Llama 3.1 8B
Release in the Llama family · version 3.1 8B
Llama 3.1 8B is the smallest model in Meta's Llama 3.1 collection, released on July 23, 2024 in pretrained and instruction-tuned versions. It is a text-only, auto-regressive transformer with grouped-query attention and a 128K context length. The model card lists eight supported languages and says the collection was pretrained on about 15 trillion tokens with a December 2023 cutoff.15
- Documentation: Model card (external site: github.com)
- Model hub: Hugging Face (pretrained, gated) (external site: huggingface.co)
- Model hub: Hugging Face (instruction-tuned, gated) (external site: huggingface.co)
- License: Llama 3.1 Community License Agreement (external site: github.com)
- Repository: Reference implementation (Llama 3) (external site: github.com)
Availability and license
Overall availability
Weights are downloadable after an access request and license acceptance. The Hugging Face repositories are gated with manual approval and ask for name, date of birth, country, affiliation, and job title; Meta's own download instructions send a signed, expiring URL by email once a request is approved.78105
Availability is separate from permission: read the license before using or redistributing.
Llama 3.1 Community License Agreement (external site: github.com)217
Plain-language guide to the Llama 3.1 Community License Agreement
Custom Meta license dated July 23, 2024; the Hugging Face metadata lists it as llama3.1. It grants a royalty-free, non-exclusive license to use, modify, and redistribute. Redistributors must include the agreement, show "Built with Llama", and keep an attribution notice, and distributed models trained or fine-tuned with Llama materials or outputs must start their names with "Llama". Use must follow the incorporated Llama 3.1 Acceptable Use Policy. Licensees whose products had more than 700 million monthly active users in the month before the release date must request a separate license. The license ends for anyone who sues alleging that the Llama materials or outputs infringe their rights. California law governs.237
Component reuse rights
- weights
- Unknown — no complete fact-level rights review
- code
- Unknown — no complete fact-level rights review
- data
- Unknown — no complete fact-level rights review
- documentation
- Unknown — no complete fact-level rights review
No complete system-rights review is recorded for this release.
Model-disclosure tier
The weights can be obtained only by request, with approval, or by some users — for example a gated download that the publisher reviews. Not counted as open-weight.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Partial | BF16 safetensors (8,030,261,248 parameters per the repository metadata) plus original consolidated weights, in Hugging Face repositories gated with manual approval.710 |
| Inference codeIs code for running the model published? | Public | Meta's llama-models repository includes Llama 3 model, generation, tokenizer, and quantization code with example scripts, and the Hugging Face metadata identifies the Transformers LlamaForCausalLM architecture.657 |
| Training codeIs the code used to train the model published? | Unknown | The model card says Meta used custom training libraries; this review found no published training code.1 |
| Training-data informationDoes the information cover provenance, scope, acquisition, selection, labeling, processing, and where data or alternatives can be obtained? Access alone does not establish completeness. | Partial | The card describes about 15 trillion pretraining tokens from publicly available sources with a December 2023 cutoff, and fine-tuning data that includes public instruction datasets and over 25 million synthetic examples. The data is not released, and this review does not judge the description complete.1 |
| Training-data accessCan the training data be obtained? This is independent of information completeness and reuse rights; original unshareable data need not be downloadable. | Unknown | Not assessed. |
| Complete training pipelineIs the complete base-training and preprocessing pipeline published, including configuration? Fine-tuning code or an inference SDK alone is insufficient. | Unknown | Not assessed. |
| Legacy data assessment (v0.1)Historical assessment combining download access and disclosure. Preserved for traceability; excluded from the v0.2 tier calculation. See the new separate assessments above. | Unknown | Not assessed. |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Partial | The card says the tuned versions use supervised fine-tuning and reinforcement learning with human feedback, and that pretraining used Meta's custom libraries and GPU cluster. Meta's Llama 3 technical paper was not assessed in this review.1 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Partial | The card reports benchmark results, and Meta's evaluation notes describe the settings and point to released evaluation outputs and a reproduction recipe using lm-evaluation-harness. The evaluation datasets on Hugging Face are gated with manual approval, and the recipe was not checked.149 |
What it is useful for
The model card lists commercial and research use in its supported languages: assistant-style chat for the instruction-tuned model, adapting the pretrained model to other text generation tasks, and using model outputs for synthetic data generation and distillation. Other developers have used it as a base model; Ai2's Llama 3.1 Tülu 3.1 8B, for example, is post-trained from it.112
Run and use notes
- llama.cpp's quantize README lists GGUF sizes for Llama 3.1 8B of 14.96 GiB at F16 (16 bits per weight), 7.95 GiB for Q8_0 (about 8.5 bits per weight), and 4.58 GiB for Q4_K_M (about 4.89 bits per weight). It says memory and disk requirements are currently the same because models are fully loaded into memory.11
- The model card treats languages beyond its eight supported ones as out of scope unless a developer fine-tunes for them in line with the license and acceptable use policy.1
Organization context
Other releases in the Llama family
- Llama 4 Maverick (17Bx128E)Model-disclosure tier (USASI rubric v0.2): Restricted weights
- Llama 4 Scout (17Bx16E)Model-disclosure tier (USASI rubric v0.2): Restricted weights
- Llama Guard 4 12BModel-disclosure tier (USASI rubric v0.2): Restricted weights
What this catalog does not know
- Training code: unknown.
- Training-data access: unknown.
- Complete training pipeline: unknown.
- Legacy data assessment (v0.1): unknown.
Have a primary source? How to report a correction.
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.