Available Engines

Rule-Based

• RTRuleRegex
• RTNewLine

LLM-Based (Boundary Inference)

• RTLLMOffsetBased
• RTLLMSegUnitBased
• RTLLMForcedDecoderBased
• RTLLMSurprisal
• RTLLMEntropy
• RTLLMTopKShift
• RTLLMFlatnessBreak

Discourse / Reasoning Schemas

• RTLLMThoughtAnchor
• RTLLMReasoningFlow
• RTLLMArgument

Topic / Semantic / NLI

• RTBERTopicSegmentation
• RTEmbeddingBasedSemanticShift
• RTEntailmentBasedSegmentation
• RTZeroShotSeqClassification
• RTZeroShotSeqClassificationRF
• RTZeroShotSeqClassificationTA

PRM-Based

• RTPRMBase

---

Engine Configuration

You can override engine parameters at call time:

from rt_seg import RTSeg
from rt_seg import RTRuleRegex
from rt_seg import RTBERTopicSegmentation
from rt_seg import OffsetFusionGraph

segmentor = RTSeg(
    engines=[RTRuleRegex, RTBERTopicSegmentation],
    aligner=OffsetFusionGraph,
    label_fusion_type="concat",  # or "majority"
    seg_base_unit="clause",
)
trace = "..."
offsets, labels = segmentor(
    trace,
    model_name="Qwen/Qwen2.5-7B-Instruct",
    chunk_size=200,
)

---

Table of contents

• Rule-Based Engines
• Probabilistic Engines
• Downstream LLM Engines
• Semantic Engines