Next down
—
—
No adjacent representable value is available in that direction.
Why use this float inspector?
- Inspect IEEE 754 layouts locally in the browser without uploading data.
- Switch between float16, bfloat16, float32, and float64 to compare precision and range.
- See how decimal input rounds into stored bits and what the next representable values are.
- Decode exact hex or binary bit patterns when a bug report gives raw bytes instead of a decimal number.
Wave 4 dev-data expansion
Inspect a floating-point value
Choose decimal, hex, or binary input. The copied share URL stores only the selected mode and format, not the input value.
Inputs
Decimal input lets you see the rounding error introduced by the selected format.
Decoded result
Source value
—
Bit layout
Exponent bits
—
Exponent raw
—
Mantissa bits
—
Next up
—
—
No adjacent representable value is available in that direction.
Use this page for storage questions, not just base conversion
A base converter rewrites the same integer in a different numeral system. Floating-point inspection is different: the decimal value may need rounding before it fits into the chosen sign, exponent, and mantissa layout. That is why this page shows both the stored value and the rounding error.
Why 0.1 changes
The decimal number 0.1 does not have a finite binary expansion. Every binary floating-point format stores the nearest representable value instead. Wider mantissas reduce the error, but they do not make 0.1 exact.
float16 versus bfloat16
float16 keeps more mantissa bits, so it is usually more precise near 1. bfloat16 keeps the same exponent width as float32, so it reaches much larger and smaller magnitudes with less precision. That trade-off is why bfloat16 appears in modern ML workflows.
Special values
When every exponent bit is 1, the value is either infinity or NaN depending on the mantissa. When every exponent bit is 0 and the mantissa is non-zero, the value is subnormal. Signed zero is also real in IEEE 754, so -0 and +0 can have different bit patterns.
Frequently asked questions
How is this different from a base converter?
A base converter rewrites the same integer into binary, octal, decimal, or hexadecimal. Float inspector shows how a floating-point number is rounded and stored with sign, exponent, and mantissa bits under IEEE 754 rules.
Why does 0.1 not stay exact in binary floating point?
Because 0.1 has no finite binary expansion. The selected format stores the nearest representable binary fraction, which introduces a small rounding error.
What is the difference between float16 and bfloat16?
float16 keeps more mantissa bits, so it is more precise near 1. bfloat16 keeps the same exponent width as float32, so it preserves a wider dynamic range but with fewer fraction bits.
Why does the share URL not include my input value?
This page copies a settings-only URL so test values stay out of the address bar, browser history, and shared links. The input remains local unless you copy it yourself.
Related
Comments (optional)
To reduce load, comments are fetched only when needed.