Agentic Inference Scaffolding

The paper's inference framework that reformulates multi-turn tool-calling as single-turn contextual QA for Qwen models and implements context memory management.

Neighborhood — ranked by edge-count

Frameworks (1)

framework

SFR-DeepResearch
implements
The paper's core contribution: an RL-based framework for training autonomous single-agent LLMs to perform deep research with web search, browsing, and code execution.

Findings (1)

finding

Single-turn agentic workflow yields 10% absolute improvement on FRAMES for QwQ-32B over default multi-turn template.
supports
Result demonstrating inference-time architectural gains from reformulating multi-turn interactions as single-turn contextual QA.

Methods (1)

method

Memory Management System
implements
Enables agents to self-manage internal context window by providing a clean_memory tool that selectively preserves important information when approaching token limits.

Hypotheses (1)

hypothesis

QwQ and Qwen models have been extensively post-trained to excel at single-step tasks, causing degradation in long multi-turn interactions.
gates
Proposed explanation for why single-turn reformulation improves performance: models' training distribution is concentrated on single-turn reasoning.

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

agentic reasoningconcept0.769
Reasoning approach using code or tool calls executed by an agent.
Ecological Scaffoldingframework0.758
Theory proposing extrinsic ecological conditions temporarily create grouped populations supporting higher-level selection; later replaced by endogenous adaptations.
Agentic Visual Reasoningconcept0.757
Paradigm where VLM acts as controller generating code or tool calls to external modules for visual operations, incurring context-switching latency.
Cognitive Scaffoldingconcept0.754
The hypothesis that tools extend and structure human cognitive processes.
mirroring / scaffoldingmethod0.739
Method of cultivating introspective behavior by mirroring back a model's self-discoveries, creating feedback loops via ICL.
inference of sentienceconcept0.738
Attributing subjective experience based on observable embodied behaviours.
agentic methodsconcept0.731
Methods that use agentic reasoning; incur context-switching latency from external execution.
Embodied Inferenceconcept0.729
Process by which organism's material states and internal dynamics realize variational inference through action