Quantized version of gemma-2-9b-it. It achieves an average score of 73.49 on the OpenLLM benchmark (version 1), whereas the unquantized model achieves 73.23.
This model was obtained by quantizing the weights and activations of gemma-2-9b-it to FP8 data type, ready for inference with vLLM >= 0.5.1. This optimization reduces the number of bits per parameter from 16 to 8, reducing the disk size and GPU memory requirements by approximately 50%.
Only the weights and activations of the linear operators within transformers blocks are quantized. Symmetric per-tensor quantization is applied, in which a single linear scaling maps the FP8 representations of the quantized weights and activations. AutoFP8 is used for quantization with a single instance of every token in random order.
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 = "RedHatAI/gemma-2-9b-it-FP8"
sampling_params = SamplingParams(temperature=0.6, top_p=0.9, max_tokens=256)
tokenizer = AutoTokenizer.from_pretrained(model_id)
messages = [
{"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)
outputs = llm.generate(prompts, sampling_params)
generated_text = outputs[0].outputs[0].text
print(generated_text)
vLLM also supports OpenAI-compatible serving. See the documentation for more details.
podman run --rm -it --device nvidia.com/gpu=all -p 8000:8000 \
--ipc=host \
--env "HUGGING_FACE_HUB_TOKEN=$HF_TOKEN" \
--env "HF_HUB_OFFLINE=0" -v ~/.cache/vllm:/home/vllm/.cache \
--name=vllm \
registry.access.redhat.com/rhaiis/rh-vllm-cuda \
vllm serve \
--tensor-parallel-size 8 \
--max-model-len 32768 \
--enforce-eager --model RedHatAI/gemma-2-9b-it-FP8
​​See Red Hat AI Inference Server documentation for more details.
# Download model from Red Hat Registry via docker
# Note: This downloads the model to ~/.cache/instructlab/models unless --model-dir is specified.
ilab model download --repository docker://registry.redhat.io/rhelai1/gemma-2-9b-it-FP8:1.5
# Serve model via ilab
ilab model serve --model-path ~/.cache/instructlab/models/gemma-2-9b-it-FP8
# Chat with model
ilab model chat --model ~/.cache/instructlab/models/gemma-2-9b-it-FP8
See Red Hat Enterprise Linux AI documentation for more details.
# Setting up vllm server with ServingRuntime
# Save as: vllm-servingruntime.yaml
apiVersion: serving.kserve.io/v1alpha1
kind: ServingRuntime
metadata:
name: vllm-cuda-runtime # OPTIONAL CHANGE: set a unique name
annotations:
openshift.io/display-name: vLLM NVIDIA GPU ServingRuntime for KServe
opendatahub.io/recommended-accelerators: '["nvidia.com/gpu"]'
labels:
opendatahub.io/dashboard: 'true'
spec:
annotations:
prometheus.io/port: '8080'
prometheus.io/path: '/metrics'
multiModel: false
supportedModelFormats:
- autoSelect: true
name: vLLM
containers:
- name: kserve-container
image: quay.io/modh/vllm:rhoai-2.20-cuda # CHANGE if needed. If AMD: quay.io/modh/vllm:rhoai-2.20-rocm
command:
- python
- -m
- vllm.entrypoints.openai.api_server
args:
- "--port=8080"
- "--model=/mnt/models"
- "--served-model-name={{.Name}}"
env:
- name: HF_HOME
value: /tmp/hf_home
ports:
- containerPort: 8080
protocol: TCP
# Attach model to vllm server. This is an NVIDIA template
# Save as: inferenceservice.yaml
apiVersion: serving.kserve.io/v1beta1
kind: InferenceService
metadata:
annotations:
openshift.io/display-name: gemma-2-9b-it-FP8 # OPTIONAL CHANGE
serving.kserve.io/deploymentMode: RawDeployment
name: gemma-2-9b-it-FP8 # specify model name. This value will be used to invoke the model in the payload
labels:
opendatahub.io/dashboard: 'true'
spec:
predictor:
maxReplicas: 1
minReplicas: 1
model:
modelFormat:
name: vLLM
name: ''
resources:
limits:
cpu: '2' # this is model specific
memory: 8Gi # this is model specific
nvidia.com/gpu: '1' # this is accelerator specific
requests: # same comment for this block
cpu: '1'
memory: 4Gi
nvidia.com/gpu: '1'
runtime: vllm-cuda-runtime # must match the ServingRuntime name above
storageUri: oci://registry.redhat.io/rhelai1/modelcar-gemma-2-9b-it-FP8:1.5
tolerations:
