| <!doctype html> |
| <html lang="en"> |
| <head> |
| <meta charset="utf-8" /> |
| <meta name="viewport" content="width=device-width, initial-scale=1" /> |
| <title>dcoach — prove a CPU model recalls its training · mindXtrain</title> |
| <link rel="stylesheet" href="/coach/static/style.css" /> |
| <style> |
| body { max-width: 1040px; margin: 0 auto; padding: 1.5rem; } |
| .dc-grid { display: grid; grid-template-columns: 1fr; gap: 1rem; } |
| .dc-card { border: 1px solid var(--border); border-radius: 8px; padding: 1rem; |
| background: var(--card); margin-bottom: 1rem; } |
| .dc-card > h3 { margin: 0 0 0.6rem; font-size: 1rem; } |
| .dc-row { display: flex; flex-direction: column; gap: 0.25rem; margin-bottom: 0.7rem; } |
| .dc-row label { font-size: 0.82rem; color: var(--muted); } |
| .dc-row select, .dc-row input[type=text], .dc-row input[type=number] { |
| background: var(--bg); color: var(--fg); border: 1px solid var(--border); |
| border-radius: 6px; padding: 0.4rem 0.6rem; font-family: inherit; } |
| .dc-skills { display: grid; grid-template-columns: repeat(auto-fit, minmax(180px, 1fr)); gap: 0.3rem; } |
| .dc-skills label { display: flex; align-items: center; gap: 0.4rem; font-size: 0.85rem; color: var(--fg); } |
| .dc-run { background: var(--accent); color: #111; border: none; border-radius: 6px; |
| padding: 0.6rem 1.2rem; font-weight: 600; font-size: 0.95rem; cursor: pointer; } |
| .dc-run:disabled { opacity: 0.5; cursor: progress; } |
| .dc-phases { font-size: 0.8rem; max-height: 240px; overflow-y: auto; } |
| .dc-phase { display: flex; gap: 0.5rem; padding: 0.15rem 0; border-bottom: 1px solid var(--border); } |
| .dc-phase .tag { color: var(--accent); min-width: 84px; font-weight: 600; } |
| .dc-verdict { display: grid; grid-template-columns: repeat(auto-fit, minmax(150px, 1fr)); gap: 0.6rem; } |
| .dc-metric { border: 1px solid var(--border); border-radius: 6px; padding: 0.6rem; text-align: center; } |
| .dc-metric .v { font-size: 1.4rem; font-weight: 700; } |
| .dc-metric .k { font-size: 0.72rem; color: var(--muted); text-transform: uppercase; letter-spacing: 0.03em; } |
| .pass { color: var(--ok); } .fail { color: var(--bad); } |
| .dc-net { display: grid; grid-template-columns: repeat(auto-fit, minmax(280px, 1fr)); gap: 0.7rem; } |
| .dc-net-card { border: 1px solid var(--border); border-radius: 8px; padding: 0.8rem; background: var(--bg); } |
| .dc-net-card h4 { margin: 0 0 0.4rem; font-size: 0.9rem; color: var(--accent-2); } |
| .dc-net-card p { margin: 0.2rem 0; font-size: 0.78rem; line-height: 1.4; } |
| .dc-net-card .meta { color: var(--muted); } |
| .dc-net-card .fit { color: var(--fg); border-left: 2px solid var(--accent); padding-left: 0.5rem; margin-top: 0.4rem; } |
| table.dc-fit { width: 100%; border-collapse: collapse; font-size: 0.78rem; } |
| table.dc-fit th, table.dc-fit td { border: 1px solid var(--border); padding: 0.4rem 0.5rem; text-align: left; vertical-align: top; } |
| table.dc-fit th { color: var(--accent); } |
| .dc-nav a { color: var(--accent-2); margin-right: 1rem; font-size: 0.85rem; } |
| .lede { color: var(--muted); } |
| .hidden { display: none; } |
| </style> |
| </head> |
| <body> |
| <header> |
| <div class="dc-nav"> |
| <a href="/coach/">← Coach</a> |
| <a href="/coach/prompts">Prompt tools</a> |
| <a href="/coach/modelfile">Modelfile builder</a> |
| </div> |
| <h1>dcoach <span class="muted">— prove the training stuck</span></h1> |
| <p class="lede"> |
| Imprint a persona onto a tiny model on <strong>CPU</strong>, then prove the model |
| <em>recalls</em> it: the <strong>classroom</strong> measures recall before vs after |
| training, the <strong>boardroom</strong> rules success or failure, and the verdict |
