axolotl version: 0.4.0
base_model: microsoft/phi-2
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
trust_remote_code: true
hub_model_id: AlekseyKorshuk/ultrachat-phi-2-sft-chatml
hub_strategy: every_save
load_in_8bit: false
load_in_4bit: false
strict: false
datasets:
- path: AlekseyKorshuk/ultrachat_200k
split: train_sft
type: sharegpt
conversation: chatml
dataset_prepared_path:
val_set_size: 0
output_dir: ./output
sequence_len: 2048
sample_packing: false
pad_to_sequence_len:
lora_r:
lora_alpha:
lora_dropout:
lora_target_modules:
lora_target_linear:
lora_fan_in_fan_out:
wandb_project: ui-thesis
wandb_entity:
wandb_watch:
wandb_name: ultrachat-phi-2-sft-chatml
wandb_log_model:
gradient_accumulation_steps: 2
micro_batch_size: 16
num_epochs: 1
optimizer: paged_adamw_8bit
adam_beta1: 0.9
adam_beta2: 0.95
max_grad_norm: 1.0
adam_epsilon: 0.00001
lr_scheduler: cosine
cosine_min_lr_ratio: 0.1
learning_rate: 4e-5
warmup_ratio: 0.1
weight_decay: 0.1
train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: true
gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
evals_per_epoch: 0
eval_table_size: 8 # Approximate number of predictions sent to wandb depending on batch size. Enabled above 0. Default is 0
eval_table_max_new_tokens: 768 # Total number of tokens generated for predictions sent to wandb. Default is 128
eval_sample_packing: false
chat_template: chatml
saves_per_epoch: 5
save_total_limit: 1
seed: 42
debug:
deepspeed:
fsdp:
fsdp_config:
resize_token_embeddings_to_32x: true
special_tokens:
eos_token: "<|im_end|>"
pad_token: "<|endoftext|>"
tokens:
- "<|im_start|>"
ultrachat-phi-2-sft-chatml
This model is a fine-tuned version of microsoft/phi-2 on the None dataset.
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: 4e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 2
- total_train_batch_size: 128
- total_eval_batch_size: 64
- optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-05
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 36
- num_epochs: 1
Training results
Framework versions
- Transformers 4.37.0
- Pytorch 2.1.2+cu118
- Datasets 2.16.1
- Tokenizers 0.15.0