FULL STORY
HF's Harness Adapters: From Live Training to Release
Hugging Face's harness-hub team live-streamed the training of LoRA+RL harness adapters to make smaller models better agents, then officially released the first adapters, including Qwen3.8-27B-pi-adapted, roughly doubling solve rates.
2026-11-06 ~ 2026-11-17 · 2 episodes · 22 posts
Episode 1 · Hugging Face's Harness Adapters: Cheap LoRA+RL Tune Small Models per Agent Framework (2026-11-06, 18 posts)
Hugging Face's harness-hub team (yb2698, ariG23498) released and live-streamed the training of "harness adapters": instead of modifying agent harnesses, a small LoRA adapter is trained for each model×harness pair with verifiable-task RL, executed inside unmodified real coding environments (NousResearch's hermes, pi, opencode). Covered models include LiquidAI's LFM2.5-24B-A2B (MoE, 2B active), LFM3-1.7B and thewebAI's TwIL-8B. Lysandre (Liquid AI) and Aritra (HF) introduced the work, and HF published an accompanying blog.
Confirmed
- Single-GPU LoRA fine-tuning of LFM3-1.7B: after a 20-step smoke run, on 150 held-out tasks (paired vs. baseline, 95% CI) the opencode harness gained 9.4 ± 6.3 points pass@1 (69%→79%)
- Compute-saving setup: one vLLM per model, one trainer per harness, each mounting LoRA adapters on shared weights; the three LFM3-1.7B trainers share one vLLM, and evaluating 1350 agent sessions cost only $5
- Reward design: reward = task verifier score + 0.1 × (1 − calls/15), applied only to correct rollouts; ablations show a 0.3 coefficient hurts accuracy while 0.1 cuts tool calls by 11–17% with no accuracy loss
- Environment synthesis: seeded with public agentic trajectories from the HF Hub, converting only trajectories containing verifiable units of work into tasks, yielding Repair/Build/Extend/QA environments, difficulty-stratified via LFM2.5-24B self-assessment to offset the old environments' data-analysis bias
- Training uses the safeagents anti-cheating scheme with all tool calls proxied and audited to prevent reward hacking; environments and training pipeline are open-sourced alongside the stream
Unconfirmed
- The exploration directions (new RL tasks from trajectories, adding the opencode harness, cross-domain transfer) remain goals of this round; full live-stream results pending
- Hesamation's prediction that 2027 will be the year of extreme model–harness co-design is personal speculation without evidence
Why it matters
- The work shows small models can gain clear improvements on a specific agent framework via extremely cheap ($5-scale) targeted RL + LoRA, offering a reproducible open-source paradigm for model–framework co-optimization; the open-sourced environment synthesis and anti-cheating tooling lower the barrier for community replication
- Livestreamed RL training of LoRA harness adapters for Hermes, pi and opencode, fully open-sourced — yb2698 · 2026-11-06
- Hugging Face livestreams open RL training: trajectory-built envs, safeagents, LoRA evals — yb2698 · 2026-11-06
- Hugging Face synthesizes RL envs from open trajectories: Repair, Build, Extend, QA — yb2698 · 2026-11-06
- Synthesizing RL Envs from Open Trajectories: LFM2.5-24B Self-Scores Task Difficulty — yb2698 · 2026-11-06
- Reward Shaping Ablation: 0.1 Bonus Cuts Tool Calls 11-17% at No Accuracy Cost — yb2698 · 2026-11-06
- Evaluating 1,350 Agent Sessions on One GPU for $5 via Shared LoRA Adapters — yb2698 · 2026-11-06
- 20-Step RL on LFM3-1.7B Lifts opencode Pass@1 by 9.4 Points — yb2698 · 2026-11-06
- HF live-streams agent RL: LoRA adapters trained inside real harnesses like opencode — yb2698 · 2026-11-06
- Team livestreams open RL training of LoRA harness adapters for hermes, pi and opencode — yb2698 · 2026-11-06
- HF team live-trains harness adapters: LoRA + RL to make small models better at agentic tasks — yb2698 · 2026-11-06
- Four research directions in HF's harness adapter training: trajectory tasks, new harnesses, transfer — yb2698 · 2026-11-06
- HF details harness adapters: one LoRA per agent harness, every tool call vetted by safeagents — yb2698 · 2026-11-06
- Harness adapters: per-agent-framework LoRAs cut tool calls, $5 for 1,350 eval sessions — LysandreJik · 2026-11-06
- Hugging Face launches Harness Adapters: per-task LoRAs for coding harnesses — Disastrous-Work-1632 · 2026-11-06
- Hugging Face releases harness adapters: safe RL inside pi, opencode and hermes — yb2698 · 2026-11-06
- Harness adapters training goes live: RL-tuned LoRAs for LFM2.5, LFM3 and Twil-9B — yb2698 · 2026-11-06
- Harness launches live LoRA RL training on LFM2.5, LFM3 and Twil-9B for agent tasks — omarsar0 · 2026-11-07
- Harness adapters: RL-trained LoRAs inside each agent harness beat the base model — yb2698 · 2026-11-07
Episode 2 · Harness Adapters launch: LoRA doubles Qwen3.8-27B coding scores (2026-11-17, 4 posts)
Hugging Face and harness-hub released Harness Adapters, RL-trained LoRA adapters that tune models like Qwen3.8-27B to specific agent harnesses. On the Pi harness, solve rates doubled, with SWE-bench Pro rising from 29.4% to 62.7% and tool calls cut by 55%.
- Pi harness adapter via RL: SWE-bench Pro 29.4%→62.7% with 55% fewer tool calls — yb2698 · 2026-11-17
- harness-hub releases first RL-trained harness adapter for agentic coding, full recipe coming Friday — yb2698 · 2026-11-17
- Harness Adapters: RL-trained LoRA lifts Qwen3.8-27B on Pi from 29% to 63% on SWE-bench Pro — LysandreJik · 2026-11-17
- LoRA adapters tuned per harness double Qwen3.8-27B agent benchmark scores — yb2698 · 2026-11-17