llama.cpp on a jailbroken PS5: run any GGUF model natively on the console, starting with Qwen 3.8
Site: the app as it is now, with its icon, screens and video, and decision models powered by Ollaya
Mert Cobanovdfe1d07de459
diff --git a/site/assets/style.css b/site/assets/style.cssindex ae63c12..9bce801 100644--- a/site/assets/style.css+++ b/site/assets/style.css@@ -4,9 +4,9 @@* Same family as cobanov.dev and herdrchat.cobanov.dev: the dark ground, the* self-hosted Recursive (mono cut for the wordmark, headings and buttons, sans* cut for reading), hairline rules between sections, near-square corners, the- * 1160px measure and the periwinkle accent. The one thing this page adds is- * the console itself: the demo video in the hero, and the boot chain set as a- * terminal panel, since that is how the project actually runs.+ * 1160px measure and the periwinkle accent. The app's icon and home screen+ * background were drawn from this palette, so the hero carries the same grid+ * and glow, and the screens are the app's own.*/@font-face {@@ -89,6 +89,9 @@ a:focus-visible, .btn:focus-visible, video:focus-visible {}.nav { display: flex; align-items: center; gap: 1.25rem; min-height: 60px; }.brand {+ display: inline-flex;+ align-items: center;+ gap: 0.6rem;font-family: var(--mono);font-weight: 700;letter-spacing: -0.04em;@@ -96,6 +99,8 @@ a:focus-visible, .btn:focus-visible, video:focus-visible {text-decoration: none;font-size: 1.05rem;}+.brand img { border-radius: 6px; }+.brand em, .display span em { font-style: normal; color: var(--accent); }.nav-spacer { flex: 1; }.nav-links { display: flex; gap: 1.15rem; font-size: 0.9rem; }.nav-links a { color: var(--muted); text-decoration: none; }@@ -125,30 +130,59 @@ a:focus-visible, .btn:focus-visible, video:focus-visible {/* ---------- hero ---------- */-.hero { padding: 4.5rem 0 4rem; border-bottom: 1px solid var(--line); }-.hero-grid {- display: grid;- grid-template-columns: minmax(0, 1fr) auto;- gap: 4rem;+/* The console's home screen background: a faint grid and a periwinkle glow. */+.hero {+ padding: 4.5rem 0 4rem;+ border-bottom: 1px solid var(--line);+ background:+ radial-gradient(60% 55% at 80% 30%, rgba(130, 136, 240, 0.16), rgba(130, 136, 240, 0.03) 55%, transparent),+ linear-gradient(90deg, var(--bg) 0, transparent 45%),+ repeating-linear-gradient(0deg, rgba(235, 235, 245, 0.035) 0 1px, transparent 1px 60px),+ repeating-linear-gradient(90deg, rgba(235, 235, 245, 0.035) 0 1px, transparent 1px 60px),+ var(--bg);+}+.pill {+ display: inline-flex;align-items: center;+ gap: 0.6rem;+ padding: 0.3rem 0.8rem 0.3rem 0.35rem;+ border: 1px solid var(--line-strong);+ border-radius: 999px;+ background: var(--bg-elevated);+ color: var(--ink-soft);+ font-size: 0.85rem;+ text-decoration: none;+ margin-bottom: 1.6rem;+}+.pill b {+ font-family: var(--mono);+ font-weight: 600;+ font-size: 0.75rem;+ background: var(--accent);+ color: var(--accent-ink);+ border-radius: 999px;+ padding: 0.1rem 0.55rem;}+.pill:hover { border-color: var(--muted-2); color: var(--ink); }.display {font-family: var(--mono);font-weight: 700;letter-spacing: -0.045em;- line-height: 1.05;- font-size: clamp(2.3rem, 5vw, 3.6rem);+ line-height: 1.08;+ font-size: clamp(2.1rem, 5vw, 3.6rem);margin: 0 0 1.4rem;+ max-width: 22ch;}-.lede { font-size: 1.15rem; color: var(--ink-soft); max-width: 34rem; margin: 0; }-.hero-note { color: var(--muted); font-size: 0.9rem; margin: 1.1rem 0 0; max-width: 34rem; }+.display span { color: var(--accent); }+.lede { font-size: 1.15rem; color: var(--ink-soft); max-width: 40rem; margin: 0; }+.hero-note { color: var(--muted); font-size: 