See axolotl config

axolotl version: 0.4.0

base_model: meta-llama/Llama-2-7b-hf
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer
is_llama_derived_model: true

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: utrgvseniorproject/MeditronGuidelines
    type: completion
dataset_prepared_path: /home/josegomez15/med-llm/Llama_Preprocess_MeditronGuideLines_txt
val_set_size: 0.05
output_dir: ./Llama2-7B-MeditronGuideLines-txt-epochs-1-lr-000002

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

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

wandb_project: Llama2-7B-MeditronGuideLines
wandb_entity: utrgvmedai
wandb_watch:
wandb_name: Llama2-7B-MeditronGuideLines-txt-epochs-1-lr-000002
wandb_log_model:

gradient_accumulation_steps: 1
micro_batch_size: 1
num_epochs: 1
#saves_per_epoch: 10
save_steps: 800
#save_total_limit: 4
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.000002

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

gradient_checkpointing: false
early_stopping_patience:
resume_from_checkpoint: true

local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
flash_attn_cross_entropy: false
flash_attn_rms_norm: true
flash_attn_fuse_qkv: false
flash_attn_fuse_mlp: true

warmup_steps: 2000
evals_per_epoch: 4
eval_table_size:
eval_sample_packing: False
debug:
deepspeed: /home/josegomez15/axolotl/deepspeed_configs/zero2.json
weight_decay: 0.1
fsdp:
fsdp_config:
special_tokens:

Llama2-7B-MeditronGuideLines-txt-epochs-1-lr-000002

This model is a fine-tuned version of meta-llama/Llama-2-7b-hf on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.3911

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: 2e-06
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 8
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 2000
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
1.3307	0.0	1	1.5317
1.4702	0.25	1141	1.4162
1.3621	0.5	2282	1.4039
1.4502	0.75	3423	1.3953
1.4184	1.0	4564	1.3911

Framework versions

• Transformers 4.38.0.dev0
• Pytorch 2.0.1+cu117
• Datasets 2.18.0
• Tokenizers 0.15.0