| <!DOCTYPE html> |
| <html lang="en"> |
| <head> |
| <meta charset="utf-8"> |
| <title>Qwopus3.6-35B-A3B-v1 — Evaluation Suite</title> |
| <style> |
| :root { color-scheme: dark; } |
| * { box-sizing: border-box; } |
| body { |
| font-family: -apple-system, BlinkMacSystemFont, "SF Pro Text", "Inter", system-ui, sans-serif; |
| background: radial-gradient(ellipse at top, #0e1a30 0%, #07080d 60%); |
| color: #e9ecf4; |
| padding: 4rem 1.5rem 6rem; |
| margin: 0; min-height: 100vh; line-height: 1.55; |
| } |
| .wrap { max-width: 920px; margin: 0 auto; } |
| header { margin-bottom: 3rem; } |
| .eyebrow { |
| color: #7dd3fc; font-size: 0.78rem; letter-spacing: 0.22em; |
| text-transform: uppercase; font-weight: 600; |
| } |
| h1 { |
| font-size: clamp(1.8rem, 4vw, 2.6rem); |
| font-weight: 700; letter-spacing: -0.025em; |
| margin: 0.35rem 0 0.4rem; |
| background: linear-gradient(135deg, #ffffff 0%, #7dd3fc 100%); |
| -webkit-background-clip: text; background-clip: text; |
| -webkit-text-fill-color: transparent; |
| } |
| .byline { color: #7dd3fc; font-size: 0.92rem; margin: 0.1rem 0 0.9rem; } |
| .byline a { color: inherit; text-decoration: none; |
| border-bottom: 1px solid rgba(125,211,252,0.25); } |
| .byline a:hover { border-bottom-color: #7dd3fc; } |
| .sub { color: #9aa3bb; max-width: 64ch; } |
| |
| .cta-row { display: flex; flex-wrap: wrap; gap: 0.7rem; align-items: center; |
| margin-top: 1.5rem; } |
| .report-link { |
| display: inline-flex; align-items: center; gap: 0.5rem; |
| padding: 0.85rem 1.3rem; |
| background: linear-gradient(135deg, #0ea5e9 0%, #6366f1 100%); |
| color: #fff; border-radius: 10px; text-decoration: none; |
| font-weight: 600; font-size: 0.95rem; |
| transition: transform 0.2s ease, filter 0.2s ease; |
| } |
| .report-link:hover { transform: translateY(-1px); filter: brightness(1.1); } |
| .report-link::after { content: " →"; } |
| .compare-link { |
| display: inline-flex; align-items: center; gap: 0.5rem; |
| padding: 0.85rem 1.2rem; |
| background: rgba(255,255,255,0.05); |
| border: 1px solid rgba(255,255,255,0.1); |
| color: #e9ecf4; border-radius: 10px; text-decoration: none; |
| font-weight: 600; font-size: 0.95rem; |
| transition: transform 0.2s ease, background 0.2s ease, border-color 0.2s ease; |
| } |
| .compare-link:hover { transform: translateY(-1px); |
| background: rgba(255,255,255,0.09); |
| border-color: rgba(255,255,255,0.2); } |
| .follow-link { |
| display: inline-flex; align-items: center; gap: 0.55rem; |
| padding: 0.85rem 1.2rem; |
| background: rgba(255,255,255,0.055); |
| border: 1px solid rgba(255,255,255,0.1); |
| color: #e9ecf4; border-radius: 10px; |
| text-decoration: none; font-weight: 600; font-size: 0.95rem; |
| transition: transform 0.2s ease, background 0.2s ease, border-color 0.2s ease; |
| } |
| .follow-link:hover { transform: translateY(-1px); background: rgba(255,255,255,0.09); |
| border-color: rgba(255,255,255,0.2); } |
| .follow-link svg { width: 15px; height: 15px; } |
| |
| .stats { |
| display: grid; |
| grid-template-columns: repeat(auto-fit, minmax(150px, 1fr)); |
| gap: 1rem; margin: 2rem 0 3rem; |
| padding: 1.25rem; |
| background: rgba(255,255,255,0.03); |
| border: 1px solid rgba(255,255,255,0.06); |
| border-radius: 14px; |
| } |
| .stat b { display:block; font-size: 1.6rem; font-weight: 700; |
| letter-spacing: -0.02em; color: #ffffff; } |
| .stat span { color: #7e8aa5; font-size: 0.78rem; |
| text-transform: uppercase; letter-spacing: 0.12em; } |
| .stat .delta { display: block; color: #86efac; font-size: 0.85rem; |
| font-weight: 600; margin-top: 0.25rem; } |
| |
| h2 { |
| font-size: 0.85rem; color: #9fa7c0; |
| text-transform: uppercase; letter-spacing: 0.14em; |
| font-weight: 600; margin: 3rem 0 1rem; |
| padding-bottom: 0.6rem; |
| border-bottom: 1px solid rgba(255,255,255,0.07); |
| } |
| |
| .grid { |
| display: grid; |
| grid-template-columns: repeat(auto-fill, minmax(280px, 1fr)); |
| gap: 0.9rem; |
| } |
| .card { |
| display: block; padding: 1.1rem 1.2rem; |
