Edit - Retraining model messed up the output. Maybe cz of my chat template. I will fine tune and update this. Stay Tuned :)

axolotl version: 0.3.0

base_model: NousResearch/Llama-2-7b-hf
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer
is_llama_derived_model: true
hub_model_id: MathLlama-7b

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: zorooo/Eval_Math_Derivatives
    type: alpaca
dataset_prepared_path:
val_set_size: 0.05
output_dir: ./qlora-out-2

adapter: qlora
lora_model_dir:

sequence_len: 2048
sample_packing: true
pad_to_sequence_len: true

lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules:
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project: axolotl_run_1_math_llama
wandb_entity:
wandb_watch:
wandb_name: math_llama_run2
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 5
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 0.0002

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

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

warmup_steps: 100
evals_per_epoch: 4
eval_table_size:
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

MathLlama-7b

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

• Loss: 0.1702

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss
0.1242	0.04	1	0.1574
0.1265	0.27	7	0.1573
0.1644	0.54	14	0.1574
0.1213	0.82	21	0.1566
0.1219	1.06	28	0.1560
0.111	1.33	35	0.1577
0.1289	1.6	42	0.1562
0.1241	1.87	49	0.1551
0.1254	2.12	56	0.1592
0.1376	2.39	63	0.1646
0.132	2.66	70	0.1611
0.1165	2.93	77	0.1568
0.1047	3.18	84	0.1698
0.0918	3.46	91	0.1717
0.1022	3.73	98	0.1677
0.1136	4.0	105	0.1661
0.0856	4.25	112	0.1733
0.0834	4.52	119	0.1702

Framework versions

• PEFT 0.7.2.dev0
• Transformers 4.37.0.dev0
• Pytorch 2.0.1+cu118
• Datasets 2.16.1
• Tokenizers 0.15.0