https://blog.lukesalamone.com/tags/rlvr/ Recent content in Rlvr on Luke Salamone's Blog Hugo en-us Sun, 11 Oct 2026 18:06:06 -0700 https://blog.lukesalamone.com/posts/chess-reasoning-llm/ Sun, 11 Oct 2026 18:06:06 -0700 https://blog.lukesalamone.com/posts/chess-reasoning-llm/ <p><strong>TLDR: I trained an LLM reasoning agent to play chess, with model sizes 0.8B, 3B and 9B. The agent calls tools which help it keep track of the board state, which enables good performance with a relatively small number of parameters.</strong></p> <p><strong>The demo code is <a href="https://github.com/lukesalamone/chess_reasoner_demo">available on Github</a>.</strong></p> <video width="760" height="430" autoplay muted loop playsinline> <source src="../../video/chess_reasoner_0_8b.mp4" type="video/mp4"> Your browser does not support the video tag. </video> <!-- ## Todo before publishing - publish 0.8B model to huggingface - publish demo code. pulls down 0.8B model, installs depdencies, starts demo server --> <h2 id="background">Background</h2> <p>I became excited about the power of RLVR after the <a href="../../posts/notes-on-deepseek-r1/">Deepseek R1 paper</a> came out. They demonstrated that through the use of &ldquo;thinking&rdquo; tokens, an LLM could solve verifiable tasks like math problems. Additionally, they described the necessity of bootstrapping the thinking process with supervised examples, which prevents the model from generating a nonsense chain of thought.</p>