| background: rgba(255,255,255,0.035); |
| border: 1px solid rgba(255,255,255,0.07); |
| border-radius: 11px; |
| color: inherit; text-decoration: none; |
| transition: transform 0.22s ease, border-color 0.22s ease, background 0.22s ease; |
| } |
| .card:hover { |
| transform: translateY(-2px); |
| border-color: rgba(125, 211, 252, 0.45); |
| background: rgba(125, 211, 252, 0.07); |
| } |
| .card strong { |
| display: block; font-weight: 600; font-size: 0.98rem; |
| color: #f3f5fa; margin-bottom: 0.25rem; |
| } |
| .card span { color: #8b93ab; font-size: 0.82rem; } |
| .meta { |
| font-family: ui-monospace, SFMono-Regular, Menlo, Consolas, monospace; |
| font-size: 0.72rem; color: #6c7693; margin-top: 0.6rem; |
| } |
| |
| footer { |
| margin-top: 5rem; color: #5a6380; |
| font-size: 0.82rem; text-align: center; |
| } |
| |
| .preview-banner { |
| display: flex; align-items: flex-start; gap: 0.9rem; |
| margin: 2rem 0 0; |
| padding: 1.1rem 1.3rem; |
| background: linear-gradient(135deg, rgba(56,189,248,0.09), rgba(99,102,241,0.07)); |
| border: 1px solid rgba(56,189,248,0.28); |
| border-radius: 12px; |
| font-size: 0.92rem; color: #e9ecf4; line-height: 1.55; |
| } |
| .preview-banner .icon { |
| flex-shrink: 0; |
| width: 28px; height: 28px; border-radius: 50%; |
| background: rgba(56,189,248,0.18); |
| display: grid; place-items: center; |
| font-size: 0.95rem; color: #7dd3fc; font-weight: 700; |
| } |
| .preview-banner .pv-title { |
| font-weight: 700; color: #bae6fd; letter-spacing: 0.02em; |
| display: block; margin-bottom: 0.2rem; |
| } |
| .preview-banner a { color: #bae6fd; text-decoration: none; |
| border-bottom: 1px solid rgba(186,230,253,0.35); } |
| </style> |
| </head> |
| <body> |
| <div class="wrap"> |
| <header> |
| <div class="eyebrow">MoE A3B Fine-tune · 35B Total / 3B Active</div> |
| <h1>Qwopus3.6-35B-A3B-v1</h1> |
| <p class="byline">by <a href="https://x.com/KyleHessling1" target="_blank" rel="noopener">Kyle Hessling</a> · built on <a href="https://huggingface.co/Jackrong/Qwopus3.6-35B-A3B-v1" target="_blank" rel="noopener">Jackrong's MoE fine-tune</a></p> |
| <p class="sub">Same 17-prompt suite as the Qwopus3.6-27B v1-preview eval — 5 agentic + 1 nothink rerun, 5 web-design, 6 canvas/WebGL. Q5_K_M on a single RTX 5090 via llama.cpp. Headline: <b>162 tok/s avg</b>, <b>129k tokens generated</b>, <b>13.3 min runtime</b>. The MoE is <b>2.6× faster</b> than the dense 27B preview at a larger quant. The web-design outputs are some of the best one-shot HTML I've seen out of any open model in this size class — verbose in a good way, the pages feel <i>complete</i> on the first attempt where most models land surface-level.</p> |
| <div class="preview-banner"> |
| <div class="icon">i</div> |
| <div> |
| <span class="pv-title">A real upgrade for one-shot front-end on a 5090</span> |
| Per the <a href="https://huggingface.co/Jackrong/Qwopus3.6-35B-A3B-v1" target="_blank" rel="noopener">model card</a>, this is a 9%-trainable LoRA fine-tune over Qwen3.6-35B-A3B with three-stage curriculum SFT. The MoE speedup (3 B active of 35 B) is a massive practical win, and an updated <b>Qwopus3.6-27B</b> is in the works that should carry similar improvements forward. |
| </div> |
| </div> |
| <div class="cta-row"> |
| <a href="report.html" class="report-link">Read the full report</a> |
| <a href="https://huggingface.co/Jackrong/Qwopus3.6-35B-A3B-v1-GGUF" target="_blank" rel="noopener" class="compare-link">GGUF on Hugging Face</a> |
| <a href="https://huggingface.co/spaces/KyleHessling1/qwopus36-eval" target="_blank" rel="noopener" class="compare-link">Compare · Qwopus3.6-27B preview eval</a> |
| <a href="https://x.com/KyleHessling1" target="_blank" rel="noopener" class="follow-link"> |
| <svg viewBox="0 0 24 24" fill="currentColor" aria-hidden="true"><path d="M18.244 2.25h3.308l-7.227 8.26 8.502 11.24H16.17l-5.214-6.817L4.99 21.75H1.68l7.73-8.835L1.254 2.25H8.08l4.713 6.231zm-1.161 17.52h1.833L7.084 4.126H5.117z"/></svg> |
| Follow @KyleHessling1 |
| </a> |
| </div> |
| </header> |
