from transformers import AutoTokenizer, AutoModelForCausalLM
from datasets import load_dataset
from llmcompressor import oneshot
from llmcompressor.modifiers.quantization import QuantizationModifier
from llmcompressor.modifiers.smoothquant import SmoothQuantModifier
from llmcompressor.transformers.compression.helpers import calculate_offload_device_map

model_id  = "Qwen/Qwen2.5-Coder-14B-Instruct"
model_out = "Qwen2.5-Coder-14B-Instruct.w8a8"

num_samples = 128
max_seq_len = 4096

tokenizer = AutoTokenizer.from_pretrained(model_id)

def preprocess_fn(example):
  return {"text": tokenizer.apply_chat_template(example["messages"], add_generation_prompt=False, tokenize=False)}

ds = load_dataset("neuralmagic/LLM_compression_calibration", split="train")
ds = ds.shuffle().select(range(num_samples))
ds = ds.map(preprocess_fn)

recipe = [
  SmoothQuantModifier(smoothing_strength=0.8),
  QuantizationModifier(targets="Linear", scheme="W8A8", ignore=["lm_head"], dampening_frac=0.1)
]

device_map = calculate_offload_device_map(
    model_id, reserve_for_hessians=True, num_gpus=1, torch_dtype="bfloat16"
)

model = AutoModelForCausalLM.from_pretrained(
  model_id,
  device_map=device_map,
  torch_dtype="bfloat16",
)

oneshot(
  model=model,
  dataset=ds,
  recipe=recipe,
  max_seq_length=max_seq_len,
  num_calibration_samples=num_samples,
  output_dir=model_out,
)

nytopop/Qwen2.5-Coder-14B-Instruct.w8a8

nytopop/
Qwen2.5-Coder-14B-Instruct.w8a8