Unverified50% confidenceFactExact time
When fine-tuning Qwen on a single RTX 3090, the number of correct answers on the dev set rose from 7/28 to 13/28, but the average negative log-likelihood (NLL) worsened from 2.215 to 3.916.
1
Sources
50%
Confidence
Long-term
Relevance
10/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedExperience internalization fine-tuning improved Qwen3-8B-27B's one-shot score from 37.07 to 48.22, while reducing generated tokens from 6.41 million to 570,000 (an 11-fold reduction).76% similarUnverifiedQwen2.5-Max-0902 has a maximum output of 130,000 tokens, with the limit rising to 260,000 tokens in reasoning mode.70% similarUnverifiedOn 300 practice exam papers labeled by Qwen3-32B, fine-tuned v3 of Laya raised the judgment macro-F1 across 11 questions from 0.33 to 0.57 versus the zero-shot version, cut calibration error from 0.15 to 0.07, and reduced danger-level MAE from 2.78 to 1.56.70% similarUnverifiedQwen3 235B Instruct 2507 relied more on price gains (+18.6 points) than the other 7 models but was not judged as chasing momentum.69% similarUnverifiedScores across 19 models ranged from 0.55 to 1.00, with Qwen 3 235B A22B Instruct ranking first at 1.00.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1064010API
curl https://kongchang.com/api/v1/knowledge/claims/1064010MCP
get_claim(id=1064010)