See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: TinyLlama/TinyLlama_v1.1
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 2bfee70655ebbccb_train_data.json
  ds_type: json
  field: content
  path: /workspace/input_data/2bfee70655ebbccb_train_data.json
  type: completion
debug: null
distributed_type: ddp
early_stopping_patience: null
eval_max_new_tokens: 128
eval_table_size: null
evals_per_epoch: 4
flash_attention: true
fp16: null
gradient_accumulation_steps: 8
gradient_checkpointing: true
group_by_length: false
hub_model_id: fats-fme/6d20e916-0301-420f-ac45-15677684fd9f
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0002
load_in_4bit: false
load_in_8bit: true
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
lr_scheduler: cosine
max_memory_MB: 40000
max_steps: 1800
micro_batch_size: 4
mlflow_experiment_name: /tmp/2bfee70655ebbccb_train_data.json
model_type: AutoModelForCausalLM
num_devices: 2
num_epochs: 3
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
saves_per_epoch: 4
sequence_len: 2048
special_tokens:
  pad_token: </s>
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: 6d20e916-0301-420f-ac45-15677684fd9f
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 6d20e916-0301-420f-ac45-15677684fd9f
warmup_steps: 100
world_size: 2
xformers_attention: null

6d20e916-0301-420f-ac45-15677684fd9f

This model is a fine-tuned version of TinyLlama/TinyLlama_v1.1 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.4870

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
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• total_eval_batch_size: 8
• 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: 100
• training_steps: 82

Training results

Training Loss	Epoch	Step	Validation Loss
1.4795	0.0369	1	1.4870
1.4298	0.2581	7	1.4870
1.4654	0.5161	14	1.4870
1.4517	0.7742	21	1.4870
2.7678	1.0323	28	1.4870
1.3889	1.2903	35	1.4870
1.4608	1.5484	42	1.4870
1.4305	1.8065	49	1.4870
1.5073	2.0645	56	1.4870
1.4382	2.3226	63	1.4870
1.4362	2.5806	70	1.4870
1.4102	2.8387	77	1.4870

Framework versions

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