0.9rem; margin: 1.1rem 0 0; max-width: 40rem; }-/* The console, filmed: portrait phone video in a plain frame. */-.demo { margin: 0; width: min(340px, 80vw); }+/* The console, filmed on the TV. */+.demo { margin: 3rem 0 0; }.demo video {display: block;width: 100%;- aspect-ratio: 9 / 16;+ aspect-ratio: 16 / 9;border-radius: var(--radius-lg);border: 1px solid var(--line-strong);background: #000;@@ -202,12 +236,6 @@ code {.panel .dim { color: var(--term-dim); }.panel .you { color: var(--accent); }-/* The boot chain: what runs, in order, on the console. */-.chain { list-style: none; padding: 0; margin: 0; }-.chain li { display: grid; grid-template-columns: 8.5rem 1fr; gap: 0 1rem; }-.chain li + li { margin-top: 0.35rem; }-.chain b { color: var(--accent); font-weight: 600; }-.grid {display: grid;grid-template-columns: repeat(auto-fit, minmax(230px, 1fr));@@ -215,28 +243,53 @@ code {}.stat .value {font-family: var(--mono);- font-weight: 700;- letter-spacing: -0.03em;- font-size: 1.15rem;+ font-weight: 300;+ letter-spacing: -0.05em;+ font-size: clamp(2.6rem, 6vw, 3.6rem);+ line-height: 1;color: var(--ink);- margin: 0 0 0.3rem;+ margin: 0 0 0.6rem;}+.stat .value small { font-size: 0.95rem; font-weight: 500; letter-spacing: -0.02em; color: var(--muted); margin-left: 0.4rem; }.stat p { color: var(--muted); margin: 0; font-size: 0.94rem; }+.note { color: var(--muted); font-size: 0.9rem; margin-top: 2rem; }-ul { padding-left: 1.15rem; max-width: var(--measure); color: var(--ink-soft); }-li { margin: 0.4rem 0; }+.eyebrow { font-family: var(--mono); font-size: 0.85rem; color: var(--muted); margin: 0 0 0.4rem; }-.roadmap { list-style: none; padding: 0; max-width: var(--measure); margin: 0; }-.roadmap li {- display: grid;- grid-template-columns: 4rem 1fr;- gap: 0 1rem;- padding: 0.7rem 0;- border-bottom: 1px solid var(--line);- margin: 0;+/* The app's own screens, two to a row. */+.shots { display: grid; grid-template-columns: repeat(2, minmax(0, 1fr)); gap: 2rem 1.5rem; margin-top: 2rem; }+.shots figure { margin: 0; }+.shots img {+ display: block;+ width: 100%;+ height: auto;+ border-radius: var(--radius-lg);+ border: 1px solid var(--line);}-.roadmap .state { font-family: var(--mono); font-size: 0.85rem; color: var(--muted); }-.roadmap .state.done { color: var(--done); }+.shots figcaption { color: var(--muted); font-size: 0.9rem; margin-top: 0.7rem; }+.shots figcaption b { font-family: var(--mono); color: var(--ink); margin-right: 0.4rem; }++.cards { gap: 1rem; }+.card {+ background: var(--panel);+ border: 1px solid var(--line);+ border-radius: var(--radius-lg);+ padding: 1.1rem 1.2rem;+}+.card p { color: var(--muted); font-size: 0.94rem; margin: 0; }+.card code { font-size: 0.8em; }++pre.panel { white-space: pre; }+pre.panel code { background: none; border: 0; padding: 0; color: inherit; font-size: inherit; }++.table { overflow-x: auto; max-width: var(--measure); margin: 0 0 1.5rem; }+table { border-collapse: collapse; width: 100%; font-size: 0.94rem; }+th, td { text-align: left; padding: 0.55rem 0.9rem 0.55rem 0; border-bottom: 1px solid var(--line); color: var(--ink-soft); vertical-align: top; }+th { font-family: var(--mono); font-weight: 600; font-size: 0.85rem; color: var(--muted); }+td:last-child { color: var(--muted); white-space: nowrap; }++ul { padding-left: 1.15rem; max-width: var(--measure); color: var(--ink-soft); }+li { margin: 0.4rem 0; }.faq h3 { margin-top: 1.6rem; }.faq h3:first-of-type { margin-top: 0; }@@ -249,19 +302,13 @@ li { margin: 0.4rem 0; }.site-footer p { color: var(--muted); max-width: none; }.foot-row { display: flex; flex-wrap: