Model Overview
Model Features
Model Capabilities
Use Cases
đ Llama 4 Model
The Llama 4 models are natively multimodal AI models, enabling text and multimodal experiences. They leverage a mixture - of - experts architecture, offering industry - leading performance in text and image understanding.
đ Quick Start
Prerequisites
Please, make sure you have transformers
v4.51.0
installed, or upgrade using pip install -U transformers
.
Example Code
from transformers import AutoProcessor, Llama4ForConditionalGeneration
import torch
model_id = "meta-llama/Llama-4-Scout-17B-16E-Instruct"
processor = AutoProcessor.from_pretrained(model_id)
model = Llama4ForConditionalGeneration.from_pretrained(
model_id,
attn_implementation="flex_attention",
device_map="auto",
torch_dtype=torch.bfloat16,
)
url1 = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/0052a70beed5bf71b92610a43a52df6d286cd5f3/diffusers/rabbit.jpg"
url2 = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/datasets/cat_style_layout.png"
messages = [
{
"role": "user",
"content": [
{"type": "image", "url": url1},
{"type": "image", "url": url2},
{"type": "text", "text": "Can you describe how these two images are similar, and how they differ?"},
]
},
]
inputs = processor.apply_chat_template(
messages,
add_generation_prompt=True,
tokenize=True,
return_dict=True,
return_tensors="pt",
).to(model.device)
outputs = model.generate(
**inputs,
max_new_tokens=256,
)
response = processor.batch_decode(outputs[:, inputs["input_ids"].shape[-1]:])[0]
print(response)
print(outputs[0])
⨠Features
- Multimodal Capability: The Llama 4 models are natively multimodal, enabling text and multimodal experiences, with strong performance in text and image understanding.
- Mixture - of - Experts Architecture: Leveraging this architecture to offer industry - leading performance.
- Model Variants: Two efficient models in the Llama 4 series are launched, Llama 4 Scout and Llama 4 Maverick.
đĻ Installation
Ensure you have transformers
v4.51.0
installed. You can upgrade it using the following command:
pip install -U transformers
đģ Usage Examples
Basic Usage
from transformers import AutoProcessor, Llama4ForConditionalGeneration
import torch
model_id = "meta-llama/Llama-4-Scout-17B-16E-Instruct"
processor = AutoProcessor.from_pretrained(model_id)
model = Llama4ForConditionalGeneration.from_pretrained(
model_id,
attn_implementation="flex_attention",
device_map="auto",
torch_dtype=torch.bfloat16,
)
url1 = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/0052a70beed5bf71b92610a43a52df6d286cd5f3/diffusers/rabbit.jpg"
url2 = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/datasets/cat_style_layout.png"
messages = [
{
"role": "user",
"content": [
{"type": "image", "url": url1},
{"type": "image", "url": url2},
{"type": "text", "text": "Can you describe how these two images are similar, and how they differ?"},
]
},
]
inputs = processor.apply_chat_template(
messages,
add_generation_prompt=True,
tokenize=True,
return_dict=True,
return_tensors="pt",
).to(model.device)
outputs = model.generate(
**inputs,
max_new_tokens=256,
)
response = processor.batch_decode(outputs[:, inputs["input_ids"].shape[-1]:])[0]
print(response)
print(outputs[0])
đ Documentation
Model Information
- Model developer: Meta
- Model Architecture: The Llama 4 models are auto - regressive language models that use a mixture - of - experts (MoE) architecture and incorporate early fusion for native multimodality.
