This model is a fine-tuned version of
jsfs11/WestOrcaNeuralMarco-DPO-v2-DARETIES-7B on the OpenHermes2.5-dpo-binarized-alpha dataset.
The following hyperparameters were used during training:
base_model: jsfs11/WestOrcaNeuralMarco-DPO-v2-DARETIES-7B
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
is_mistral_derived_model: true
load_in_8bit: false
load_in_4bit: true
strict: false
rl: dpo
chat_template: chatml
datasets:
- path: mlabonne/chatml-OpenHermes2.5-dpo-binarized-alpha
split: train
type: chatml.intel
dataset_prepared_path:
val_set_size: 0.01
output_dir: ./out
adapter: qlora
lora_model_dir:
sequence_len: 1800
sample_packing: false
pad_to_sequence_len: false
lora_r: 32
lora_alpha: 32
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:
lora_target_modules:
wandb_project: axolotl
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:
gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 1
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 5e-7
train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: true
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: 1
eval_table_size:
eval_table_max_new_tokens: 128
save_steps: 1080
max_steps: 1080
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
Training results
"train/loss": 0.4733,
"train/grad_norm": 15.831088066101074,
"train/learning_rate": 0,
"train/rewards/chosen": -0.6122800707817078,
"train/rewards/rejected": -1.650345802307129,
"train/rewards/accuracies": 0.875,
"train/rewards/margins": 1.0380656719207764,
"train/logps/rejected": -379.778564453125,
"train/logps/chosen": -250.2126007080078,
"train/logits/rejected": -2.0232465267181396,
"train/logits/chosen": -2.1629369258880615,
"train/epoch": 2.08594881699662,
"train/global_step": 1080,
"_timestamp": 1717044966.608197,
"_runtime": 12949.461512088776,
"_step": 1080,
"train_runtime": 12950.5619,
"train_samples_per_second": 1.334,
"train_steps_per_second": 0.083,
"total_flos": 0,
"train_loss": 0.560937881635295,
