Unverified50% confidenceFactExact time
In practice, LoRA often chooses a rank between 4 and 64, together with a scaling factor alpha to control the magnitude of the update.
1
Sources
50%
Confidence
Long-term
Relevance
11/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWhen the rank r is between 4 and 64, the number of trainable parameters in LoRA is usually on the order of one-thousandth of the original parameters.66% similarUnverifiedThe author's fine-tuning used the QLoRA method, with LoRA rank/alpha of 32/64, a learning rate of 1e-4, batch size 1 with gradient accumulation of 16, AdamW 8-bit optimizer, and training for 1 epoch.62% similarUnverifiedINT4 quantization compresses weights that were originally represented by 65,536 distinct values (FP16) down to only 16 discrete levels.61% similarUnverifiedIn Artificial Analysis's AA-Omniscience benchmark (which penalizes incorrect answers), the top score rose from 4.8 at launch to 46 currently.60% similarUnverifiedState-of-the-art supervised models on the LIAR benchmark still hover around 28%–32% six-class accuracy.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1131604API
curl https://kongchang.com/api/v1/knowledge/claims/1131604MCP
get_claim(id=1131604)