Property | Details |
---|---|
Model Type | Llama 4 Scout (17Bx16E), Llama 4 Maverick (17Bx128E) |
Training Data | A mix of publicly available, licensed data and information from Meta's products and services. This includes publicly shared posts from Instagram and Facebook and people's interactions with Meta AI. Learn more in the Privacy Center. |
Params (Llama 4 Scout) | 17B (Activated), 109B (Total) |
Params (Llama 4 Maverick) | 17B (Activated), 400B (Total) |
Input modalities | Multilingual text and image |
Output modalities | Multilingual text and code |
Context length (Llama 4 Scout) | 10M |
Context length (Llama 4 Maverick) | 1M |
Token count (Llama 4 Scout) | ~40T |
Token count (Llama 4 Maverick) | ~22T |
Knowledge cutoff | August 2024 |
Supported languages: Arabic, English, French, German, Hindi, Indonesian, Italian, Portuguese, Spanish, Tagalog, Thai, and Vietnamese.
Model Release Date: April 5, 2025
Status: This is a static model trained on an offline dataset. Future versions of the tuned models may be released as we improve model behavior with community feedback.
License: A custom commercial license, the Llama 4 Community License Agreement, is available at: [https://github.com/meta - llama/llama - models/blob/main/models/llama4/LICENSE](https://github.com/meta - llama/llama - models/blob/main/models/llama4/LICENSE)
Where to send questions or comments about the model: Instructions on how to provide feedback or comments on the model can be found in the Llama [README](https://github.com/meta - llama/llama - models/blob/main/README.md). For more technical information about generation parameters and recipes for how to use Llama 4 in applications, please go [here](https://github.com/meta - llama/llama - cookbook).
Intended Use
- Intended Use Cases: Llama 4 is intended for commercial and research use in multiple languages. Instruction tuned models are intended for assistant - like chat and visual reasoning tasks, whereas pretrained models can be adapted for natural language generation. For vision, Llama 4 models are also optimized for visual recognition, image reasoning, captioning, and answering general questions about an image. The Llama 4 model collection also supports the ability to leverage the outputs of its models to improve other models including synthetic data generation and distillation. The Llama 4 Community License allows for these use cases.
- Out - of - scope: Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in any other way that is prohibited by the Acceptable Use Policy and Llama 4 Community License. Use in languages or capabilities beyond those explicitly referenced as supported in this model card.
Hardware and Software
- Training Factors: Custom training libraries, Meta's custom built GPU clusters, and production infrastructure were used for pretraining. Fine - tuning, quantization, annotation, and evaluation were also performed on production infrastructure.
- Training Energy Use: Model pre - training utilized a cumulative of 7.38M GPU hours of computation on H100 - 80GB (TDP of 700W) type hardware. Training time is the total GPU time required for training each model and power consumption is the peak power capacity per GPU device used, adjusted for power usage efficiency.
- Training Greenhouse Gas Emissions: Estimated total location - based greenhouse gas emissions were 1,999 tons CO2eq for training. Since 2020, Meta has maintained net zero greenhouse gas emissions in its global operations and matched 100% of its electricity use with clean and renewable energy; therefore, the total market - based greenhouse gas emissions for training were 0 tons CO2eq.
Model Name | Training Time (GPU hours) | Training Power Consumption (W) | Training Location - Based Greenhouse Gas Emissions (tons CO2eq) | Training Market - Based Greenhouse Gas Emissions (tons CO2eq) |
---|---|---|---|---|
Llama 4 Scout | 5.0M | 700 | 1,354 | 0 |
Llama 4 Maverick | 2.38M | 700 | 645 | 0 |
Total | 7.38M | - | 1,999 | 0 |
The methodology used to determine training energy use and greenhouse gas emissions can be found here.
