diff options
| author | Laurent Mazare <laurent.mazare@gmail.com> | 2023-08-27 08:17:38 +0100 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2023-08-27 08:17:38 +0100 |
| commit | fa0d75b18d2eaa8662be52b88991bb5c87472a93 (patch) | |
| tree | 968f72f8432b87e8803612bea48944011d160306 /candle-core/src/quantized/k_quants.rs | |
| parent | 28658054ff213a5ec243ad310ada2b61019481e5 (diff) | |
| download | candle-fa0d75b18d2eaa8662be52b88991bb5c87472a93.tar.gz candle-fa0d75b18d2eaa8662be52b88991bb5c87472a93.tar.bz2 candle-fa0d75b18d2eaa8662be52b88991bb5c87472a93.zip | |
Quantization tests + fix some issues. (#616)
Diffstat (limited to 'candle-core/src/quantized/k_quants.rs')
| -rw-r--r-- | candle-core/src/quantized/k_quants.rs | 12 |
1 files changed, 6 insertions, 6 deletions
diff --git a/candle-core/src/quantized/k_quants.rs b/candle-core/src/quantized/k_quants.rs index f94ff72f..177047b6 100644 --- a/candle-core/src/quantized/k_quants.rs +++ b/candle-core/src/quantized/k_quants.rs @@ -302,9 +302,9 @@ impl GgmlType for BlockQ4_1 { ys.d = f16::from_f32(d); ys.m = f16::from_f32(min); - for (j, q) in ys.qs.iter_mut().enumerate() { - let x0 = (xs[i * qk + j] - min) * id; - let x1 = (xs[i * qk + qk / 2 + j] - min) * id; + for (j, q) in ys.qs.iter_mut().take(qk / 2).enumerate() { + let x0 = (xs[j] - min) * id; + let x1 = (xs[qk / 2 + j] - min) * id; let xi0 = u8::min(15, (x0 + 0.5) as u8); let xi1 = u8::min(15, (x1 + 0.5) as u8); @@ -496,9 +496,9 @@ impl GgmlType for BlockQ5_1 { ys.m = f16::from_f32(min); let mut qh = 0u32; - for (j, q) in ys.qs.iter_mut().enumerate() { - let x0 = (xs[i * qk + j] - min) * id; - let x1 = (xs[i * qk + qk / 2 + j] - min) * id; + for (j, q) in ys.qs.iter_mut().take(qk / 2).enumerate() { + let x0 = (xs[j] - min) * id; + let x1 = (xs[qk / 2 + j] - min) * id; let xi0 = (x0 + 0.5) as u8; let xi1 = (x1 + 0.5) as u8; |