| feeds the <strong>autotune feedback loop</strong> that tunes the next run. This is |
| mindXtrain's first-run proof — and the on-ramp to decentralized training. |
| </p> |
| </header> |
|
|
| <main class="dc-grid"> |
| <section class="dc-card"> |
| <h3>1 · Imprint & Prove</h3> |
| <div class="dc-row"> |
| <label for="dc-persona">Persona (the voice to imprint)</label> |
| <select id="dc-persona"></select> |
| </div> |
| <div class="dc-row"> |
| <label>Skills (extra demonstrations, optional)</label> |
| <div id="dc-skills" class="dc-skills"></div> |
| </div> |
|
|
| <details> |
| <summary class="muted" style="cursor:pointer;font-size:0.85rem;">Advanced parameters</summary> |
| <div style="margin-top:0.6rem;"> |
| <div class="dc-row"> |
| <label for="dc-base">Base model (the actor)</label> |
| <input id="dc-base" type="text" value="HuggingFaceTB/SmolLM2-135M" /> |
| </div> |
| <div class="dc-row"> |
| <label for="dc-board">Boardroom preset</label> |
| <select id="dc-board"> |
| <option value="classic_triad">classic_triad</option> |
| <option value="solo">solo</option> |
| <option value="quintet">quintet</option> |
| </select> |
| </div> |
| <div class="dc-row"> |
| <label for="dc-board-model">Boardroom model (blank = tally classroom verdict)</label> |
| <input id="dc-board-model" type="text" placeholder="e.g. qwen3:0.6b" /> |
| </div> |
| <div class="dc-row"> |
| <label for="dc-tokens">Recall generation length (max_new_tokens)</label> |
| <input id="dc-tokens" type="number" value="48" min="8" max="256" /> |
| </div> |
| </div> |
| </details> |
|
|
| <button id="dc-run" class="dc-run">Imprint & Prove</button> |
| <span id="dc-status" class="muted" style="margin-left:0.8rem;font-size:0.85rem;"></span> |
| </section> |
|
|
| <section id="dc-live" class="dc-card hidden"> |
| <h3>2 · Live proof loop</h3> |
| <div id="dc-phases" class="dc-phases"></div> |
| </section> |
|
|
| <section id="dc-result" class="dc-card hidden"> |
| <h3>3 · Verdict</h3> |
| <div class="dc-verdict"> |
| <div class="dc-metric"><div class="v" id="r-before">–</div><div class="k">recall before</div></div> |
| <div class="dc-metric"><div class="v" id="r-after">–</div><div class="k">recall after</div></div> |
| <div class="dc-metric"><div class="v" id="r-delta">–</div><div class="k">imprint Δ</div></div> |
| <div class="dc-metric"><div class="v" id="r-persona">–</div><div class="k">persona kept</div></div> |
| <div class="dc-metric"><div class="v" id="r-board">–</div><div class="k">boardroom</div></div> |
| <div class="dc-metric"><div class="v" id="r-passed">–</div><div class="k">proof</div></div> |
| </div> |
| <p id="r-rationale" class="lede" style="margin-top:0.7rem;"></p> |
| <p id="r-next" class="muted" style="font-size:0.82rem;"></p> |
| </section> |
|
|
| <section class="dc-card"> |
| <h3>How mindXtrain fits decentralized training (2026)</h3> |
| <p id="dc-thesis" class="lede"></p> |
| <table class="dc-fit"> |
| <thead><tr><th>Primitive</th><th>mindXtrain</th><th>Maps to</th></tr></thead> |
| <tbody id="dc-fit-body"></tbody> |
| </table> |
| <h4 style="margin:1rem 0 0.5rem;font-size:0.9rem;">The networks</h4> |
| <div id="dc-net" class="dc-net"></div> |
| <p class="muted" style="font-size:0.75rem;margin-top:0.7rem;"> |
| Read-only. mindXtrain does not mine on these networks — every one is CUDA-first / |
| hardware-gated. It exposes a verifiable, payable training surface compatible with |
| their verification primitives. Source: <code>docs/decentralized-training-deep-dive-2026.md</code>. |
| </p> |
| </section> |
| </main> |
|
|
| <script src="/coach/static/dcoach.js"></script> |
| </body> |
| </html> |
|
|