Benchmarks
Pre - trained models
Category | Benchmark | # Shots | Metric | Llama 3.1 70B | Llama 3.1 405B | Llama 4 Scout | Llama 4 Maverick |
---|---|---|---|---|---|---|---|
Reasoning & Knowledge | MMLU | 5 | macro_avg/acc_char | 79.3 | 85.2 | 79.6 | 85.5 |
MMLU - Pro | 5 | macro_avg/em | 53.8 | 61.6 | 58.2 | 62.9 | |
MATH | 4 | em_maj1@1 | 41.6 | 53.5 | 50.3 | 61.2 | |
Code | MBPP | 3 | pass@1 | 66.4 | 74.4 | 67.8 | 77.6 |
Multilingual | TydiQA | 1 | average/f1 | 29.9 | 34.3 | 31.5 | 31.7 |
Image | ChartQA | 0 | relaxed_accuracy | No multimodal support | 83.4 | 85.3 | |
DocVQA | 0 | anls | 89.4 | 91.6 |
Instruction tuned models
Category | Benchmark | # Shots | Metric | Llama 3.3 70B | Llama 3.1 405B | Llama 4 Scout | Llama 4 Maverick |
---|---|---|---|---|---|---|---|
Image Reasoning | MMMU | 0 | accuracy | No multimodal support | 69.4 | 73.4 | |
MMMU Pro^ | 0 | accuracy | 52.2 | 59.6 | |||
MathVista | 0 | accuracy | 70.7 | 73.7 | |||
Image Understanding | ChartQA | 0 | relaxed_accuracy | 88.8 | 90.0 | ||
DocVQA (test) | 0 | anls | 94.4 | 94.4 | |||
Coding | LiveCodeBench (10/01/2024 - 02/01/2025) | 0 | pass@1 | 33.3 | 27.7 | 32.8 | 43.4 |
Reasoning & Knowledge | MMLU Pro | 0 | macro_avg/acc | 68.9 | 73.4 | 74.3 | 80.5 |
GPQA Diamond | 0 | accuracy | 50.5 | 49.0 | 57.2 | 69.8 | |
Multilingual | MGSM | 0 | average/em | 91.1 | 91.6 | 90.6 | 92.3 |
Long context | MTOB (half book) eng->kgv/kgv->eng | - | chrF | Context window is 128K | 42.2/36.6 | 54.0/46.4 | |
MTOB (full book) eng->kgv/kgv->eng | - | chrF | 39.7/36.3 | 50.8/46.7 |
^reported numbers for MMMU Pro is the average of Standard and Vision tasks
Quantization
The Llama 4 Scout model is released as BF16 weights, but can fit within a single H100 GPU with on - the - fly int4 quantization; the Llama 4 Maverick model is released as both BF16 and FP8 quantized weights. The FP8 quantized weights fit on a single H100 DGX host while still maintaining quality.
Intended Use Notes
â ī¸ Important Note
- Llama 4 has been trained on a broader collection of languages than the 12 supported languages (pre - training includes [200 total languages](https://ai.meta.com/research/no - language - left - behind/)). Developers may fine - tune Llama 4 models for languages beyond the 12 supported languages provided they comply with the Llama 4 Community License and the Acceptable Use Policy. Developers are responsible for ensuring that their use of Llama 4 in additional languages is done in a safe and responsible manner.
- Llama 4 has been tested for image understanding up to 5 input images. If leveraging additional image understanding capabilities beyond this, Developers are responsible for ensuring that their deployments are mitigated for risks and should perform additional testing and tuning tailored to their specific applications.
đ§ Technical Details
Training Factors
We used custom training libraries, Meta's custom built GPU clusters, and production infrastructure for pretraining. Fine - tuning, quantization, annotation, and evaluation were also performed on production infrastructure.
Training Energy Use
Model pre - training utilized a cumulative of 7.38M GPU hours of computation on H100 - 80GB (TDP of 700W) type hardware. Training time is the total GPU time required for training each model and power consumption is the peak power capacity per GPU device used, adjusted for power usage efficiency.
Training Greenhouse Gas Emissions
Estimated total location - based greenhouse gas emissions were 1,999 tons CO2eq for training. Since 2020, Meta has maintained net zero greenhouse gas emissions in its global operations and matched 100% of its electricity use with clean and renewable energy; therefore, the total market - based greenhouse gas emissions for training were 0 tons CO2eq.
đ License
A custom commercial license, the Llama 4 Community License Agreement, is available at: [https://github.com/meta - llama/llama - models/blob/main/models/llama4/LICENSE](https://github.com/meta - llama/llama - models/blob/main/models/llama4/LICENSE)







