CleverShovel/Mistral-7B-v0.1-paper-reviews-lora

See axolotl config

axolotl version: 0.4.0

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

load_in_8bit: false
load_in_4bit: true
strict: false

bnb_config_kwargs:
  llm_int8_has_fp16_weight: true
  bnb_4bit_quant_type: nf4
  bnb_4bit_use_double_quant: false

datasets:
  - path: CleverShovel/paper_reviews
    type: alpaca
dataset_prepared_path: CleverShovel/paper_reviews
val_set_size: 0.05
output_dir: ./llm_train/test_out

#using lora for lower cost
adapter: qlora
lora_r: 8
lora_alpha: 32
lora_dropout: 0.05
lora_target_modules:
  - q_proj
  - v_proj

sequence_len: 2048
sample_packing: false
pad_to_sequence_len: true

wandb_project: paper_review
wandb_entity:
wandb_watch:
wandb_name: base
wandb_log_model: checkpoint

gradient_accumulation_steps: 2
micro_batch_size: 5
max_steps: 300
num_epochs: 1
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false

float16: true
bf16: false
fp16: false
tf32: false

save_safetensors: true
save_steps: 100

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 100
xformers_attention:
flash_attention: true

warmup_ration: 0.05
evals_steps: 100
eval_table_size:
eval_table_max_new_tokens: 128
debug:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

llm_train/test_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: 2.0276

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: 5
• eval_batch_size: 5
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 10
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 9
• training_steps: 300

Training results

Training Loss	Epoch	Step	Validation Loss
2.0121	0.13	300	2.0276

Framework versions

• PEFT 0.8.2
• Transformers 4.38.0.dev0
• Pytorch 2.1.2+cu118
• Datasets 2.16.1
• Tokenizers 0.15.0