mikudev/MN-12B-Lyra-v3-FP8

Original Model: https://huggingface.co./Sao10K/MN-12B-Lyra-v3

Quantized with FP8 using https://github.com/neuralmagic/AutoFP8

Script:

from datasets import load_dataset
from transformers import AutoTokenizer

from auto_fp8 import AutoFP8ForCausalLM, BaseQuantizeConfig

pretrained_model_dir = "Sao10K/MN-12B-Lyra-v3"
quantized_model_dir = "MN-12B-Lyra-v3-FP8"

tokenizer = AutoTokenizer.from_pretrained(pretrained_model_dir, use_fast=True, model_max_length=4096)
tokenizer.pad_token = tokenizer.eos_token

ds = load_dataset("mgoin/ultrachat_2k", split="train_sft").select(range(512))
examples = [tokenizer.apply_chat_template(batch["messages"], tokenize=False) for batch in ds]
examples = tokenizer(examples, padding=True, truncation=True, return_tensors="pt").to("cuda")

quantize_config = BaseQuantizeConfig(
    quant_method="fp8",
    activation_scheme="static",
    ignore_patterns=["re:.*lm_head"],
)

model = AutoFP8ForCausalLM.from_pretrained(
    pretrained_model_dir, quantize_config=quantize_config
)

model.quantize(examples)
model.save_quantized(quantized_model_dir)