You can deploy this model and start earning money today!

RedHatAI/
Llama-4-Scout-17B-16E-Instruct-quantized.w4a16

Name: Llama-4-Scout-17B-16E-Instruct-quantized.w4a16
Availability: InStock

safetensors

llama4

facebook

Llama-4-Scout-17B-16E-Instruct-quantized.w4a16
Model Icon

Model Overview

Model Architecture: Llama4ForConditionalGeneration
- Input: Text / Image
- Output: Text
Model Optimizations:
- Activation quantization: None
- Weight quantization: INT4
Release Date: 04/25/2025
Version: 1.0
Validated on: RHOAI 2.20, RHAIIS 3.0, RHELAI 1.5
Model Developers: Red Hat (Neural Magic)

Model Optimizations

This model was obtained by quantizing weights of Llama-4-Scout-17B-16E-Instruct to INT4 data type. This optimization reduces the number of bits used to represent weights from 16 to 4, reducing GPU memory requirements by approximately 75%. Weight quantization also reduces disk size requirements by approximately 75%. The llm-compressor library is used for quantization.

Deployment

This model can be deployed efficiently on vLLM, Red Hat Enterprise Linux AI, and Openshift AI, as shown in the example below.

Deploy on vLLM

python

from vllm import LLM, SamplingParams
from transformers import AutoTokenizer

model_id = "RedHatAI/Llama-4-Scout-17B-16E-Instruct-quantized.w4a16"
number_gpus = 4

sampling_params = SamplingParams(temperature=0.7, top_p=0.8, max_tokens=256)

tokenizer = AutoTokenizer.from_pretrained(model_id)

prompt = "Give me a short introduction to large language model."

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

outputs = llm.generate(prompt, sampling_params)

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

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

Deploy on Red Hat AI Inference Server

bash

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/Llama-4-Scout-17B-16E-Instruct-quantized.w4a16

See Red Hat AI Inference Server documentation for more details.

Deploy on Red Hat Enterprise Linux AI

bash

# 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/llama-4-scout-17b-16e-instruct-quantized-w4a16:1.5

bash

# Serve model via ilab
ilab model serve --model-path ~/.cache/instructlab/models/llama-4-scout-17b-16e-instruct-quantized-w4a16
  
# Chat with model
ilab model chat --model ~/.cache/instructlab/models/llama-4-scout-17b-16e-instruct-quantized-w4a16

See Red Hat Enterprise Linux AI documentation for more details.

Deploy on Red Hat Openshift AI

python

# 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

python

# 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: Llama-4-Scout-17B-16E-Instruct-quantized.w4a16 # OPTIONAL CHANGE
    serving.kserve.io/deploymentMode: RawDeployment
  name: Llama-4-Scout-17B-16E-Instruct-quantized.w4a16          # 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-llama-4-scout-17b-16e-instruct-quantized-w4a16:1.5
    tolerations:
    - effect: NoSchedule
      key: nvidia.com/gpu
      operator: Exists

bash

# 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

python

# 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": "Llama-4-Scout-17B-16E-Instruct-quantized.w4a16",
    "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.

Evaluation

The model was evaluated on the OpenLLM leaderboard tasks (v1 and v2), long context RULER, multimodal MMMU, and multimodal ChartQA. All evaluations are obtained through lm-evaluation-harness.

Evaluation details

OpenLLM v1

lm_eval \
  --model vllm \
  --model_args pretrained="RedHatAI/Llama-4-Scout-17B-16E-Instruct-quantized.w4a16",dtype=auto,add_bos_token=True,max_model_len=4096,tensor_parallel_size=8,gpu_memory_utilization=0.7,enable_chunked_prefill=True,trust_remote_code=True \
  --tasks openllm \
  --batch_size auto

OpenLLM v2

lm_eval \
  --model vllm \
  --model_args pretrained="RedHatAI/Llama-4-Scout-17B-16E-Instruct-quantized.w4a16",dtype=auto,add_bos_token=False,max_model_len=16384,tensor_parallel_size=8,gpu_memory_utilization=0.5,enable_chunked_prefill=True,trust_remote_code=True \
  --tasks leaderboard \
  --apply_chat_template \
  --fewshot_as_multiturn \
  --batch_size auto

