Unverified60% confidenceFactExact time
The conversion from FP8 to BF16 is lossless because BF16 can exactly represent all values representable by FP8 E4M3, while the reverse is not true
2
Sources
60%
Confidence
Long-term
Relevance
8/6/2026
First Seen
Sources
Q8_K_XL与MXFP4命名之争：FP8≠Q8_0的量化真相
Related Claims
VerifiedBF16 uses 1 sign bit, 8 exponent bits, and 7 mantissa bits, sacrificing precision while retaining the same dynamic range as FP32.77% similarUnverifiedUnder BF16, if the result of any floating-point operation cannot be exactly represented in BF16, it must be rounded to the nearest representable value, producing a relative error on the order of about 2^-8.69% similarUnverifiedEmbeddingGemma 2 inference must use BF16 or FP32 and should not use FP16, because activation values exceed the dynamic range of FP16, which can cause NaN values or quality degradation without any error being raised.67% similarVerifiedFP32每个参数占4字节，FP16/BF16占2字节，INT8量化占1字节，INT4量化占0.5字节65% similarUnverifiedThe maximum value of FP16 is approximately 65504; its dynamic range is smaller than that of BF16 and is prone to overflow issues.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/697311API
curl https://kongchang.com/api/v1/knowledge/claims/697311MCP
get_claim(id=697311)