See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: bigscience/bloom-560m
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - d15508f8089bb58e_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/d15508f8089bb58e_train_data.json
  type:
    field_input: alt_text
    field_instruction: question
    field_output: response
    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: false
gradient_accumulation_steps: 8
gradient_checkpointing: true
group_by_length: false
hub_model_id: Alphatao/54e6ea23-b5e1-4619-b29d-902656097671
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:
- query_key_value
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 7680
micro_batch_size: 4
mlflow_experiment_name: /tmp/d15508f8089bb58e_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: e655a30b-fca7-4301-b464-31f5d361a493
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: e655a30b-fca7-4301-b464-31f5d361a493
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

54e6ea23-b5e1-4619-b29d-902656097671

This model is a fine-tuned version of bigscience/bloom-560m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.3220

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: 4676

Training results

Training Loss	Epoch	Step	Validation Loss
22.4204	0.0004	1	2.7565
13.7242	0.0428	100	1.6310
12.6846	0.0855	200	1.5751
11.8994	0.1283	300	1.5361
11.7018	0.1711	400	1.5088
13.5725	0.2139	500	1.4896
12.0607	0.2566	600	1.4739
10.9681	0.2994	700	1.4602
11.1942	0.3422	800	1.4502
10.2645	0.3849	900	1.4407
12.0345	0.4277	1000	1.4308
11.543	0.4705	1100	1.4215
10.9802	0.5133	1200	1.4173
10.6541	0.5560	1300	1.4101
10.0251	0.5988	1400	1.4006
10.9978	0.6416	1500	1.3946
11.9791	0.6843	1600	1.3887
10.4467	0.7271	1700	1.3834
10.8237	0.7699	1800	1.3796
10.5302	0.8127	1900	1.3746
10.5822	0.8554	2000	1.3683
10.7599	0.8982	2100	1.3670
9.556	0.9410	2200	1.3615
10.5026	0.9837	2300	1.3549
9.6549	1.0265	2400	1.3543
10.5016	1.0693	2500	1.3523
10.5743	1.1121	2600	1.3481
9.8328	1.1548	2700	1.3470
11.0661	1.1976	2800	1.3439
11.0951	1.2404	2900	1.3412
10.086	1.2831	3000	1.3390
9.8004	1.3259	3100	1.3364
9.525	1.3687	3200	1.3342
10.3603	1.4115	3300	1.3327
10.1241	1.4542	3400	1.3308
9.146	1.4970	3500	1.3288
10.2162	1.5398	3600	1.3271
10.3733	1.5825	3700	1.3260
10.2542	1.6253	3800	1.3246
10.2883	1.6681	3900	1.3239
10.0702	1.7109	4000	1.3234
9.5546	1.7536	4100	1.3233
9.1898	1.7964	4200	1.3218
9.8766	1.8392	4300	1.3222
10.6581	1.8820	4400	1.3218
10.9446	1.9247	4500	1.3220
11.5194	1.9675	4600	1.3220

Framework versions

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