See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: JackFram/llama-68m
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 3b022bbf876bb022_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/3b022bbf876bb022_train_data.json
  type:
    field_instruction: instruction
    field_output: output
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 2
eval_max_new_tokens: 128
eval_steps: 100
eval_table_size: null
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 8
gradient_checkpointing: true
group_by_length: false
hub_model_id: Alphatao/48b93c3f-b584-4eea-af10-bdcd3b475793
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_grad_norm: 1.0
max_steps: 2520
micro_batch_size: 4
mlflow_experiment_name: /tmp/3b022bbf876bb022_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: 2048
special_tokens:
  pad_token: </s>
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: 9a2de9cd-181f-4b68-b9fa-bbcc319584cc
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 9a2de9cd-181f-4b68-b9fa-bbcc319584cc
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

48b93c3f-b584-4eea-af10-bdcd3b475793

This model is a fine-tuned version of JackFram/llama-68m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.4406

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: 2520

Training results

Training Loss	Epoch	Step	Validation Loss
5.1347	0.0006	1	5.3534
3.8199	0.0648	100	3.8628
3.6453	0.1295	200	3.4683
3.0499	0.1943	300	3.2360
3.001	0.2591	400	3.0761
2.8141	0.3239	500	2.9496
2.6098	0.3886	600	2.8526
2.5753	0.4534	700	2.7764
2.8254	0.5182	800	2.7130
2.5672	0.5829	900	2.6640
2.4898	0.6477	1000	2.6243
2.9367	0.7125	1100	2.5887
2.5477	0.7773	1200	2.5608
2.5939	0.8420	1300	2.5375
2.4766	0.9068	1400	2.5170
2.3303	0.9716	1500	2.4973
2.4749	1.0368	1600	2.4847
2.4359	1.1016	1700	2.4736
2.4272	1.1664	1800	2.4631
2.4509	1.2312	1900	2.4559
2.6858	1.2959	2000	2.4515
2.1665	1.3607	2100	2.4459
2.4758	1.4255	2200	2.4433
2.4798	1.4902	2300	2.4416
2.2094	1.5550	2400	2.4407
2.193	1.6198	2500	2.4406

Framework versions

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