gemma-2-9b-it-quantized.w4a16

Model Overview

• Model Architecture: Gemma-2
  • Input: Text
  • Output: Text
• Model Optimizations:
  • Weight quantization: INT4
• Intended Use Cases: Intended for commercial and research use in English. Similarly to gemma-2-9b-it, this models is intended for assistant-like chat.
• Out-of-scope: Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in languages other than English.
• Release Date: 8/16/2024
• Version: 1.0
• License(s): Gemma
• Model Developers: Neural Magic

Quantized version of gemma-2-9b-it. It achieves an average score of 73.62 on the OpenLLM benchmark (version 1), whereas the unquantized model achieves 73.23.

Model Optimizations

This model was obtained by quantizing the weights of gemma-2-9b-it to INT4 data type. This optimization reduces the number of bits per parameter from 16 to 4, reducing the disk size and GPU memory requirements by approximately 75%.

Only the weights of the linear operators within transformers blocks are quantized. Symmetric group-wise quantization is applied, in which a linear scaling per group maps the INT4 and floating point representations of the quantized weights. The GPTQ algorithm is applied for quantization, as implemented in the llm-compressor library. Quantization is performed with 1% damping factor, group-size as 128 and 512 sequences sampled from Open-Platypus.

Deployment

Use with vLLM

This model can be deployed efficiently using the vLLM backend, as shown in the example below.

from vllm import LLM, SamplingParams
from transformers import AutoTokenizer

model_id = "neuralmagic/gemma-2-9b-it-quantized.w4a16"

sampling_params = SamplingParams(temperature=0.6, top_p=0.9, max_tokens=256)

tokenizer = AutoTokenizer.from_pretrained(model_id)

messages = [
    {"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
    {"role": "user", "content": "Who are you? Please respond in pirate speak."},
]

prompts = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

llm = LLM(model=model_id, tensor_parallel_size=2)

outputs = llm.generate(prompts, sampling_params)

generated_text = outputs[0].outputs[0].text
print(generated_text)

vLLM aslo supports OpenAI-compatible serving. See the documentation for more details.

Use with transformers

The following example contemplates how the model can be deployed in Transformers using the generate() function.

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "neuralmagic/gemma-2-9b-it-quantized.w4a16"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)

messages = [
    {"role": "system", "content": "You are a pirate chatbot who always responds in pirate speak!"},
    {"role": "user", "content": "Who are you? Please respond in pirate speak"},
]

input_ids = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

terminators = [
    tokenizer.eos_token_id,
    tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

outputs = model.generate(
    input_ids,
    max_new_tokens=256,
    eos_token_id=terminators,
    do_sample=True,
    temperature=0.6,
    top_p=0.9,
)
response = outputs[0][input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))

Creation

This model was created by using the llm-compressor library as presented in the code snipet below.

from transformers import AutoTokenizer
from llmcompressor.transformers import SparseAutoModelForCausalLM, oneshot
from llmcompressor.modifiers.quantization import GPTQModifier
from datasets import load_dataset
import random

model_id = "google/gemma-2-9b-it"

num_samples = 512
max_seq_len = 4096

tokenizer = AutoTokenizer.from_pretrained(model_id)

preprocess_fn = lambda example: {"text": "Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n{text}".format_map(example)}

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

examples = [
    tokenizer(
        example["text"], padding=False, max_length=max_seq_len, truncation=True,
    ) for example in ds
]

recipe = GPTQModifier(
  targets="Linear",
  scheme="W4A16",
  ignore=["lm_head"],
  dampening_frac=0.01,
)

model = SparseAutoModelForCausalLM.from_pretrained(
  model_id,
  device_map="auto",
  trust_remote_code=True,
)

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

model.save_pretrained("gemma-2-9b-it-quantized.w4a16")

Evaluation

The model was evaluated on the OpenLLM leaderboard tasks (version 1) with the lm-evaluation-harness (commit 383bbd54bc621086e05aa1b030d8d4d5635b25e6) and the vLLM engine, using the following command:

lm_eval \
  --model vllm \
  --model_args pretrained="neuralmagic/gemma-2-9b-it-quantized.w4a16",dtype=auto,tensor_parallel_size=2,gpu_memory_utilization=0.4,add_bos_token=True,max_model_len=4096,trust_remote_code=True \
  --tasks openllm \
  --batch_size auto

Accuracy

Open LLM Leaderboard evaluation scores

Benchmark	gemma-2-9b-it	gemma-2-9b-it-quantized.w4a16(this model)	Recovery
MMLU (5-shot)	72.28	71.36	98.72%
ARC Challenge (25-shot)	71.5	70.98	99.27%
GSM-8K (5-shot, strict-match)	76.26	79.83	104.68%
Hellaswag (10-shot)	81.91	81.29	99.24%
Winogrande (5-shot)	77.11	78.29	101.53%
TruthfulQA (0-shot)	60.32	59.97	99.42%
Average	73.23	73.62	100.53%