wrap; gap: 1rem; align-items: center; margin-bottom: 0.8rem; }-@media (max-width: 860px) {- .hero-grid { grid-template-columns: 1fr; gap: 2.5rem; }- .demo { justify-self: center; }+@media (max-width: 760px) {+ .shots { grid-template-columns: 1fr; }}@media (max-width: 620px) {.nav-links { display: none; }.hero { padding: 3rem 0 2.5rem; }- .chain li { grid-template-columns: 1fr; }- .chain li + li { margin-top: 0.7rem; }- .roadmap li { grid-template-columns: 1fr; }}-.lessons { margin-top: 2.5rem; }-.steps-list { max-width: var(--measure); color: var(--ink-soft); padding-left: 1.3rem; margin: 0 0 1.25rem; }.steps-list li { margin: 0.5rem 0; }diff --git a/site/index.html b/site/index.htmlindex 91cbbe4..2c2ddec 100644--- a/site/index.html+++ b/site/index.html@@ -3,17 +3,19 @@<head><meta charset="utf-8"><meta name="viewport" content="width=device-width, initial-scale=1">-<title>PS5LM: llama.cpp on a PlayStation 5</title>-<meta name="description" content="PS5LM ports llama.cpp to the PlayStation 5. Open language models run on the console's own CPU, and you chat with them in the PS5's browser.">+<title>PS5LM: a model server on your PS5</title>+<meta name="description" content="PS5LM runs llama.cpp on the PS5's GPU and serves language and decision models to every device on your network, with OpenAI and Anthropic APIs and a screen made for the TV."><meta name="theme-color" content="#0e0e14"><link rel="canonical" href="https://ps5lm.cobanov.dev/">+<link rel="icon" type="image/png" href="/media/favicon.png">+<link rel="apple-touch-icon" href="/media/icon.png"><link rel="preload" as="font" type="font/woff2" href="/assets/fonts/recursive-sans-v1.woff2" crossorigin><link rel="preload" as="font" type="font/woff2" href="/assets/fonts/recursive-mono-v1.woff2" crossorigin>-<link rel="stylesheet" href="/assets/style.css?v=2">+<link rel="stylesheet" href="/assets/style.css?v=3"><meta property="og:type" content="website"><meta property="og:url" content="https://ps5lm.cobanov.dev/">-<meta property="og:title" content="PS5LM: llama.cpp on a PlayStation 5">-<meta property="og:description" content="Open language models running on the PS5's own CPU, with a chat page in the console's browser. Open source, GPL-3.0.">+<meta property="og:title" content="PS5LM: a model server on your PS5">+<meta property="og:description" content="llama.cpp on the PS5's GPU: Qwen 3.5 9B at 55 tokens a second, decision models powered by Ollaya, OpenAI and Anthropic APIs. Open source, GPL-3.0."><meta property="og:image" content="https://ps5lm.cobanov.dev/media/og.jpg"><meta property="og:image:width" content="1200"><meta property="og:image:height" content="630">@@ -22,12 +24,12 @@<body><header class="site-header"><div class="wrap nav">- <a class="brand" href="/">ps5lm</a>+ <a class="brand" href="/"><img src="/media/icon.png" width="28" height="28" alt=""><span>PS5<em>LM</em></span></a><span class="nav-spacer"></span><nav class="nav-links" aria-label="Primary">- <a href="#how">How it works</a>- <a href="#numbers">Numbers</a>- <a href="#roadmap">Roadmap</a>+ <a href="#screens">The app</a>+ <a href="#features">Features</a>+ <a href="#decisions">Decisions</a><a href="#install">Install</a><a href="#faq">FAQ</a></nav>@@ -37,110 +39,123 @@<main><section class="hero">- <div class="wrap hero-grid">- <div>- <h1 class="display">llama.cpp, running on a PlayStation 5.</h1>- <p class="lede">PS5LM runs open language models on the console's own CPU. You chat with them in the PS5's browser, typing with the DualSense.