MCPcopy Create free account
hub / github.com/LLMQuant/quant-mind / PaperFlowCfg

Class PaperFlowCfg

quantmind/configs/paper.py:60–65  ·  view source on GitHub ↗

Knobs specific to paper_flow.

Source from the content-addressed store, hash-verified

58 """Chunking and summarization controls for the semantic paper build.
59
60 Selects the source-first chunk/summary shape when bound to
61 ``PaperFlow``: ``PaperFlow(PaperSemanticCfg(...)).build(input)`` returns a
62 ``PaperSemanticResult``.
63
64 Summarization is a deterministic map-reduce: code tiles the chunk set into
65 ``summary_research_group_size`` groups (so coverage is guaranteed by
66 construction), fans out one research agent per group with
67 ``summary_concurrency`` parallelism, then runs one reducer. Per-agent output
68 is bounded by ``max_summary_output_tokens`` through ``ModelSettings``; there

Calls

no outgoing calls