Dataset procedure

• GPT4-generated dataset
• size: 80
• per_device_train_batch_size=4,
• gradient_accumulation_steps=4,
• warmup_steps=100,
• max_steps=200,
• learning_rate=2e-4,
• fp16=True,
• logging_steps=1,
• output_dir='outputs',

Training procedure

The following bitsandbytes quantization config was used during training:

• quant_method: bitsandbytes
• load_in_8bit: True
• load_in_4bit: False
• llm_int8_threshold: 6.0
• llm_int8_skip_modules: None
• llm_int8_enable_fp32_cpu_offload: False
• llm_int8_has_fp16_weight: False
• bnb_4bit_quant_type: fp4
• bnb_4bit_use_double_quant: False
• bnb_4bit_compute_dtype: float32

LoraConfig procedure

r=16, #attention heads
lora_alpha=32, #alpha scaling
# target_modules=["q_proj", "v_proj"], #if you know the
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM" # set this for CLM or Seq2Seq

Framework versions

• PEFT 0.6.0.dev0