</p>- <div class="cta">- <a class="btn btn-primary" href="https://github.com/cobanov/PS5LM/releases/latest">Download PS5LM</a>- <a class="btn" href="https://github.com/cobanov/PS5LM">Source on GitHub</a>- </div>- <p class="hero-note">One payload, <code>ps5lm.elf</code>: pick a model, it downloads on the console, chat. Open source under GPL-3.0, for a PS5 you own on firmware 13.60 or lower.</p>+ <div class="wrap">+ <a class="pill" href="#decisions"><b>New</b> Decision models, powered by Ollaya</a>+ <h1 class="display">Your PS5 runs the language model.<br><span>Every device on your network uses it.</span></h1>+ <p class="lede">PS5LM loads a GGUF model onto the console's GPU and serves it to your laptop, your phone and your coding tools, with OpenAI and Anthropic APIs and a screen made for the TV. No PC while it runs, no cloud, no account.</p>+ <div class="cta">+ <a class="btn btn-primary" href="https://github.com/cobanov/PS5LM/releases/latest">Get the app</a>+ <a class="btn" href="https://github.com/cobanov/PS5LM">Source on GitHub</a></div>+ <p class="hero-note">Free and open source under GPL-3.0, for a jailbroken PS5 you own on firmware 7.00 to 13.60.</p><figure class="demo">- <video id="demo" src="/media/ps5lm-demo.mp4" poster="/media/poster.jpg" muted loop playsinline preload="metadata"- aria-label="A PS5 on a TV: someone types with a DualSense in the PS5 browser, and Qwen3.5 answers, including a haiku about the PS5."></video>- <figcaption>Qwen3.5 0.8B on a PS5 Slim, answering at 21 tokens a second.</figcaption>+ <video id="demo" src="/media/ps5lm-home.mp4" poster="/media/poster.jpg" muted loop playsinline preload="metadata"+ aria-label="A PS5 on a TV: PS5LM's Home screen shows Qwen 3.5 9B generating at 52 tokens a second while requests from a laptop stream in."></video>+ <figcaption>Qwen 3.5 9B answering requests from a laptop on a PS5 Slim.</figcaption></figure></div></section>- <section id="how">+ <section id="numbers"><div class="wrap">- <h2>How it works</h2>- <p>Nothing here is emulated. The PS5 has an 8-core AMD Zen 2 CPU and runs a FreeBSD-based system, so llama.cpp builds for it almost unchanged: three build flags, one small library shim and one patch. This is what runs on the console, in order:</p>- <div class="panel">- <ol class="chain">- <li><b>relapse</b><span>the browser exploit opens the console (firmware 7.00 to 13.60)</span></li>- <li><b>loader</b><span>a payload loader accepts homebrew programs</span></li>- <li><b>llama-server</b><span>upstream llama.cpp, built with the open ps5-payload-dev SDK</span></li>- <li><b>model</b><span>a GGUF file from Hugging Face, read into memory</span></li>- <li><b>browser</b><span>the PS5's own browser opens the chat page the console serves</span></li>- </ol>+ <h2>Measured on the console</h2>+ <div class="grid stats">+ <div class="stat"><p class="value">207<small>tok/s</small></p><p>Qwen 3.5 0.8B, Q4_K_M</p></div>+ <div class="stat"><p class="value">55<small>tok/s</small></p><p>Qwen 3.5 9B, Q4_K_M, with a 64k context</p></div>+ <div class="stat"><p class="value">20<small>tok/s</small></p><p>Qwen 3.8 27B, UD-Q2_K_XL, the biggest that fits</p></div>+ <div class="stat"><p class="value">0.3<small>s</small></p><p>Three decisions from d1 Omni 600M; a 720p image and seven questions in 2.9 s</p></div></div>- <p>A phone or laptop on the same network can open the same chat page, and the server speaks the OpenAI API, so other apps can use the PS5 as a small model server.</p>+ <p class="note">A PS5 Slim, every layer on its RDNA 2 GPU through llama.cpp's Vulkan backend, with 11.4 GB of the shared memory for the model and its context.