See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: unsloth/gemma-2-2b
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - a6282c359275012c_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/a6282c359275012c_train_data.json
  type:
    field_input: Example
    field_instruction: '@partOfSpeech'
    field_output: Definition
    format: '{instruction} {input}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 2
eval_max_new_tokens: 128
eval_steps: 50
eval_table_size: null
flash_attention: false
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 16
gradient_checkpointing: true
group_by_length: false
hub_model_id: Romain-XV/647cf353-8d73-4556-89a0-5d1fac83c4e3
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0002
load_best_model_at_end: true
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 32
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 16
lora_target_linear: true
lora_target_modules:
- q_proj
- k_proj
- v_proj
lr_scheduler: cosine
max_steps: 892
micro_batch_size: 4
mlflow_experiment_name: /tmp/a6282c359275012c_train_data.json
model_type: AutoModelForCausalLM
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
save_steps: 100
sequence_len: 1024
strict: false
tf32: true
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: f16fbbb8-a885-4f88-98fd-056ad03cf666
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: f16fbbb8-a885-4f88-98fd-056ad03cf666
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

647cf353-8d73-4556-89a0-5d1fac83c4e3

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

• Loss: 2.5626

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: 0.0002
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 64
• optimizer: Use OptimizerNames.ADAMW_BNB 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: 10
• training_steps: 892

Training results

Training Loss	Epoch	Step	Validation Loss
6.183	0.0006	1	6.4935
2.7821	0.0281	50	2.7672
2.6933	0.0561	100	2.7114
2.6583	0.0842	150	2.6771
2.7439	0.1122	200	2.6637
2.562	0.1403	250	2.6464
2.7245	0.1684	300	2.6365
2.6579	0.1964	350	2.6249
2.8156	0.2245	400	2.6096
2.5292	0.2526	450	2.6016
2.5435	0.2806	500	2.5919
2.6033	0.3087	550	2.5848
2.5197	0.3367	600	2.5782
2.6235	0.3648	650	2.5724
2.3961	0.3929	700	2.5675
2.5195	0.4209	750	2.5650
2.5028	0.4490	800	2.5633
2.4243	0.4770	850	2.5626

Framework versions

• PEFT 0.13.2
• Transformers 4.46.0
• Pytorch 2.5.0+cu124
• Datasets 3.0.1
• Tokenizers 0.20.1