Unverified50% confidenceEventExact time
In three rounds of controlled experiments on a 32-expert tiny character model, Expert Choice routing achieved the lowest evaluation loss among the three rounds.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Understanding MoE Routing: Three Core Problems and Two Fixes
redditr/deeplearning10/15/2026
Related Entities
Related Claims
UnverifiedIn an experiment with a 32-expert miniature character model, when Token Choice routing had no load balancing loss, a single expert swallowed 57% of the tokens in the early stage of training, and 17 of the 32 experts were almost ignored.80% similarUnverifiedIn the most difficult single-concept quantitative category, most models' accuracy falls below 30%70% similarUnverifiedIn the article, under DeepEval's default mode, 7 responses had their pass/fail conclusion flip at least once across three re-runs; after manually fixing evaluation_steps, the number of flips dropped to 3, and the agreement rate rose from 31/40 to 35/40.70% similarUnverified智能筛选回归用例Agent能将回归测试的用例数量压缩到全量的20%-30%，同时保持95%以上的缺陷发现率69% similarUnverifiedIn earlier tests, a rewrite that raised a tool's score from 31 to 74 instead made both models' selection performance worse.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1035490API
curl https://kongchang.com/api/v1/knowledge/claims/1035490MCP
get_claim(id=1035490)