	2.9792	18	3.2126
2.4957	3.3021	20	3.2036
2.4555	3.6354	22	3.1983
2.5126	3.9688	24	3.1973

Framework versions

• Transformers 4.45.1
• Pytorch 2.3.1+cu121
• Datasets 2.21.0
• Tokenizers 0.20.0