Verified65% confidenceFactExact time
Training strategies for mathematical LLMs include continued pre-training on large-scale mathematical corpora on top of general pre-training, followed by supervised fine-tuning with chain-of-thought reasoning data, combined with reinforcement learning using process reward models.
3
Sources
65%
Confidence
Long-term
Relevance
7/14/2026
First Seen
Sources
语音驱动几何交互：LLM语义解析与Function Calling架构实践
Related Claims
UnverifiedThe core difference between reasoning models and ordinary LLMs lies in the training approach rather than the architecture; reasoning models introduce a reinforcement learning stage on top of pretraining and supervised fine-tuning.80% similarUnverified在代码数据上训练的LLM不仅提升了编程能力，还显著增强了模型的逻辑推理能力，代码训练和逻辑推理能力之间存在正向迁移效应78% similarUnverifiedThe LLM routing space already has precedents in RouteLLM, Martian, and others; the core idea is to train lightweight classifiers to predict task difficulty and type before inference, then map the results to a model selection strategy.76% similarUnverified数学LLM通常指经过数学推理数据强化训练的模型，如DeepSeek-Math、Qwen-Math，以及经过工具增强的GPT系列76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/509872API
curl https://kongchang.com/api/v1/knowledge/claims/509872MCP
get_claim(id=509872)