Unverified50% confidenceFactExact time
Accordion-6B-A0.5B was pre-trained on 3.6T tokens.
1
Sources
50%
Confidence
Long-term
Relevance
10/26/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAccording to the Chinchilla scaling law, the optimal training data volume for a 30B model is approximately 600B–1T tokens.66% similarUnverified千问3.8 27B基于千问3.5的整套架构体系，架构层面与3.5完全一致，仅做了继续预训练和后训练（含SFT与RL强化）65% similarUnverifiedThe LLaMA series adopts an over-training strategy, training a 7B-parameter model on 1T tokens, far exceeding the Chinchilla-optimal ratio of 140B tokens62% similarUnverifiedGPT-3 has 175B parameters and was trained on 300B tokens; Gopher has 280B parameters and was trained on 300B tokens — both are in an over-parameterized and under-trained state.61% similarUnverifiedCompared with Qwen3-30B-A3B, Hy-Memory-30B-A3B improves by an average of 11 points, and still scores 56.89 on the EverMemOS framework, which it never saw in training.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1098761API
curl https://kongchang.com/api/v1/knowledge/claims/1098761MCP
get_claim(id=1098761)