Architectural evolution
Architectural techniques do not appear at random: they diffuse through the field one laboratory at a time. This page traces two curves — the evolution of attention itself, and the adoption timeline of the operators that make modern LLMs trainable and cheap to serve.
Attention: from MHA to hybrid linear
Multi-head attention (MHA) computes one key–value head per query head. Multi-query (MQA) shares a single KV head across all queries; grouped-query (GQA) generalises this to a few shared heads and became the default for dense models after Llama 2 70B. Multi-head latent attention (MLA), introduced with DeepSeek-V2, compresses keys and values into a low-rank latent vector instead. Most recently, hybrid designs interleave linear-attention layers with a minority of full-attention layers.
multi-head attention (MHA) · 10
The original formulation: one key–value head per query head. Standard from GPT-2 through most pre-2023 dense models.
GPT-2 GPT-3 175B (paper) LLaMA 1 7B DeepSeek-Coder 6.7B DeepSeek LLM 7B DeepSeekMoE 16B Qwen1.5 7B Step-3 Step-3.5-Flash Step-3.7-Flash
multi-query attention (MQA) · 3
A single shared KV head serves all query heads — aggressive KV-cache savings at some quality cost; rare in frontier open models.
grouped-query attention (GQA) · 46
A few KV heads shared across groups of query heads. The default for dense open models since Llama 2 70B.
Llama 2 70B Mistral 7B ChatGLM3 6B Mixtral 8x22B GLM-4 9B Qwen2 72B Llama 3.1 70B Llama 3.1 8B Qwen2.5 72B Qwen2.5 7B MiniMax-Text-01 Qwen3 0.6B Qwen3 235B A22B Qwen3 30B A3B Qwen3 32B Qwen3 8B MiniMax-M1 GLM-4.5 GLM-4.5-Air Hunyuan 7B GPT-OSS 120B GPT-OSS 20B Qwen3-Next 80B A3B GLM-4.6 MiniMax-M2 Qwen3-Coder-Next MiniMax-M2.5 Qwen3.5 122B A10B Qwen3.5 27B Qwen3.5 35B A3B Qwen3.5 397B A17B Qwen3.5 9B Qwen3.5 9B Base Hunyuan 3 Preview Hunyuan 3 Preview Base MiniMax-M2.7 Qwen3.6 27B Qwen3.6 35B A3B Hunyuan-TurboS 30B A3B Hunyuan-TurboS 7B MiniMax-M3 MiniMax-M3 MXFP8 Hunyuan 3 Qwen3.8 2.4T A95B Qwen3.8 27B Qwen3.8-Flash-Next
multi-head latent attention (MLA) · 21
Keys and values are compressed into a low-rank latent vector, shrinking the KV cache dramatically; introduced with DeepSeek-V2.
DeepSeek-V2 DeepSeek-V2 Lite DeepSeek-V3 DeepSeek-V3 Base DeepSeek-R1 Moonlight 16B Kimi K2 Base DeepSeek-V3.1 DeepSeek-V3.2 Kimi K2 0905 Kimi Linear 48B A3B GLM-4.7-Flash GLM-5 Kimi K2.5 GLM-5.1 Kimi K2.6 GLM-5.2 Kimi K3 GLM-5.3 GLM-5.3-Flash Hunyuan 4 Preview
hybrid linear attention · 0
A minority of full-attention layers interleaved with linear-attention layers, trading long-context cost for near-constant memory.
Operator adoption
Each row is a technique; the year shown is when the earliest model in this corpus first shipped it in an open-weight release.
| Attention / Topic | Model | First open-weight adopter |
|---|---|---|
| MHA | 10 | GPT-2 (2019-02) |
| GQA | 46 | Llama 2 70B (2023-07) |
| MLA | 21 | DeepSeek-V2 (2024-05) |
| MQA | 3 | DeepSeek-V4-Flash (2026-04) |
| SWA | 12 | Mistral 7B (2023-09) |
| Linear attention (hybrid) | 0 | — |
| MoE | 55 | DeepSeekMoE 16B (2024-01) |
| RoPE | 58 | Mistral 7B (2023-09) |
| YaRN long-context | 0 | — |
| RMSNorm | 73 | LLaMA 1 7B (2023-02) |
| SwiGLU / SiLU | 70 | LLaMA 1 7B (2023-02) |
| Multi-Token Prediction | 39 | DeepSeek-V3 (2024-12) |
| FP8 training | 18 | DeepSeek-V3 (2024-12) |