斯坦福大学推出 TRACE：将智能体反复失败转化为合成 RL 训练环境的能力定向训练系统

斯坦福大学研究人员提出 TRACE，一种能力定向的智能体训练系统。

该系统从智能体自身的运行轨迹中诊断出反复出现的失败模式，针对每种能力合成一个可验证的训练环境，为每种能力训练一个 LoRA 适配器，并在专家之间路由模型 token。

TRACE 在 τ2-Bench 上提升了 15.3 个点，在 SWE-bench Verified 上达到 73.2% 的 Pass@1。

Agentic LLMs often fail the same way, again and again. A Stanford research team traced this to missing, reusable capabilities. Their system, TRACE , diagnoses those gaps and trains for them directly.

TRACE stands for T urning R ecurrent A gent failures into C apability-targeted training E nvironments. It was released open-source under an MIT license.

To understand the design, first consider why agents fail. They lack specific skills that tasks demand, like retrieving the right record or verifying a precondition.

Two mainstream fixes spend compute poorly. Direct RL or SFT gives sparse rewards that never say which skill was missing. Broad synthetic data is untargeted, so budget flows to skills the model already has.

However, TRACE observes that failures are not random. A small set of deficits accounts for most failed trajectories. Therefore, each recurring deficit can become its own dense, verifiable training signal.

Given that findings, TRACE runs an automated four-step pipeline. Each step is driven by an LLM agent following a markdown prompt.

The base agent generates rollouts in the target environment. An analysis agent splits them into successful and failed sets. It then labels every trajectory-capability pair as NA , PRESENT , or LACKING .

A capability is retained only when it is contrastive and high-coverage. Specifically, its contrastive gap must clear δ = 0.20 and coverage must clear ρ = 0.10. Consequently, the pipeline keeps skills whose absence concentrates in failures.

Next, a generation agent builds one synthetic environment per retained capability. Each environment isolates a single capability while preserving the target’s tool schemas and format.

Task instances are procedurally generated from random seeds. Because generation and verification are algorithmic, rewards need no human labels or LLM judge.

Then each capability gets one LoRA (Low-Rank Adaptation) adapter, trained on its synthetic environment. The training algorithm is GRPO (Group Relative Policy Optimization). The base model stays frozen throughout.

GRPO groups rollouts by shared seed, so scenarios are identical within a group. Rewards are then normalized within each group to isolate the policy’s contribution.

Finally, TRACE composes the adapters into a Mixture-of-Experts (MoE) model. The backbone and adapters stay frozen, and only lightweight token-level gates are trained.

At inference, each token is routed top-1 to a single capability adapter. This lets the model switch experts mid-trajectory.

TRACE diagnoses the capabilities an agent lacks, builds one verifiable environment per gap, trains a LoRA expert for each, then routes tokens across experts. Step through the pipeline below.

Split rollouts into pass / fail, then keep gaps that separate them.

One seeded, auto-verifiable environment is generated per capability.

Rollouts share a seed; rewards are normalized within the group.

A learned gate routes each token top-1 to a single capability expert.

Targeted training and MoE composition beat prompt optimization and single-adapter baselines.

Built from arXiv:2604.05336：https://arxiv.org/abs/2604.05336 · code：https://github.com/ScalingIntelligence/TRACE. Numbers are from the paper. • Marktechpost

In practice, on τ²-Bench, contrastive analysis recovered four deficits. These were structured data reasoning, multi-step task completion, precondition verification, and tool calling precision.

Notably, this findings stayed stable across ten independent runs. Structured data reasoning alone covered the largest share of failed tasks. Multi-step task completion followed closely behind.

To ground these capabilities, consider three concrete tasks. Each maps to a distinct failure mode that TRACE targets.

With those capabilities trained, TRACE was tested on two backbones and two benchmarks. τ²-Bench measures customer-service pass rate across 50 airline and 114 retail tasks. SWE-bench Verified measures Pass@1 on 500 real GitHub issues.

On Qwen3-30B-A3B, TRACE improved τ²-Bench by +15.3 points and SWE-bench Verified by +15 points Pass@1. It beat the strongest external baselines, GEPA and SWE-RL, by +8.6 and +8.4 points.

Moreover, TRACE is sample-efficient. Using under one-fourth the rollouts, it exceeded the final scores of GRPO and GEPA. Its final accuracy ran +10.4 and +8.6 points higher on τ²-Bench. Additionally, a third benchmark, ToolSandBox, showed the same pattern.

On Qwen3.6-27B, TRACE reached 73.2% Pass@1 on SWE-bench Verified. That 27B open-weight model surpassed GPT-5.2-Codex (72.8%), GLM 5, and Claude 4.5 Sonnet on the public leaderboard.

For AI professionals, the pipeline is benchmark-agnostic and driven by markdown prompts. After environments are generated, each capability adapter is trained against a running vLLM server.

Each LoRA adapter adds ~1.6B trainable parameters, or 5.3% of the backbone. The trained MoE gate adds just 491,760 parameters in total. Default thresholds are ρ = 0.10, δ = 0.20, and 8-of-10 cross-run consistency.

Asif Razzaq is the CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences.

Build an Agentic Event Venue Operator [Full Codes]：https://pxllnk.co/twdn5

Thanks! Our team will contact you soon