LFM2.5 Pi LoRA goes GGUF: run the RL-trained agent adapter in llama.cpp
yb2698 · x · 2026-10-30
Author released harness-hub/LFM2.5-24B-A2B-pi-GGUF — the GGUF conversion (just 37MB F16) of the LoRA adapter RL-trained inside the Pi harness, stackable on LiquidAI's official base GGUF in llama.cpp. It works like a charm in hands-on testing.
Key deployment steps from the model card:
- Download the base Q80 and adapter F16 GGUFs — the closer the base quant is to the BF16 training weights, the closer you get to the trained behavior.
- Launch llama-server with --jinja (enables tool calling, which Pi depends on) and mount the adapter via --lora, context set to 65536.
- Wire the local server into /.pi/agent/models.json, and api must be openai-completions — otherwise pi uses the Responses API the adapter was never trained with.
On tasks both solve, the adapter matches base accuracy while cutting tool calls by 39%.
Related event: LFM2.5 Pi LoRA Adapter Released in GGUF for Local llama.cpp Use(2 posts)→
More from coding & agent
- A 4B Model Verifies a Coding Agent 70x Its Size at Just 4% of Steps, 9-13% Extra Cost — petrusenko_max · 2026-10-30
- Static RAG eval sets go stale: treat your eval dataset as a pipeline, not a file — paulabartabajo_ · 2026-10-30
- What took a week now takes ~5 minutes with Sol or Opus coding agents — D3VAUX · 2026-10-30
- dhh: 'Review every line' is AI psychosis — the future is verifying outcomes, not code — CtrlAltDwayne · 2026-10-30
- Do you need an AI QA agent writing Playwright tests when AI already reviews PRs? — Infinite-Fly-2756 · 2026-10-30
- Octoverse: Claude Code tops fastest-growing repos; helium lands at #9 — uwukko · 2026-10-30