</p></div></section>- <section id="numbers">+ <section id="screens"><div class="wrap">- <h2>Measured on the console</h2>- <div class="grid">- <div class="stat">- <p class="value">13 to 21 tok/s</p>- <p>Qwen3.5 0.8B generating text. Qwen3.5 2B does about 9.</p>- </div>- <div class="stat">- <p class="value">5 of 16 cores</p>- <p>The CPU threads a homebrew program may use. System services share them too.</p>- </div>- <div class="stat">- <p class="value">about 6 GB</p>- <p>The memory one homebrew program can hold, from the pool the home screen also uses.</p>- </div>- <div class="stat">- <p class="value">12 MB/s</p>- <p>Downloading a model from Hugging Face on the console, over 16 connections.</p>- </div>+ <h2>Made for the TV</h2>+ <p>Five pages you move between with L1 and R1. Everything is in reach of the DualSense, and every request another device makes shows up on Home as it happens.</p>+ <div class="shots">+ <figure><img src="/media/app/home.jpg" width="1600" height="900" loading="lazy" alt="Home: the model, its speed, and the requests as a chat"><figcaption><b>Home</b> The model, its speed, and every request and its answer as a chat. Square sends a demo, Triangle opens a keyboard.</figcaption></figure>+ <figure><img src="/media/app/models.jpg" width="1600" height="900" loading="lazy" alt="Models: suggested models measured on the console"><figcaption><b>Models</b> What is on this PS5, models measured on the console, Hugging Face search and downloads.</figcaption></figure>+ <figure><img src="/media/app/home-decision.jpg" width="1600" height="900" loading="lazy" alt="Home with a decision model loaded"><figcaption><b>Decisions</b> With a decision model, Home shows the time per decision and each answer.</figcaption></figure>+ <figure><img src="/media/app/console.jpg" width="1600" height="900" loading="lazy" alt="Console: temperature, power, GPU and CPU charts"><figcaption><b>Console</b> Temperature, power, GPU and CPU over the last two minutes or the last hour.</figcaption></figure>+ <figure><img src="/media/app/api.jpg" width="1600" height="900" loading="lazy" alt="API: the endpoints and an example"><figcaption><b>API</b> The address, the endpoints and a request to copy.</figcaption></figure>+ <figure><img src="/media/app/settings.jpg" width="1600" height="900" loading="lazy" alt="Settings with a list of choices under a row"><figcaption><b>Settings</b> Startup, context and cache, storage, and the logs under Diagnostics.</figcaption></figure></div>- <h3 class="lessons">What the console taught us</h3>- <ul>- <li>llama.cpp starts one busy-waiting thread per core. On a PS5 that takes every core homebrew gets, and the console shuts itself down. The PS5 build leaves two cores free.</li>- <li>llama.cpp's own web interface renders blank in the PS5 browser, so PS5LM ships a plain chat page.</li>- <li>While the browser is in front, the system pages a background program's memory out to disk, and a model then crawls. Locking it in memory took Qwen3.5 2B from 0.9 to 9 tokens a second; locking 2.7 GB froze the console, so the limit is 1.6 GB for now.</li>- <li>The kernel caps a connection's receive buffer at 64 KB, so one download stream to the CDN stops at 0.9 MB/s. Sixteen in parallel get about 12 MB/s.</li>- </ul>- <p>The full notes are in <a href="https://github.com/cobanov/PS5LM/blob/main/docs/CONSOLE.md">docs/CONSOLE.md</a>.