|
|
| <div class="stats"> |
| <div class="stat"><b>161.9</b><span>avg tok/s</span><span class="delta">×2.6 vs 27B preview</span></div> |
| <div class="stat"><b>14</b><span>shipped runs</span></div> |
| <div class="stat"><b>106,688</b><span>completion tokens</span></div> |
| <div class="stat"><b>~25 GB</b><span>VRAM used</span></div> |
| <div class="stat"><b>65K</b><span>context window</span></div> |
| </div> |
|
|
| <h2>Web design · open to preview</h2> |
| <div class="grid"> |
| <a class="card" href="design_saas_landing.html"><strong>SaaS landing page</strong><span>Prism — AI observability</span><div class="meta">75.9 KB · 23,839 tok · 151 s</div></a> |
| <a class="card" href="design_analytics_dashboard.html"><strong>Analytics dashboard</strong><span>Light theme, emerald accent</span><div class="meta">37.5 KB · 14,029 tok · 87 s</div></a> |
| <a class="card" href="design_designer_portfolio.html"><strong>Designer portfolio</strong><span>Maya Chen — kinetic typography</span><div class="meta">27.5 KB · 9,144 tok · 56 s</div></a> |
| <a class="card" href="design_pricing_page.html"><strong>Pricing page</strong><span>3 tiers + animated toggle + FAQ</span><div class="meta">50.1 KB · 13,862 tok · 86 s</div></a> |
| <a class="card" href="design_mobile_app_marketing.html"><strong>Mobile app marketing</strong><span>Stillwater — CSS-only iPhone mock</span><div class="meta">47.9 KB · 16,596 tok · 104 s</div></a> |
| </div> |
|
|
| <h2>Canvas / WebGL · creative coding</h2> |
| <p style="color:#9aa3bb; font-size:0.9rem; margin:0 0 1rem; max-width:64ch;">3 of 6 creative-canvas runs shipped clean and are below. The Mandelbulb shader, soft-body physics sandbox, and audio-reactive visualizer didn't render correctly on first attempt — these are prompts most one-shot models in this size class fail on, and they're typically the kind of brief that needs a second turn to fix shader errors or collision math. Excluded here for the headline; raw outputs are still in the repo if you want to inspect.</p> |
| <div class="grid"> |
| <a class="card" href="canvas_particle_attractor.html"><strong>Particle attractor</strong><span>3000-particle fluid swarm</span><div class="meta">10.6 KB · 4,152 tok · 25 s</div></a> |
| <a class="card" href="canvas_generative_flowfield.html"><strong>Generative flowfield</strong><span>Ink-line agents on simplex noise</span><div class="meta">19.3 KB · 6,934 tok · 42 s</div></a> |
| <a class="card" href="canvas_three_scene.html"><strong>Three.js crystal scene</strong><span>Transmissive glass + bloom</span><div class="meta">16.1 KB · 5,674 tok · 35 s</div></a> |
| </div> |
|
|
| <h2>Agentic reasoning · text output</h2> |
| <div class="grid"> |
| <a class="card" href="agentic_multi_step_planning.txt"><strong>Multi-step planning</strong><span>URL shortener deploy plan</span><div class="meta">thinking: 2,440 tok · 15 s</div></a> |
| <a class="card" href="agentic_self_critique.txt"><strong>Self-critique loop</strong><span>Palindrome · O(n³) → O(n²)</span><div class="meta">thinking: 4,391 tok · 27 s</div></a> |
| <a class="card" href="agentic_code_debug.txt"><strong>Code debug (4 bugs)</strong><span>k-th smallest element</span><div class="meta">thinking: 1,393 tok · 9 s</div></a> |
| <a class="card" href="agentic_structured_extraction.txt"><strong>Structured JSON extraction</strong><span>Calendar + roster from prose</span><div class="meta">thinking: 2,501 tok · 15 s · clean pass</div></a> |
| <a class="card" href="agentic_structured_extraction_nothink.txt"><strong>JSON extraction · no-think</strong><span>Same prompt, thinking off</span><div class="meta">352 tok · 2 s</div></a> |
| <a class="card" href="agentic_tool_use_json.txt"><strong>Tool-use planning</strong><span>Weather + flights + hotel</span><div class="meta">thinking: 1,381 tok · 8 s</div></a> |
| </div> |
|
|
| <footer> |
| Model: <code>Jackrong/Qwopus3.6-35B-A3B-v1-GGUF — Q5_K_M</code> · 35B total / 3B active · served via llama.cpp on a single RTX 5090 |
| </footer> |
| </div> |
| </body> |
| </html> |
|
|
