Unverified50% confidenceFactExact time
The Hugging Face blog release Apertus 1.5 discloses the complete four-stage post-training pipeline: SFT (3.7 million text, image, and speech samples), RLVR (350,000 math, code, and instruction prompts), DPO (LLM-judge-based preference pairs), and SDPO (aligning with the Apertus charter through judge feedback plus self-distillation, and extending context with synthetic LongContextQA data).
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedHugging Face has become the de facto standard platform in the NLP field, providing a unified interface for using tens of thousands of pre-trained models.72% similarUnverifiedHugging Face's TRL library provides standardized implementations of training paradigms such as SFT (Supervised Fine-Tuning), DPO (Direct Preference Optimization), and GRPO.71% similarUnverifiedSmoLLM使用Hugging Face的BPE训练器生成自定义词汇表69% similarUnverifiedMany open-weight models on Hugging Face do not include training code, allowing users to use and fine-tune them but not fully reproduce the training process.68% similarUnverifiedTraining data in the instruction tuning phase contains a large volume of creative writing, business email templates, and role-playing dialogue samples, enabling the model to master various textual formats68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1044465API
curl https://kongchang.com/api/v1/knowledge/claims/1044465MCP
get_claim(id=1044465)