LFM2.5-24B Hermes adapter ships as GGUF: local agents with 20% fewer tool calls
yb2698 · x · 2026-11-02
harness-hub released a GGUF version (37MB F16 LoRA) of its Hermes agent adapter for LiquidAI's LFM2.5-24B-A2B, runnable directly on a llama.cpp server: same accuracy as the base model but 20% fewer tool calls on solvable tasks.
How to run it:
- Download the official base GGUF (Q80 recommended, closest to the BF16 weights used in training) plus the adapter GGUF
- Serve with llama serve --lora ... at -c 65536 or higher; Hermes needs at least 64K context per session, scaled up with parallel slots
- Install NousResearch's hermes-agent and point /.hermes/config.yaml at the local custom provider, then run hermes chat
- The adapter was converted from PEFT via llama.cpp's convertloratogguf.py
A practical recipe for developers running agents locally while cutting tool-call overhead.
Related event: Hermes LoRA Adapter Released for LFM2.5-24B, Cutting Tool Calls 20%(2 posts)→
More from coding & agent
- GitHub Actions has a lesser-known ubuntu-slim runner: 1 CPU, 5GB RAM, cheaper CI — DanielLockyer · 2026-11-02
- Before automating a workflow with AI, ask if anyone understands it end to end — djgoel123 · 2026-11-02
- One stale date silently broke a WhatsApp reply agent: 19 messages unanswered — petrusenko_max · 2026-11-02
- The 10X engineer is now just a SKILLS.md: 6 workflow rules for AI coding agents — mdancho84 · 2026-11-02
- Auditing my MCP credential broker: the worst hole was the config file — Relevant_Mousse_9117 · 2026-11-02
- AWS's open-source Strands Harness claims 45% cheaper than Claude Code, reigniting harness debate — blaizedsouza · 2026-11-02