Glossary
Llama
Meta's open-weight model family, the most widely deployed open release through 2024 to 2026, released under the source-available Community License with an MAU cap and acceptable-use clause.
The model family that anchored the open weightsweightsA model release that publishes the trained parameters under some downloadable license, distinct from "open source" which (per OSAID) also requires data and training-code openness. Open full entry ecosystem from the original 7B/13B/33B/65B release in February 2023 onward. Llama 2 added chat-tuned variants and shipped under the Community License that is still in use; Llama 3 (April 2024, 8B and 70B) and Llama 3.1 (July 2024, adding the 405B) pushed competitive intelligence into the open tier; Llama 4 (April 2025) shifted to a mixture-of-experts architecture.
The license is the most-discussed aspect. The 700M MAU clause and
the acceptable-usegovernanceLicense or terms-of-service clauses that prohibit certain uses (weapons, surveillance, harassment, child sexual abuse material), common on open-weight licenses but rejected by the strict open-source definition.
Open full entry addendum exclude Llama from OSAIDgovernanceThe OSI's October 2024 definition of "open source AI," requiring not just weights but enough information about data, code, and architecture for third parties to reproduce the system.
Open full entry -compliant “open
source AI”; the practical effect on most users is none, but the
governance signal matters.
Full coverage on the project page at /projects/llama.
Sources
Mentioned in
- acceptable-use
- Apache 2.0
- Axolotl
- BPE
- dense
- fine-tuning
- Gemma
- GGUF
- GQA
- KV cache
- Llama Guard
- llama.cpp
- LM Studio
- local-first
- MAU
- MHA
- Mixtral
- mixture of experts
- on-device
- open weights
- OSAID
- post-training
- pretraining
- quantization
- RedPajama
- RLHF
- RoPE
- sliding window attention
- source-available
- throughput
- tokenizer
- Unsloth
- vLLM
- layer Weights