#!/usr/bin/env python3
"""Check model size and quantization status."""

import torch
from transformers import AutoModelForCausalLM

model_path = "/data/models/Ornith-1.0-35B-bnb-4bit"

print(f"Loading model from: {model_path}")
model = AutoModelForCausalLM.from_pretrained(
    model_path,
    device_map="cpu",
    torch_dtype=torch.float16,
    trust_remote_code=True,
)

# Count parameters
total_params = sum(p.numel() for p in model.parameters())
print(f"Total parameters: {total_params / 1e9:.2f}B")

# Check for quantized parameters
bnb_params = 0
bf16_params = 0
for name, p in model.named_parameters():
    if hasattr(p, 'quant_state') and p.quant_state is not None:
        bnb_params += p.numel()
    else:
        bf16_params += p.numel()

print(f"BnB 4-bit parameters: {bnb_params / 1e9:.2f}B")
print(f"BF16 parameters: {bf16_params / 1e9:.2f}B")
print(f"Estimated size: {(bnb_params * 0.5 + bf16_params * 2) / 1e9:.2f} GB")

del model
import gc
gc.collect()