awilliamson/phi-smol

See axolotl config

axolotl version: 0.3.0

base_model: phi-600M-cont/checkpoint-5000
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
trust_remote_code: true

load_in_8bit: false
load_in_4bit: false
strict: false

# max_steps: 8000
#pretraining_dataset: nampdn-ai/tiny-strange-textbooks 
datasets:
  - path: math-ai/StackMathQA
    name: stackmathqa100k
    type:
      system_prompt: ""
      field_system: system
      field_instruction: Q
      field_output: A
      format: "[INST] {instruction} [/INST]"
      no_input_format: "[INST] {instruction} [/INST]"
    train_on_split: train[:10%]
  - path: SciPhi/textbooks-are-all-you-need-lite
    type: completion
    field: completion
    train_on_split: train[:10%]


dataset_prepared_path:
val_set_size: 0.001
output_dir: ./phi-600M-mix

sequence_len: 2048
sample_packing: true  # currently unsupported
pad_to_sequence_len:

adapter:
lora_model_dir:
lora_r:
lora_alpha:
lora_dropout:
lora_target_linear:
lora_fan_in_fan_out:
lora_modules_to_save:

wandb_project: phine
wandb_entity: willfulbytes
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 1
num_epochs: 1
optimizer: paged_adamw_8bit
adam_beta2: 0.98
adam_epsilon: 0.0000001
max_grad_norm: 1.0
lr_scheduler: cosine
learning_rate: 1e-4
cosine_min_lr_ratio: 0.2

train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: true

gradient_checkpointing: true
early_stopping_patience: false
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 0
evals_per_epoch: 100
saves_per_epoch: 10
save_steps:
debug:
deepspeed:
weight_decay: 0.1
fsdp:
fsdp_config:
resize_token_embeddings_to_32x: true
special_tokens:
  pad_token: "<|endoftext|>"

phi-600M-mix

This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.6549

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: 0.0001
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 4
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-07
• lr_scheduler_type: cosine
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
3.366	0.0	1	3.3037
2.5809	0.01	84	2.5172
2.5684	0.02	168	2.3902
2.6054	0.03	252	2.3144
2.2944	0.04	336	2.2658
2.2836	0.05	420	2.2178
2.4438	0.06	504	2.1837
2.1093	0.07	588	2.1460
2.1831	0.08	672	2.1220
2.3081	0.09	756	2.0990
1.9909	0.1	840	2.0850
2.114	0.11	924	2.0550
1.8529	0.12	1008	2.0410
2.1594	0.13	1092	2.0215
2.0632	0.14	1176	2.0035
1.9221	0.15	1260	1.9906
2.0664	0.16	1344	1.9861
1.931	0.17	1428	1.9708
1.9948	0.18	1512	1.9533
1.9229	0.19	1596	1.9464
2.0231	0.2	1680	1.9332
2.2535	0.21	1764	1.9232
1.8994	0.22	1848	1.9140
1.9913	0.23	1932	1.8935
1.8613	0.24	2016	1.8916
1.9724	0.25	2100	1.8790
1.9965	0.26	2184	1.8653
2.0012	0.27	2268	1.8648
1.9752	0.28	2352	1.8572
1.9709	0.29	2436	1.8504
1.7314	0.3	2520	1.8432
1.7373	0.31	2604	1.8470
1.93	0.32	2688	1.8353
1.7185	0.33	2772	1.8210
1.8435	0.34	2856	1.8201
1.8117	0.35	2940	1.8118
2.1292	0.36	3024	1.8095
1.7536	0.37	3108	1.8023
1.7596	0.38	3192	1.7956
1.9481	0.39	3276	1.7890
1.7915	0.4	3360	1.7872
1.8639	0.41	3444	1.7782
1.6688	0.42	3528	1.7754
1.6312	0.43	3612	1.7669
1.8053	0.45	3696	1.7602
1.8867	0.46	3780	1.7544
1.9305	0.47	3864	1.7546
1.7926	0.48	3948	1.7496
1.8326	0.49	4032	1.7436
1.7334	0.5	4116	1.7437
1.6552	0.51	4200	1.7348
1.6622	0.52	4284	1.7330
1.9858	0.53	4368	1.7303
1.7784	0.54	4452	1.7271
1.8752	0.55	4536	1.7222
1.5931	0.56	4620	1.7186
1.6785	0.57	4704	1.7131
1.8382	0.58	4788	1.7101
1.5888	0.59	4872	1.7081
1.8055	0.6	4956	1.7062
1.6869	0.61	5040	1.7021
1.8096	0.62	5124	1.6999
1.9318	0.63	5208	1.6980
1.6153	0.64	5292	1.6963
1.6556	0.65	5376	1.6924
1.4087	0.66	5460	1.6908
1.7946	0.67	5544	1.6881
1.6097	0.68	5628	1.6867
1.6397	0.69	5712	1.6847
1.7799	0.7	5796	1.6828
1.6216	0.71	5880	1.6809
1.5052	0.72	5964	1.6790
1.6931	0.73	6048	1.6773
1.5936	0.74	6132	1.6762
1.803	0.75	6216	1.6737
1.5175	0.76	6300	1.6719
1.6305	0.77	6384	1.6711
1.715	0.78	6468	1.6698
1.8779	0.79	6552	1.6686
1.6844	0.8	6636	1.6669
1.3624	0.81	6720	1.6658
1.5534	0.82	6804	1.6650
1.8579	0.83	6888	1.6648
1.6093	0.84	6972	1.6632
1.5325	0.85	7056	1.6618
1.6753	0.86	7140	1.6619
1.3612	0.87	7224	1.6611
1.4817	0.88	7308	1.6606
1.7252	0.89	7392	1.6599
1.7463	0.9	7476	1.6586
1.8894	0.91	7560	1.6581
1.545	0.92	7644	1.6575
1.7251	0.93	7728	1.6572
1.7265	0.94	7812	1.6572
1.7813	0.95	7896	1.6564
1.7005	0.96	7980	1.6560
1.6444	0.97	8064	1.6555
1.5202	0.98	8148	1.6552
1.8648	0.99	8232	1.6549

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.0.1
• Datasets 2.16.1
• Tokenizers 0.15.0