From 927e251a12fa287e13c6bd9667ee97d783486c09 Mon Sep 17 00:00:00 2001 From: Iwan Kawrakow Date: Tue, 18 Jun 2024 20:08:28 +0300 Subject: Bitnet(1.75 bpw): higher precision fp8 scale Use 3 bits for the exponent and 5 bits for the mantissa. This makes PPL to be the same as fp16 (but the previous version with 4 bits for the exponent and mantissa was good enough for any practical purposes). --- iqk-quantize.h | 46 ++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 46 insertions(+) create mode 100644 iqk-quantize.h (limited to 'iqk-quantize.h') diff --git a/iqk-quantize.h b/iqk-quantize.h new file mode 100644 index 00000000..b89c9427 --- /dev/null +++ b/iqk-quantize.h @@ -0,0 +1,46 @@ +#pragma once + +#include + +typedef union { + float f; + uint32_t i; +} iq1bn_scale_t; + +#ifdef __cplusplus +extern "C" { +#endif + +#ifdef BITNET_IQ1BN_4x4 +static inline float iq1bn_min_value(void) { return 1.9074e-06f; } +static inline float iq1bn_max_value(void) { return 0.12109f; } +#else +static inline float iq1bn_min_value(void) { return 0.000488281f; } +static inline float iq1bn_max_value(void) { return 0.123047f; } +#endif + +static inline uint8_t iq1bn_float_to_fp8(float f) { + if (f <= iq1bn_min_value()) return 0; + if (f >= iq1bn_max_value()) return 255; + iq1bn_scale_t s; + s.f = f; +#ifdef BITNET_IQ1BN_4x4 + return ((((s.i >> 23) + 132) & 0xf) << 4) | ((s.i >> 19) & 0xf); +#else + return ((s.i >> 18) & 0x1f) | (((s.i >> 23) - 116) << 5); +#endif +} + +static inline float iq1bn_fp8_to_float(uint8_t fp8) { + iq1bn_scale_t s; +#ifdef BITNET_IQ1BN_4x4 + s.i = ((((fp8 >> 4) | 0xf0) - 132) << 23) | ((fp8 & 0x0f) << 19); +#else + s.i = (((fp8 >> 5) + 116) << 23) | ((fp8 & 0x1f) << 18); +#endif + return s.f; +} + +#ifdef __cplusplus +} +#endif -- cgit v1.2.3