# Llama-2-7b
# base_model: daryl149/llama-2-7b-chat-hf
# model_type: LlamaForCausalLM
# tokenizer_type: LlamaTokenizer
# is_llama_derived_model: true

#Mistral-7b
base_model: mistralai/Mistral-7B-v0.1
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
is_mistral_derived_model: true

# git clone https://github.com/OpenAccess-AI-Collective/axolotl
# cd axolotl

# pip3 install packaging
# pip3 install -e '.[flash-attn,deepspeed]'

# accelerate launch -m axolotl.cli.train ./llama_7b_config.yaml

# accelerate launch -m axolotl.cli.inference ./llama_7b_config.yaml \
#     --lora_model_dir="dohonba/mistral_7b_fingpt"

load_in_8bit: true
load_in_4bit: false
strict: false

datasets:
  - path: dohonba/combi
    type: context_qa.load_v2
  # - path: dohonba/tfns
  #   type: context_qa.load_v2
  # - path: dohonba/auditor_sentiment
  #   type: context_qa.load_v2
  # - path: dohonba/tfns
  #   type: context_qa.load_v2
    
dataset_prepared_path:
val_set_size: 0.05
output_dir: ./lora-out

sequence_len: 512
sample_packing: true
pad_to_sequence_len: true

adapter: lora
lora_model_dir:
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 1
micro_batch_size: 14
# max_steps: 1000
num_epochs: 2
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

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

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
s2_attention:

warmup_steps: 50
evals_per_epoch: 0
eval_table_size:
eval_table_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

lora-out

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.0917

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: 14
eval_batch_size: 14
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 50
num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss
0.08	1.02	566	0.0986
0.0919	1.98	1110	0.0917

Framework versions

PEFT 0.7.1
Transformers 4.37.0
Pytorch 2.0.1
Datasets 2.16.1
Tokenizers 0.15.0