See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: unsloth/tinyllama-chat
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - daed85532ae01daa_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/daed85532ae01daa_train_data.json
  type:
    field_input: input
    field_instruction: instruction
    field_output: output
    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/8d2d892c-ff6e-4848-9b91-107eb3356cd9
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: 3436
micro_batch_size: 4
mlflow_experiment_name: /tmp/daed85532ae01daa_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
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: d7c344bd-e406-41ee-ad84-5b799adb7e49
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: d7c344bd-e406-41ee-ad84-5b799adb7e49
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

8d2d892c-ff6e-4848-9b91-107eb3356cd9

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

• Loss: 1.0690

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: 3436

Training results

Training Loss	Epoch	Step	Validation Loss
1.7226	0.0004	1	1.8629
1.5389	0.0413	100	1.5883
1.6151	0.0826	200	1.5001
1.4608	0.1239	300	1.4470
1.391	0.1652	400	1.4021
1.4795	0.2065	500	1.3675
1.3089	0.2478	600	1.3393
1.2863	0.2891	700	1.3107
1.2662	0.3304	800	1.2906
1.0741	0.3717	900	1.2695
1.3393	0.4130	1000	1.2517
1.2022	0.4543	1100	1.2325
1.2035	0.4956	1200	1.2130
1.1916	0.5369	1300	1.2001
1.194	0.5782	1400	1.1885
1.1436	0.6195	1500	1.1722
1.1657	0.6608	1600	1.1645
1.2633	0.7022	1700	1.1506
1.1023	0.7435	1800	1.1378
1.083	0.7848	1900	1.1263
1.1029	0.8261	2000	1.1195
1.2289	0.8674	2100	1.1089
1.0197	0.9087	2200	1.1006
1.0328	0.9500	2300	1.0929
1.0541	0.9913	2400	1.0857
0.9552	1.0326	2500	1.0857
0.8904	1.0739	2600	1.0828
0.9669	1.1152	2700	1.0780
1.0052	1.1565	2800	1.0762
0.9452	1.1978	2900	1.0731
0.876	1.2391	3000	1.0715
0.937	1.2804	3100	1.0701
1.0547	1.3217	3200	1.0692
0.8701	1.3630	3300	1.0691
1.1223	1.4043	3400	1.0690

Framework versions

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