AITOPAI热报
PapersSelected

Meta 与 CMU 论文：RL 训练的 Qwen3 8B 工具选择胜过 402B Llama-4-Maverick

rohanpaul_ai@rohanpaul_ai
Post· 2 sources
View the post
Why we picked it

Meta 和 CMU 把 Qwen3 8B 用 RL 训成了工具选择专家，ToolBench 上 41.5% 精确命中率反超 402B 的 Maverick，做大型工具目录 agent 的可以看看。

Meta 和 Carnegie Mellon 的论文用强化学习训练 Qwen3 小模型做工具选择，流程是把请求拆成步骤、每步用关键词搜索、再选出 1 个工具。在 ToolBench 上，训练后的 8B 模型精确选对工具的比例为 41.5%，而 402B 的 Llama-4-Maverick 最优提示设置为 30.6%。不重新训练的情况下，该模型在 4 个新工具基准中的 3 个上超过 Maverick，且换更强检索器后成绩继续提升。训练数据先用教师撰写示例，再用奖励有用搜索步骤和精确选择的 RL。

Original · rohanpaul_ai

New Meta and Carnegie Mellon paper finds that small models trained with RL pick the right tools more often than a prompted 402B Llama-4-Maverick.

Agents can face more than 10,000 tools, and their descriptions cannot all fit in context. Retrievers only rank by relevance, so recent setups have Claude or ChatGPT pick the tools.

They trained Qwen3 models to split a request into steps, search each step with plain keyword search, and pick 1 tool per step. Training used teacher-written examples, then RL that rewards useful search steps and exact picks.

On ToolBench, the trained 8B model picked exactly the right tools 41.5% of the time, versus 30.6% for Maverick's best prompted setup. With no retraining, it beat Maverick on 3 of 4 new tool benchmarks and improved when given a stronger retriever.

If your agent chooses from a large tool catalog, an RL-trained small model can handle selection instead of a prompted large one.

#Qwen3#Llama-4-Maverick#ToolBench#强化学习#智能体

Keep reading

Recent stories with the same tag or section
PapersBuildersSelected22:35
Meta trains 8B Qwen3 model to beat larger LLMs at tool selection

Small models trained specifically for tool selection can outperform much larger prompted LLMs on the routing step, so agent builders can cut cost and latency by replacing the proprietary LLM in their tool-selection pipeline. The selector's ability to work with different retrievers and unseen domains makes it practical to slot into existing RAG-style stacks. Teams building large-scale agent tooling should compare this approach before defaulting to a frontier model for routing.

#Qwen3#Meta#ToolBench#agents
Event
DAIR.AI@dair_ai2 sourcesSource
PapersBuildersSelected13:58
Meta paper: RSD lets agents learn from unlabeled runs

RSD turns the huge pool of unlabeled agent runs, failures included, into usable training signal without any reward modelling or human annotation. Teams running their own agents can retain completed trajectories and retrain on hindsight reviews, refreshing the reviews each round to avoid the 59.7% versus 67.9% gap seen with stale notes. The comparison against reward-based methods on ALFWorld makes it a practical reference point for anyone choosing a training recipe.

#Meta#Qwen3#RSD#agents
rohanpaul_ai@rohanpaul_aiSource
ProductsBuildersSelected02:35
Woodcock speeds up Qwen3-1.7B nearly 3x on M4 Max

A permissively licensed drop-in alternative to Eagle 4 matters because Eagle's licensing can block commercial local-inference deployments. Developers running small models on Apple Silicon can benchmark Woodcock against their current speculative decoding setup to see if the 2.5x gain holds on their workloads. It competes directly with Eagle 4 rather than a hosted service.

#Woodcock#Eagle 4#Qwen3#speculative decoding
@atomic_chat_hq@atomic_chat_hqSource
PapersBuildersSelected13:32
Meta and Michigan State paper finds LLM-built reasoning graphs unreliable

The non-obvious point is that most reasoning-graph pipelines assume the LLM's decomposition is correct without ever verifying it, so confidence scores inherit hidden graph errors. Builders of agent workflows and RAG-style decomposition systems should add graph-level validation before trusting computed confidence. The Qwen3-32B versus Qwen3-8B result is a concrete warning that scaling a model does not guarantee better sub-question answering.

#Qwen3#Meta#reasoning models#AI safety
rohanpaul_ai@rohanpaul_aiSource