LLM Context Window Explained: From 4K to 1M Tokens

LLM Context Window Explained: From 4K to 1M Tokens

Context window size is one of the most important — and most misunderstood — specs for local LLM inference. More context means higher memory usage, slower inference, and sometimes lower quality on short tasks. Understanding the tradeoffs helps you choose the right context size for your workload rather than defaulting to maximum.

Key fact: Every additional 1K tokens of context adds ~1MB to KV cache (in FP16, for a 7B model). Running a 7B model at 32K context vs 4K context adds 28MB to VRAM usage. For hardware limits by context size, see LLMHardware.io's VRAM calculator.

How Context Window Works

The context window is the total amount of text the model can "see" at once — including your system prompt, conversation history, and the current message. When it fills up, older content is dropped (truncated from the beginning by default in most tools). The KV cache stores attention states for all context tokens in VRAM, which is why larger contexts use more memory proportionally.

Context vs Quality: The Lost in the Middle Problem

Research shows most LLMs degrade in quality when important information is in the middle of very long contexts. Models reliably recall information from the beginning and end of context, but forget content buried in the middle. This "lost in the middle" effect means a 128K context window doesn't guarantee the model effectively uses all 128K tokens. For RAG applications, smaller focused context often outperforms large unfocused context.

Practical Recommendations

For conversation: 8K-16K is sufficient for most interactions. For document analysis: match to document size but cap at 32K unless needed. For code generation with large codebases: 32K-64K. For document summarization pipelines: use chunking + RAG rather than hoping context alone works. Enable --flash-attn in llama.cpp when using contexts over 8K — it significantly reduces memory pressure and speeds up prefill.

Comments

Post a Comment

Popular posts from this blog

AMD RX 9070 XT for Local LLMs: ROCm Performance Guide (2026)

The AMD RX 9070 XT launched in early 2026 at $599 with 16GB GDDR6 and 648 GB/s memory bandwidth. For local LLM users who want a mid-range GPU without NVIDIA prices, this changes the calculus significantly. Here is what it actually delivers. RX 9070 XT Specs vs Competition GPU VRAM Bandwidth Price LLM Score RX 9070 XT 16 GB 648 GB/s $599 ~60 tok/s RTX 4070 Ti Super 16 GB 672 GB/s $799 ~62 tok/s RTX 4080 Super 16 GB 736 GB/s $999 ~68 tok/s RTX 4090 24 GB 1,008 GB/s ~$1,100 ~90 tok/s At $599 with 16GB VRAM, the RX 9070 XT is the best value GPU for local LLMs in the mid-range tier in 2026. See the full comparison at LLMHardware.io's AMD GPU guide . ROCm Setup on Linux (2026) # Ubuntu 22.04/24.04 sudo apt install rocm-hip-libraries hip-base # or use the AMD installer: wget https://repo.radeon.com/amdgpu-install/6.3/ubuntu/jammy/amdgpu-install_6.3.60300-1_all.deb sudo dpkg -i amdgpu-install*.deb sudo amdgpu-install --usecase=rocm,lrt,openclsdk # Verify ROCm sees your GPU r...
Read more

llama.cpp flash attention: how to enable and benchmark

Image
Flash attention computes attention with far less memory traffic. The math is identical; the memory movement is not. In llama.cpp it matters most in two situations: long context windows and tight video memory (VRAM). With flash attention enabled, the key-value cache shrinks substantially. A 32k-token context window becomes realistic on a 16GB card, and the prompt-processing phase speeds up, so long documents get ingested faster. The catch is compatibility: not every model and quantization combination supports it, and on some setups the speedup is modest. If you run short chats on a strong GPU, you may barely notice. Where it shines is retrieval-augmented generation (RAG) and long-document work on mid-range cards. Enable it, compare your tokens per second, and keep it if your workload benefits. For the common case of pasting a long PDF and asking questions, it is close to free performance. See how context length changes VRAM needs .
Read more

SillyTavern with Local LLMs: Complete Setup for AI Companions

Image
SillyTavern is the most feature-rich frontend for character-based AI conversations. What began as an AI roleplay tool has evolved into a powerful platform with persona management, memory systems, extensive prompt engineering tools, and connections to every major local inference backend. Here's how to set it up with local models. Quick connect: SillyTavern connects to Ollama, KoboldCpp, llama.cpp server, and TabbyAPI (ExLlamaV2) out of the box. Any of the hardware setups at LLMHardware.io provides a solid backend for SillyTavern. Installation Install via git: git clone https://github.com/SillyTavern/SillyTavern && cd SillyTavern && npm install then node server.js . Access at localhost:8000. For persistent installation, use the included start.sh script. SillyTavern uses Node.js and runs on any platform. The companion app (STMP) enables mobile access to your SillyTavern instance via Tailscale or local network. Character Cards and the TavernAI Format Chara...
Read more