Unverified50% confidenceBenchmarkExact time
VIDRAFT performed LoRA training on standard Transformer blocks across all 32 chips of the Korean NPU server, achieving 0.9999 consistency with standard PyTorch training.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/13/2026
First Seen
Valid until: 10/27/2026
Sources
Related Entities
Related Claims
UnverifiedVIDRAFT's VKORE software acceleration layer, without changing the chip, both increases NPU inference performance several times over and makes Transformer LoRA training run on this kind of inference-specific hardware, with numerical precision aligned with standard PyTorch at 0.9999.72% similarUnverifiedNVIDIA A100、Google TPU v4等专用AI训练芯片针对Transformer的矩阵运算模式进行了专项优化70% similarUnverifiedGShard于2020年首次在Transformer中实现了600B参数的MoE训练66% similarUnverifiedThe Hopper architecture of the H100 introduced a dedicated Transformer Engine supporting FP8 precision computation.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1013619API
curl https://kongchang.com/api/v1/knowledge/claims/1013619MCP
get_claim(id=1013619)