Origin Model

Model Load

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

peft_model_id = 'dev7halo/falcon-7b-sharded-bf16-KoAlpaca'
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.float16,
)
model = AutoModelForCausalLM.from_pretrained(
    config.base_model_name_or_path,
    quantization_config=bnb_config,
    trust_remote_code=True
)
model = PeftModel.from_pretrained(model, peft_model_id)
tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)

model = model.to('cuda')
model.eval()

Generate

inputs = tokenizer("광해군은 폭군이었나요 ?", return_tensors="pt")

with torch.no_grad():
    outputs = model.generate(input_ids=inputs["input_ids"].to("cuda"), max_new_tokens=512)
    print(tokenizer.batch_decode(outputs.detach().cpu().numpy(), skip_special_tokens=True)[0])

Training procedure

The following bitsandbytes quantization config was used during training:

• load_in_8bit: False
• load_in_4bit: True
• 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: nf4
• bnb_4bit_use_double_quant: False
• bnb_4bit_compute_dtype: float16

Framework versions

• PEFT 0.4.0.dev0