Unverified50% confidenceFactExact time
Hydrogen-1.0 chose post-training from GLM-5.3 rather than training from scratch, following a "base model + domain specialization" route.
1
Sources
50%
Confidence
Long-term
Relevance
11/4/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedHydrogen-1.0's training method combines supervised fine-tuning (SFT) with reinforcement learning (RL), with data sources including domain data, scientist feedback, in-house laboratory records and discovery trajectories.72% similarUnverifiedHydrogen-1.0's training data includes domain data, laboratory records and agent discovery trajectories.69% similarUnverifiedZhipu did not retrain the model base, but retained GLM-5.2's base and only expanded and reinforced it during the post-training phase.64% similarUnverified对于从零开始训练模型(而非仅部署已有模型)的场景,TensorFlow C++不是理想的入门选择62% similarVerifiedThe GLM series originally adopted an autoregressive blank-filling pre-training framework that accommodates both understanding and generation capabilities.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1134465API
curl https://kongchang.com/api/v1/knowledge/claims/1134465MCP
get_claim(id=1134465)