See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: microsoft/Phi-3-mini-4k-instruct
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 09e4f1971d3a0fdc_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/09e4f1971d3a0fdc_train_data.json
  type:
    field_instruction: premise
    field_output: hypothesis
    format: '{instruction}'
    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/a4d6c702-99f9-4a83-bd22-ffba734a7c58
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: 2958
micro_batch_size: 4
mlflow_experiment_name: /tmp/09e4f1971d3a0fdc_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: 0.029679755438815184
wandb_entity: null
wandb_mode: online
wandb_name: 5b3ba6f9-b3e0-4796-bb0b-1e66dec49f77
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 5b3ba6f9-b3e0-4796-bb0b-1e66dec49f77
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

a4d6c702-99f9-4a83-bd22-ffba734a7c58

This model is a fine-tuned version of microsoft/Phi-3-mini-4k-instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.4409

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: 2958

Training results

Training Loss	Epoch	Step	Validation Loss
21.0035	0.0002	1	3.0446
13.5929	0.0196	100	1.6034
15.2066	0.0392	200	1.5797
11.9544	0.0587	300	1.5673
15.0186	0.0783	400	1.5544
15.2379	0.0979	500	1.5472
11.9004	0.1175	600	1.5425
12.0731	0.1370	700	1.5289
13.5655	0.1566	800	1.5247
11.0814	0.1762	900	1.5171
12.0226	0.1958	1000	1.5113
13.4871	0.2153	1100	1.5027
12.4213	0.2349	1200	1.4974
11.609	0.2545	1300	1.4948
10.51	0.2741	1400	1.4853
10.1529	0.2936	1500	1.4808
12.5564	0.3132	1600	1.4773
11.5684	0.3328	1700	1.4699
11.8717	0.3524	1800	1.4686
15.0797	0.3719	1900	1.4606
11.5057	0.3915	2000	1.4571
12.0039	0.4111	2100	1.4533
10.9814	0.4307	2200	1.4501
9.6352	0.4502	2300	1.4471
11.4737	0.4698	2400	1.4448
13.1873	0.4894	2500	1.4429
11.5528	0.5090	2600	1.4418
10.4684	0.5285	2700	1.4411
10.4001	0.5481	2800	1.4409
10.6807	0.5677	2900	1.4409

Framework versions

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