llama.cpp on a jailbroken PS5: run any GGUF model natively on the console, starting with Qwen 3.8
Chat: tell the model what it is and where it runs
Mert Cobanova7debaf060f8
diff --git a/ps5/webui/index.html b/ps5/webui/index.htmlindex 2fa26a1..5b52cdf 100644--- a/ps5/webui/index.html+++ b/ps5/webui/index.html@@ -95,8 +95,17 @@<script>(function () {- var SYSTEM = 'You are a helpful assistant running on a PlayStation 5. ' +- 'Answer in one to three short sentences, in the language the user writes in.';+ // What the model knows about itself. The name fills in from /v1/models.+ var modelName = 'a language model';+ // Small models copy an example answer far better than they follow a rule.+ function systemPrompt() {+ return 'You are ' + modelName + ', an open-weight language model running locally on this ' ++ 'PlayStation 5 with PS5LM, a port of llama.cpp. You run on the console\'s own CPU, offline, and ' ++ 'the conversation stays on this console. If someone asks who you are or where you run, answer in ' ++ 'their language along the lines of: "I\'m ' + modelName + ', running right here on this PS5\'s own ' ++ 'processor with PS5LM." Keep answers to one to three short sentences unless asked for more, and ' ++ 'always reply in the language the user writes in.';+ }var log = document.getElementById('log');var scroller = document.getElementById('scroll');var form = document.getElementById('form');@@ -114,8 +123,9 @@try {var id = JSON.parse(info.responseText).data[0].id;// Qwen3.5-2B-Q4_K_M.gguf reads as Qwen3.5 2B.- document.getElementById('model').textContent = id.split('/').pop().replace(/\.gguf$/, '')+ modelName = id.split('/').pop().replace(/\.gguf$/, '').replace(/-(UD-)?(I?Q[0-9].*|BF16|F16|F32)$/i, '').replace(/-/g, ' ');+ document.getElementById('model').textContent = modelName;} catch (e) {}};info.send();@@ -205,7 +215,7 @@// Small models on a few CPU cores: short answers, no thinking phase.xhr.send(JSON.stringify({- messages: [{ role: 'system', content: SYSTEM }].concat(messages),+ messages: [{ role: 'system', content: systemPrompt() }].concat(messages),stream: true,max_tokens: 256,chat_template_kwargs: { enable_thinking: false }