Unverified50% confidenceFactExact time
On Qwen3, romanization (ITRANS transliteration) reduces Hindi token overhead from 4.44x to 2.06x.
1
Sources
50%
Confidence
Long-term
Relevance
10/19/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedFor the same Hindi sentence using different tokenizers, the token cost multipliers are: Qwen3/Qwen2.5 4.44 times, Sarvam-M 1.79 times, and Gemma-3 1.31 times.64% similarUnverifiedQwen3 4B-instruct scored 93% on math tasks under the normal-style prompt, dropping sharply to 43% when switched to the florid style.60% similarUnverifiedIn the Qwen3-4B comparative experiment, updating the model directly with unfiltered skills without using SGUID filtering decreased performance by 0.3 percentage points on HMMT25.58% similarUnverifiedThe IQ4XS quantized version of Qwen3 27B shows only approximately 6% performance degradation compared to the original full version.57% similarUnverifiedQwen3-1.7B's overall causal faithfulness is only 54.8%, far below that of Qwen3-4B.57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1067577API
curl https://kongchang.com/api/v1/knowledge/claims/1067577MCP
get_claim(id=1067577)