| <!DOCTYPE html> |
| <html lang="en"> |
| <head> |
| <meta charset="utf-8"> |
| <meta name="viewport" content="width=device-width, initial-scale=1"> |
| <title>Speculative Decoding on a 16 GB Consumer GPU — paper</title> |
| <style> |
| :root { --ink:#1a1d21; --mut:#5b6470; --acc:#0f4c81; --line:#e3e6ea; } |
| * { box-sizing: border-box; } |
| body { |
| margin:0; font-family: -apple-system, "Segoe UI", Roboto, Helvetica, Arial, sans-serif; |
| color: var(--ink); background:#f7f8fa; line-height:1.5; |
| } |
| header { |
| max-width:1000px; margin:0 auto; padding:2rem 1.25rem 0.5rem; |
| } |
| h1 { font-size:1.45rem; margin:0 0 0.25rem; letter-spacing:-0.01em; } |
| .meta { color:var(--mut); font-size:0.92rem; margin-bottom:0.75rem; } |
| .links { display:flex; flex-wrap:wrap; gap:0.5rem; margin:0.75rem 0 0.25rem; } |
| .links a { |
| text-decoration:none; font-size:0.85rem; color:var(--acc); |
| border:1px solid var(--line); background:#fff; border-radius:999px; |
| padding:0.3rem 0.85rem; |
| } |
| .links a:hover { border-color:var(--acc); } |
| .wrap { max-width:1000px; margin:0 auto; padding:0 1.25rem 2.5rem; } |
| .pdf { |
| width:100%; height:calc(100vh - 220px); min-height:640px; |
| border:1px solid var(--line); border-radius:6px; background:#fff; |
| } |
| footer { |
| max-width:1000px; margin:0 auto; padding:0 1.25rem 2rem; |
| color:var(--mut); font-size:0.82rem; |
| } |
| </style> |
| </head> |
| <body> |
| <header> |
| <h1>Speculative Decoding on a 16 GB Consumer GPU</h1> |
| <div class="meta">Acceptance, modern drafters, and quantization in Gemma 4 12B and Qwen3-8B |
| · Martin Ulises Millan Guerrero · 2026-08-08</div> |
| <div class="links"> |
| <a href="https://huggingface.co/reyden009/speculative-decoding-lab" target="_blank" rel="noopener">Companion repo (code + data)</a> |
| <a href="https://huggingface.co/reyden009/speculative-decoding-lab/resolve/main/paper/paper.pdf" target="_blank" rel="noopener">PDF (direct)</a> |
| <a href="https://huggingface.co/reyden009/speculative-decoding-lab/resolve/main/paper/arxiv-bundle.tar.gz" target="_blank" rel="noopener">arXiv-ready bundle</a> |
| <a href="https://huggingface.co/reyden009/speculative-decoding-lab/resolve/main/paper/paper.md" target="_blank" rel="noopener">Markdown source</a> |
| </div> |
| </header> |
| <div class="wrap"> |
| <iframe class="pdf" src="https://huggingface.co/reyden009/speculative-decoding-lab/resolve/main/paper/paper.pdf#toolbar=1" |
| title="Paper PDF" allowfullscreen></iframe> |
| </div> |
| <footer> |
| Setup: RTX 4060 Ti 16 GB · Intel i5-12400 · llama.cpp (CUDA) · Gemma 4 12B and Qwen3-8B GGUF · everything user-space, no sudo. |
| </footer> |
| </body> |
| </html> |
|
|