See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: Qwen/Qwen2-0.5B-Instruct
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - a4812af705a99992_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/a4812af705a99992_train_data.json
  type:
    field_instruction: prompt
    field_output: GEITje-7B-ultra
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
device_map:
  ? ''
  : 0,1,2,3,4,5,6,7
early_stopping_patience: 2
eval_max_new_tokens: 128
eval_steps: 100
eval_table_size: null
flash_attention: true
gradient_accumulation_steps: 8
gradient_checkpointing: true
group_by_length: false
hub_model_id: Alphatao/4f6569e7-34d2-4cda-999c-da683488fcad
hub_repo: null
hub_strategy: null
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
- o_proj
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 5376
micro_batch_size: 4
mlflow_experiment_name: /tmp/a4812af705a99992_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 2
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: 0.04
wandb_entity: null
wandb_mode: online
wandb_name: 7e27224a-c921-4daf-ae25-99e536d03fc6
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 7e27224a-c921-4daf-ae25-99e536d03fc6
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

4f6569e7-34d2-4cda-999c-da683488fcad

This model is a fine-tuned version of Qwen/Qwen2-0.5B-Instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.8845

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: 8
• total_train_batch_size: 32
• 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: 2984

Training results

Training Loss	Epoch	Step	Validation Loss
2.3423	0.0007	1	2.4573
2.119	0.0670	100	2.2297
2.0666	0.1341	200	2.1565
2.1467	0.2011	300	2.1080
2.1191	0.2682	400	2.0717
2.0104	0.3352	500	2.0461
2.0987	0.4022	600	2.0235
2.1524	0.4693	700	2.0040
1.9367	0.5363	800	1.9897
1.8111	0.6034	900	1.9749
1.9941	0.6704	1000	1.9641
1.9709	0.7375	1100	1.9528
2.0112	0.8045	1200	1.9429
1.9086	0.8715	1300	1.9339
1.8286	0.9386	1400	1.9262
1.7759	1.0058	1500	1.9207
2.0174	1.0728	1600	1.9153
1.9403	1.1399	1700	1.9114
1.9319	1.2069	1800	1.9070
1.7344	1.2739	1900	1.9026
1.8536	1.3410	2000	1.8989
1.9783	1.4080	2100	1.8954
2.0019	1.4751	2200	1.8928
1.5308	1.5421	2300	1.8903
1.7832	1.6092	2400	1.8882
1.5709	1.6762	2500	1.8867
1.8052	1.7432	2600	1.8857
1.8442	1.8103	2700	1.8849
1.868	1.8773	2800	1.8846
1.8549	1.9444	2900	1.8845

Framework versions

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