ForgottenMaid-12B-bnb

Code for quantization (Generated by Grok with manual editing)

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
import torch
import sys

# Define model ID
model_id = sys.argv[1]

# Configure quantization
quantization_config = BitsAndBytesConfig(
    load_in_4bit=True,  # Use 4-bit quantization (or load_in_8bit=True for 8-bit)
    bnb_4bit_quant_type="nf4",  # Normal Float 4-bit (nf4) for better precision
    bnb_4bit_compute_dtype=torch.float16,  # Compute in float16 for efficiency
    bnb_4bit_use_double_quant=True  # Double quantization for further memory savings
)

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_id)

# Load quantized model
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    quantization_config=quantization_config,
    device_map="auto",  # Automatically map layers to GPU/CPU
    torch_dtype=torch.float16
)

# Save model and tokenizer
save_path = sys.argv[2]
model.save_pretrained(save_path)
tokenizer.save_pretrained(save_path)