</p></div></section>- <section id="roadmap">+ <section id="features"><div class="wrap">- <h2>Roadmap</h2>- <ul class="roadmap">- <li><span class="state done">done</span><span>llama.cpp builds for the PS5 and runs small models on its CPU</span></li>- <li><span class="state done">done</span><span>Chat in the PS5's own browser, with the DualSense and the on-screen keyboard</span></li>- <li><span class="state done">done</span><span>The PS5LM app (v0.1): one payload, a model library in the PS5 browser, downloads on the console, chat</span></li>- <li><span class="state">next</span><span>Listed in Payload Manager, and more models measured on more consoles</span></li>- <li><span class="state">later</span><span>Bigger models, Qwen 3.8 27B first, in a native app that gets more memory than a payload</span></li>- <li><span class="state">later</span><span>The GPU, through llama.cpp's Vulkan backend on a community Vulkan driver for the PS5</span></li>- </ul>+ <h2>A model server under your TV</h2>+ <div class="grid cards">+ <div class="card"><h3>It runs on the GPU</h3><p>llama.cpp's Vulkan backend on the console's RDNA 2 GPU. Every architecture and quantization llama.cpp knows comes along.</p></div>+ <div class="card"><h3>It answers your network</h3><p>OpenAI's <code>/v1/chat/completions</code>, Anthropic's <code>/v1/messages</code> (Claude Code works against it) and a chat page in any browser, at <code>http://<console>:8081</code>.</p></div>+ <div class="card"><h3>It decides, too</h3><p>Load a decision model and the console answers <code>/v1/systemone</code>: choice, score and yes-or-no questions about text and images.</p></div>+ <div class="card"><h3>It shows its work</h3><p>Every request streams across the TV as a chat: which device asked, what it asked, and the reply as it is written.</p></div>+ <div class="card"><h3>It watches the console</h3><p>Charts of temperature, power and load, beside a table of every reading the console gives.</p></div>+ <div class="card"><h3>It picks what fits</h3><p>The planner reads each GGUF and gives it the longest context and best cache that fit. Downloads go to internal storage, the M.2 drive or USB.</p></div>+ <div class="card"><h3>It reads images</h3><p>Models with an image projector, such as Gemma 4, MiniCPM-V and d1, take pictures in chat and in decisions.</p></div>+ <div class="card"><h3>It needs nothing else</h3><p>No PC while it runs, no cloud and no account. Your prompts never leave your home.</p></div>+ </div>+ </div>+ </section>++ <section id="decisions">+ <div class="wrap">+ <p class="eyebrow">Powered by <a href="https://ollaya.dev">Ollaya</a></p>+ <h2>Decision models</h2>+ <p><a href="https://github.com/ollaya-dev/ollaya">Ollaya</a> runs open decision models on a PC the way Ollama runs language models. PS5LM brings the same idea to the console: a decision model reads a state and answers typed questions in one forward pass, with calibrated probabilities and no text generated.</p>+ <pre class="panel"><code><span class="dim">$</span> curl http://<console>:8081/v1/systemone -H 'Content-Type: application/json' -d '{+ "state": "I was charged twice this month. Refund me or I cancel.",+ "questions": {+ "intent": {"type": "choice", "instructions": "What does the customer want?",+ "criteria": {"refund": null, "information": null, "cancel": null}},+ "angry": {"type": "noul", "instructions": "Is the customer angry?"}+ }+}'</code></pre>+ <p>The Suggested list has d1 Omni 600M, d1 3B and Clef Flash. Any GGUF that carries llama.cpp's decision metadata loads the same way, and the app labels it a decision model.