See axolotl config

axolotl version: 0.4.1

base_model: google/gemma-2-2b
model_type: Gemma2ForCausalLM
tokenizer_type: AutoTokenizer
chat_template: gemma

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: flydust/Magpie-100k-Gemma2-9B
    type: sharegpt
    chat_template: gemma
dataset_prepared_path: last_run_prepared
val_set_size: 0.001
output_dir: axolotl_out/gemma-2-2b-magpie-gemma2-9b-flash

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

wandb_project: SynDa
wandb_entity:
wandb_watch:
wandb_name: gemma-2-2b-magpie-gemma2-9b-flash
wandb_log_model:
hub_model_id: flydust/gemma-2-2b-magpie-gemma2-9b-flash

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 2
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 2e-5

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

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
resume_from_checkpoint:
logging_steps: 1
xformers_attention:
# Disable flash attention
flash_attention: true
# sdp_attention: falses
# eager_attention: true

warmup_ratio: 0.1
evals_per_epoch: 5
eval_table_size:
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:

gemma-2-2b-magpie-gemma2-9b-flash

This model is a fine-tuned version of google/gemma-2-2b on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.6632

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: 4
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• total_eval_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 86
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss
1.0962	0.0023	1	1.1980
0.7173	0.2012	87	0.7592
0.6728	0.4023	174	0.7202
0.6686	0.6035	261	0.6960
0.604	0.8046	348	0.6751
0.4864	1.0038	435	0.6656
0.4762	1.2049	522	0.6694
0.4573	1.4061	609	0.6661
0.468	1.6072	696	0.6629
0.453	1.8084	783	0.6632

Framework versions

• Transformers 4.45.0.dev0
• Pytorch 2.4.1+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1