See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: Qwen/Qwen2-0.5B
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 67d196c6d8c46ccb_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/67d196c6d8c46ccb_train_data.json
  type:
    field_input: rejected
    field_instruction: prompt
    field_output: chosen
    format: '{instruction} {input}'
    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/85fa2be2-5d82-4783-acf4-71bb2a8d44c4
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: 128
lora_dropout: 0.1
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 64
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: 3414
micro_batch_size: 4
mlflow_experiment_name: /tmp/67d196c6d8c46ccb_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: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: 21997cd8-7875-4d2d-8537-d9529b0c7b3a
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 21997cd8-7875-4d2d-8537-d9529b0c7b3a
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

85fa2be2-5d82-4783-acf4-71bb2a8d44c4

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

• Loss: 1.2863

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: 3414

Training results

Training Loss	Epoch	Step	Validation Loss
1.8747	0.0006	1	1.8434
1.5259	0.0551	100	1.6113
1.6668	0.1101	200	1.5517
1.5874	0.1652	300	1.5147
1.4066	0.2203	400	1.4878
1.6357	0.2753	500	1.4642
1.6035	0.3304	600	1.4488
1.5591	0.3855	700	1.4329
1.3102	0.4406	800	1.4199
1.4886	0.4956	900	1.4045
1.4737	0.5507	1000	1.3958
1.2135	0.6058	1100	1.3838
1.4271	0.6608	1200	1.3731
1.6556	0.7159	1300	1.3631
1.3462	0.7710	1400	1.3547
1.2202	0.8260	1500	1.3442
1.1946	0.8811	1600	1.3388
1.2619	0.9362	1700	1.3305
1.2952	0.9913	1800	1.3235
1.2811	1.0466	1900	1.3222
1.268	1.1017	2000	1.3196
1.3511	1.1567	2100	1.3137
1.1941	1.2118	2200	1.3092
1.1547	1.2669	2300	1.3061
1.164	1.3220	2400	1.3022
1.112	1.3770	2500	1.2983
1.2021	1.4321	2600	1.2959
1.2954	1.4872	2700	1.2931
1.062	1.5422	2800	1.2908
0.9773	1.5973	2900	1.2889
1.1116	1.6524	3000	1.2874
1.252	1.7074	3100	1.2870
1.3042	1.7625	3200	1.2865
1.2098	1.8176	3300	1.2864
1.164	1.8727	3400	1.2863

Framework versions

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