</p></div></section><section id="install"><div class="wrap"><h2>Install</h2>+ <p>You need a PS5 on firmware 7.00 to 13.60, jailbroken, with these payloads loaded:</p>+ <div class="table">+ <table>+ <thead><tr><th>Payload</th><th>For</th><th></th></tr></thead>+ <tbody>+ <tr><td><a href="https://github.com/EchoStretch/kstuff-lite">kstuff-lite</a></td><td>the kernel patches homebrew runs on</td><td>required</td></tr>+ <tr><td><a href="https://github.com/etaHEN/etaHEN">etaHEN</a></td><td>homebrew support</td><td>required</td></tr>+ <tr><td><a href="https://github.com/drakmor/ShadowMountPlus">ShadowMountPlus</a></td><td>puts PS5LM on the home screen from <code>/data/homebrew</code></td><td>required</td></tr>+ <tr><td><a href="https://github.com/ps5-payload-dev/elfldr">elfldr</a></td><td>temperature, power and real free space</td><td>recommended</td></tr>+ <tr><td><a href="https://github.com/ps5-payload-dev/ftpsrv">ftpsrv</a></td><td>installing the app, copying models from a PC</td><td>recommended</td></tr>+ </tbody>+ </table>+ </div><ol class="steps-list">- <li>Jailbreak a PS5 you own on firmware 7.00 to 13.60, with a payload loader such as Payload Manager.</li>- <li>In Payload Manager, turn on Multiple Payload Sources in Settings and add the source <code>https://ps5lm.cobanov.dev/payloads.json</code>. Install PS5LM from the list and load it. (Or download <code>ps5lm.elf</code> from the <a href="https://github.com/cobanov/PS5LM/releases/latest">latest release</a> and load it like any payload.)</li>- <li>The PS5 browser opens the library. Press Download on a model, then Run, then Open chat.</li>+ <li>Download <code>ps5lm-app-vX.Y.Z.zip</code> from the <a href="https://github.com/cobanov/PS5LM/releases/latest">latest release</a> and unzip it.</li>+ <li>Copy the <code>PPSA99575</code> folder to <code>/data/homebrew</code> over FTP. ShadowMountPlus adds <strong>PS5LM</strong> to the home screen.</li>+ <li>Open PS5LM, go to <strong>Models > Suggested</strong> and get a model. Or copy a <code>.gguf</code> to <code>/data/PS5LM/models</code>, or to <code>PS5LM/models</code> on a USB drive.</li>+ <li>Load it. Once Home says Ready, open <code>http://<console>:8081</code> in a browser, or point your OpenAI or Anthropic client at it.</li></ol>- <p>From a phone or computer on the same network, the library is at <code>http://<console IP>:8082</code> and the chat at port 8081, which also speaks the OpenAI API. Building from source is in the <a href="https://github.com/cobanov/PS5LM#readme">README</a>.</p>+ <p>To update, copy the new folder over the old one; settings, models and downloads stay in <code>/data/PS5LM</code>. Using it from coding tools (opencode, Codex, Claude Code) is in <a href="https://github.com/cobanov/PS5LM/blob/main/docs/CLI.md">docs/CLI.md</a>, building it in <a href="https://github.com/cobanov/PS5LM/blob/main/docs/BUILDING.md">docs/BUILDING.md</a>.</p></div></section><section id="faq" class="faq"><div class="wrap"><h2>Questions</h2>- <h3>Does this run games or need any Sony code?</h3>- <p>No. PS5LM runs open-weight language models and nothing else. No games, keys, firmware or Sony files are part of it.</p>- <h3>Why the CPU and not the GPU?</h3>- <p>The CPU path needs nothing but the open payload SDK, so it came first. The GPU is the next big step, through a community Vulkan driver for the PS5.</p>+ <h3>Does it run on the GPU?</h3>+ <p>Yes, every layer. llama.cpp's Vulkan backend runs on RADV, the open Mesa driver, ported to the PS5 by PS5_Vulkan. Generation is bound by memory bandwidth, and the 9B model reads its weights at about 70% of the GPU's 448 GB/s.</p><h3>How big a model can it run?</h3>- <p>Up to 1.6 GB of model for now, which is Qwen3.5 2B at 4-bit. The model has to stay locked in memory to run at full speed, and that memory is shared with the home screen. Bigger models need a native app, which is on the roadmap.