YAML Metadata Warning: empty or missing yaml metadata in repo card (https://huggingface.co/docs/hub/model-cards#model-card-metadata)
vllm (pretrained=/home/mgoin/code/llm-compressor/examples/quantizing_moe/DeepSeek-Coder-V2-Lite-Instruct-FP8,tensor_parallel_size=2,trust_remote_code=True), gen_kwargs: (None), limit: None, num_fewshot: 5, batch_size: auto
|Tasks|Version|     Filter     |n-shot|  Metric   |   |Value |   |Stderr|
|-----|------:|----------------|-----:|-----------|---|-----:|---|-----:|
|gsm8k|      3|flexible-extract|     5|exact_match|↑  |0.7710|±  |0.0116|
|     |       |strict-match    |     5|exact_match|↑  |0.7582|±  |0.0118|

Creation

import torch
from datasets import load_dataset
from transformers import AutoTokenizer

from llmcompressor.modifiers.quantization import QuantizationModifier
from llmcompressor.transformers import SparseAutoModelForCausalLM, oneshot

# select a Mixture of Experts model for quantization
MODEL_ID = "deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct"

model = SparseAutoModelForCausalLM.from_pretrained(
    MODEL_ID, device_map="auto", torch_dtype="auto", trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)

# Select calibration dataset.
# its recommended to use more calibration samples for MoE models so each expert is hit
DATASET_ID = "HuggingFaceH4/ultrachat_200k"
DATASET_SPLIT = "train_sft"
NUM_CALIBRATION_SAMPLES = 2048
MAX_SEQUENCE_LENGTH = 2048


# Load dataset and preprocess.
ds = load_dataset(DATASET_ID, split=DATASET_SPLIT)
ds = ds.shuffle(seed=42).select(range(NUM_CALIBRATION_SAMPLES))


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


ds = ds.map(preprocess)


# Tokenize inputs.
def tokenize(sample):
    return tokenizer(
        sample["text"],
        padding=False,
        max_length=MAX_SEQUENCE_LENGTH,
        truncation=True,
        add_special_tokens=False,
    )


ds = ds.map(tokenize, remove_columns=ds.column_names)

# define a llmcompressor recipe for FP8 W8A8 quantization
# since the MoE gate layers are sensitive to quantization, we add them to the ignore
# list so they remain at full precision
recipe = [
    QuantizationModifier(
        targets="Linear",
        scheme="FP8",
        ignore=["lm_head", "re:.*mlp.gate$"],
    ),
]

SAVE_DIR = MODEL_ID.split("/")[1] + "-FP8"

oneshot(
    model=model,
    dataset=ds,
    recipe=recipe,
    max_seq_length=MAX_SEQUENCE_LENGTH,
    num_calibration_samples=NUM_CALIBRATION_SAMPLES,
    save_compressed=True,
    output_dir=SAVE_DIR,
)


print("========== SAMPLE GENERATION ==============")
SAMPLE_INPUT = ["I love quantization because"]
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
inputs = tokenizer(SAMPLE_INPUT, return_tensors="pt", padding=True).to(model.device)
output = model.generate(**inputs, max_length=50)
text_output = tokenizer.batch_decode(output)
print(text_output)
Downloads last month
4,067
Safetensors
Model size
15.7B params
Tensor type
BF16
·
F8_E4M3
·
Inference API
Unable to determine this model's library. Check the docs .