nbeerbower's picture
Create README.md
ed0d372 verified
metadata
license: apache-2.0
library_name: transformers
base_model:
  - nbeerbower/mistral-nemo-kartoffel-12B
datasets:
  - jondurbin/gutenberg-dpo-v0.1
  - nbeerbower/gutenberg2-dpo
  - nbeerbower/gutenberg-moderne-dpo
  - nbeerbower/synthetic-fiction-dpo
  - nbeerbower/Arkhaios-DPO
  - nbeerbower/Purpura-DPO
  - nbeerbower/Schule-DPO

Schreiber-mistral-nemo-12B

nbeerbower/mistral-nemo-kartoffel-12B finetuned on:

Method

ORPO tuned with 1x RTX A6000 for 3 epochs.

QLoRA config

# QLoRA config
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch_dtype,
    bnb_4bit_use_double_quant=True,
)
# LoRA config
peft_config = LoraConfig(
    r=64,
    lora_alpha=128,
    lora_dropout=0.05,
    bias="none",
    task_type="CAUSAL_LM",
    target_modules=['up_proj', 'down_proj', 'gate_proj', 'k_proj', 'q_proj', 'v_proj', 'o_proj']
)

ORPO config

orpo_args = ORPOConfig(
    learning_rate=8e-6,
    lr_scheduler_type="cosine",
    warmup_ratio=0.05,
    max_length=4096,
    max_prompt_length=1024,
    max_completion_length=4096,
    beta=0.1,
    per_device_train_batch_size=1,
    per_device_eval_batch_size=1,
    gradient_accumulation_steps=64,
    optim="paged_adamw_8bit",
    num_train_epochs=3,
    max_grad_norm=0.5,
    bf16=True,
)