</p>+ <p>About 11 GB of model and context. Qwen 3.8 27B at 2-bit fits with a 64k context; a 9B at 4-bit fits with room to spare. The planner tells you what fits before you download.</p>+ <h3>Does this run games or need any Sony code?</h3>+ <p>No. PS5LM runs open-weight models and nothing else. No games, keys, firmware or Sony SDK files are part of it.</p><h3>Can it break my console?</h3>- <p>It does not write to system files. Pushing the CPU or memory too hard can freeze the console until you hold the power button; the defaults are set to avoid that.</p>+ <p>It writes only to its own folders, never to system files. Under a long load the console runs warm and the fan spins up, as it does in a game; the Console page shows both.</p>+ <h3>Is it the same as PS5LM Plus?</h3>+ <p>PS5LM grew from PS5LM Plus, the native GPU app by abhi245y, and now lives under its own title ID, PPSA99575, so the two install side by side.</p></div></section></main>@@ -148,10 +163,12 @@<footer class="site-footer"><div class="wrap"><div class="foot-row">- <span>ps5lm, by <a href="https://cobanov.dev">Mert Cobanov</a></span>+ <span>PS5LM, by <a href="https://cobanov.dev">Mert Cobanov</a></span><a href="https://github.com/cobanov/PS5LM">GitHub</a>+ <a href="https://ollaya.dev">Ollaya</a><a href="https://github.com/ggml-org/llama.cpp">llama.cpp</a>- <a href="https://github.com/ps5-payload-dev">ps5-payload-dev</a>+ <a href="https://github.com/mihawk-99/PS5_Vulkan">PS5_Vulkan</a>+ <a href="https://github.com/blackbearreloaded/ps5-homebrew-ui">ps5-homebrew-ui</a></div><p>PS5LM is not affiliated with Sony Interactive Entertainment. PlayStation and PS5 are its trademarks.</p></div>diff --git a/site/media/app/api.jpg b/site/media/app/api.jpgnew file mode 100644index 0000000..8abeea9Binary files /dev/null and b/site/media/app/api.jpg differdiff --git a/site/media/app/console.jpg b/site/media/app/console.jpgnew file mode 100644index 0000000..00b598eBinary files /dev/null and b/site/media/app/console.jpg differdiff --git a/site/media/app/home-decision.jpg b/site/media/app/home-decision.jpgnew file mode 100644index 0000000..fe03feeBinary files /dev/null and b/site/media/app/home-decision.jpg differdiff --git a/site/media/app/home.jpg b/site/media/app/home.jpgnew file mode 100644index 0000000..912b1eeBinary files /dev/null and b/site/media/app/home.jpg differdiff --git a/site/media/app/models.jpg b/site/media/app/models.jpgnew file mode 100644index 0000000..2e8b5f7Binary files /dev/null and b/site/media/app/models.jpg differdiff --git a/site/media/app/settings.jpg b/site/media/app/settings.jpgnew file mode 100644index 0000000..7a78dfeBinary files /dev/null and b/site/media/app/settings.jpg differdiff --git a/site/media/favicon.png b/site/media/favicon.pngnew file mode 100644index 0000000..9d5a08dBinary files /dev/null and b/site/media/favicon.png differdiff --git a/site/media/icon.png b/site/media/icon.pngnew file mode 100644index 0000000..23d6e4bBinary files /dev/null and b/site/media/icon.png differdiff --git a/site/media/og.jpg b/site/media/og.jpgindex 90d6103..4ab1c52 100644Binary files a/site/media/og.jpg and b/site/media/og.jpg differdiff --git a/site/media/poster.jpg b/site/media/poster.jpgindex bd4f828..8258763 100644Binary files a/site/media/poster.jpg and b/site/media/poster.jpg differdiff --git a/site/media/ps5lm-demo.mp4 b/site/media/ps5lm-demo.mp4deleted file mode 100644index 1aa0568..0000000Binary files a/site/media/ps5lm-demo.mp4 and /dev/null differdiff --git a/site/media/ps5lm-home.mp4 b/site/media/ps5lm-home.mp4new file mode 100644index 0000000..09d677cBinary files /dev/null and b/site/media/ps5lm-home.mp4 differ