Long Context RULER

lm_eval \
  --model vllm \
  --model_args pretrained="RedHatAI/Llama-4-Scout-17B-16E-Instruct-quantized.w4a16",dtype=auto,add_bos_token=False,max_model_len=524288,tensor_parallel_size=8,gpu_memory_utilization=0.9,enable_chunked_prefill=True,trust_remote_code=True \
  --tasks ruler \
  --metadata='{"max_seq_lengths":[131072]}' \
  --batch_size auto

Multimodal MMMU

lm_eval \
  --model vllm-vlm \
  --model_args pretrained="RedHatAI/Llama-4-Scout-17B-16E-Instruct-quantized.w4a16",dtype=auto,add_bos_token=False,max_model_len=1000000,tensor_parallel_size=8,gpu_memory_utilization=0.9,enable_chunked_prefill=True,trust_remote_code=True,max_images=10 \
  --tasks mmmu_val \
  --apply_chat_template \
  --batch_size auto

Multimodal ChartQA

export VLLM_MM_INPUT_CACHE_GIB=8
lm_eval \
  --model vllm-vlm \
  --model_args pretrained="RedHatAI/Llama-4-Scout-17B-16E-Instruct-quantized.w4a16",dtype=auto,add_bos_token=False,max_model_len=1000000,tensor_parallel_size=8,gpu_memory_utilization=0.9,enable_chunked_prefill=True,trust_remote_code=True,max_images=10 \
  --tasks chartqa \
  --apply_chat_template \
  --batch_size auto

Accuracy

	Recovery (%)	meta-llama/Llama-4-Scout-17B-16E-Instruct	RedHatAI/Llama-4-Scout-17B-16E-Instruct-quantized.w4a16 (this model)
ARC-Challenge 25-shot	98.51	69.37	68.34
GSM8k 5-shot	100.4	90.45	90.90
HellaSwag 10-shot	99.67	85.23	84.95
MMLU 5-shot	99.75	80.54	80.34
TruthfulQA 0-shot	99.82	61.41	61.30
WinoGrande 5-shot	98.98	77.90	77.11
OpenLLM v1 Average Score	99.59	77.48	77.16
IFEval 0-shot avg of inst and prompt acc	99.51	86.90	86.47
Big Bench Hard 3-shot	99.46	65.13	64.78
Math Lvl 5 4-shot	99.22	57.78	57.33
GPQA 0-shot	100.0	31.88	31.88
MuSR 0-shot	100.9	42.20	42.59
MMLU-Pro 5-shot	98.67	55.70	54.96
OpenLLM v2 Average Score	99.54	56.60	56.34
MMMU 0-shot	100.6	53.44	53.78
ChartQA 0-shot exact_match	100.1	65.88	66.00
ChartQA 0-shot relaxed_accuracy	99.55	88.92	88.52
Multimodal Average Score	100.0	69.41	69.43
RULER seqlen = 131072 niah_multikey_1	98.41	88.20	86.80
RULER seqlen = 131072 niah_multikey_2	94.73	83.60	79.20
RULER seqlen = 131072 niah_multikey_3	96.44	78.80	76.00
RULER seqlen = 131072 niah_multiquery	98.79	95.40	94.25
RULER seqlen = 131072 niah_multivalue	101.6	73.75	74.95
RULER seqlen = 131072 niah_single_1	100.0	100.00	100.0
RULER seqlen = 131072 niah_single_2	100.0	99.80	99.80
RULER seqlen = 131072 niah_single_3	100.2	99.80	100.0
RULER seqlen = 131072 ruler_cwe	87.39	39.42	33.14
RULER seqlen = 131072 ruler_fwe	98.13	92.93	91.20
RULER seqlen = 131072 ruler_qa_hotpot	100.4	48.20	48.40
RULER seqlen = 131072 ruler_qa_squad	96.22	53.57	51.55
RULER seqlen = 131072 ruler_qa_vt	98.82	92.28	91.20
RULER seqlen = 131072 Average Score	98.16	80.44	78.96

RedHatAI/Llama-4-Scout-17B-16E-Instruct-quantized.w4a16

Llama-4-Scout-17B-16E-Instruct-quantized.w4a16 Model Icon

Model Overview

Model Optimizations

Deployment

Evaluation

Accuracy

RedHatAI/
Llama-4-Scout-17B-16E-Instruct-quantized.w4a16

Llama-4-Scout-17B-16E-Instruct-quantized.w4a16
Model Icon