Unverified50% confidenceFactExact time
The perplexity of the florid-style prompt was about twice that of the normal style (Qwen3 0.6B: 37.2 vs 16.7), but the perplexity of the plain style and the normal style were almost the same (15.6 vs 16.7), yet the small model performed worse on the plain style.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedOn models of 4B and below, the normal-style prompt scored highest, with the plain and florid styles both dropping about 8 points, but all confidence intervals crossed zero, so the differences cannot be strictly proven significant.73% similarUnverifiedQwen3 4B-instruct scored 93% on math tasks under the normal-style prompt, dropping sharply to 43% when switched to the florid style.65% similarUnverifiedPerplexity's decision model has 27B parameters and is fine-tuned from Qwen3.8-27B.60% similarUnverifiedSwitching the attention kernel and inference framework vLLM produces different outputs on the full-precision 16-bit QWEN 3 30B, with differences stemming from the mathematical computation path itself.60% similarUnverifiedOn the Qwen-3.5-4B-Instruct model, the top two principal axes matched the style dimensions of human spontaneous requests with a precision of 72.8% and a macro-averaged recall of 43.6%.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1018312API
curl https://kongchang.com/api/v1/knowledge/claims/1018312MCP
get_claim(id=1018312)