RecaLLM interleaves explicit in-context retrieval with reasoning to mitigate the lost-in-thought degradation in long-context LLMs, achieving gains on RULER and HELMET benchmarks with short training contexts.
The reasoning trace should use the recall tool frequently, but only to recall key information from the context
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
RecaLLM: Addressing the Lost-in-Thought Phenomenon with Explicit In-Context Retrieval
RecaLLM interleaves explicit in-context retrieval with reasoning to mitigate the lost-in-thought degradation in long-context LLMs, achieving gains on RULER and HELMET benchmarks with short training contexts.