Unverified50% confidenceFactExact time
TIDE-RS directly consumes per-parameter bf16 gradients and reduces them into fp32 shards, a mixed-precision mode that NCCL does not provide.
1
Sources
50%
Confidence
Long-term
Relevance
10/30/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedBF16 uses 1 sign bit, 8 exponent bits, and 7 mantissa bits, sacrificing precision while retaining the same dynamic range as FP32.73% similarUnverifiedRRF uses only ranks rather than raw scores, avoiding the problem that the dimensions of dense vectors and sparse BM25 are not comparable.69% similarUnverifiedGradient checkpointing and mixed-precision training (FP16/BF16) are common memory-reduction measures.68% similarUnverifiedLoading a model with N parameters at FP16 precision requires approximately 2N bytes of video memory.68% similarUnverifiedPreliminary endpoint results on the 32B model show the same direction: 35.53% for SFT and 22.37% for final RLVR.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1114844API
curl https://kongchang.com/api/v1/knowledge/claims/1114844MCP
get_claim(id=1114844)