- effect: NoSchedule
key: nvidia.com/gpu
operator: Exists
# make sure first to be in the project where you want to deploy the model
# oc project <project-name>
# apply both resources to run model
# Apply the ServingRuntime
oc apply -f vllm-servingruntime.yaml
# Apply the InferenceService
oc apply -f qwen-inferenceservice.yaml
# Replace <inference-service-name> and <cluster-ingress-domain> below:
# - Run `oc get inferenceservice` to find your URL if unsure.
# Call the server using curl:
curl https://<inference-service-name>-predictor-default.<domain>/v1/chat/completions
-H "Content-Type: application/json" \
-d '{
"model": "gemma-2-9b-it-FP8",
"stream": true,
"stream_options": {
"include_usage": true
},
"max_tokens": 1,
"messages": [
{
"role": "user",
"content": "How can a bee fly when its wings are so small?"
}
]
}'
See Red Hat Openshift AI documentation for more details.
This model was created by applying AutoFP8 with calibration samples from ultrachat, as presented in the code snipet below. Although AutoFP8 was used for this particular model, Neural Magic is transitioning to using llm-compressor which supports several quantization schemes and models not supported by AutoFP8.
from datasets import load_dataset
from transformers import AutoTokenizer
import numpy as np
import torch
from auto_fp8 import AutoFP8ForCausalLM, BaseQuantizeConfig
MODEL_DIR = "google/gemma-2-9b-it"
final_model_dir = MODEL_DIR.split("/")[-1]
CONTEXT_LENGTH = 4096
NUM_SAMPLES = 512
NUM_REPEATS = 1
pretrained_model_dir = MODEL_DIR
tokenizer = AutoTokenizer.from_pretrained(pretrained_model_dir, use_fast=True, model_max_length=CONTEXT_LENGTH)
tokenizer.pad_token = tokenizer.eos_token
tokenizer_num_tokens = len(list(tokenizer.get_vocab().values()))
total_token_samples = NUM_REPEATS * tokenizer_num_tokens
num_random_samp = -(-total_token_samples // CONTEXT_LENGTH)
input_ids = np.tile(np.arange(tokenizer_num_tokens), NUM_REPEATS + 1)[:num_random_samp * CONTEXT_LENGTH]
np.random.shuffle(input_ids)
input_ids = input_ids.reshape(num_random_samp, CONTEXT_LENGTH)
input_ids = torch.tensor(input_ids, dtype=torch.int64).to("cuda")
quantize_config = BaseQuantizeConfig(
quant_method="fp8",
activation_scheme="static",
)
examples = input_ids
model = AutoFP8ForCausalLM.from_pretrained(pretrained_model_dir, quantize_config=quantize_config)
model.quantize(examples)
quantized_model_dir = f"{final_model_dir}-FP8"
model.save_quantized(quantized_model_dir)
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-FP8",dtype=auto,gpu_memory_utilization=0.4,add_bos_token=True,max_model_len=4096 \
--tasks openllm \
--batch_size auto
| Benchmark | gemma-2-9b-it | gemma-2-9b-it-FP8(this model) | Recovery |
| MMLU (5-shot) | 72.28 | 71.99 | 99.59% |
| ARC Challenge (25-shot) | 71.50 | 71.50 | 100.0% |
| GSM-8K (5-shot, strict-match) | 76.26 | 76.87 | 100.7% |
| Hellaswag (10-shot) | 81.91 | 81.70 | 99.74% |
| Winogrande (5-shot) | 77.11 | 78.37 | 101.6% |
| TruthfulQA (0-shot) | 60.32 | 60.52 | 100.3% |
| Average | 73.23 | 73.49 | 100.36% |