huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-19 19:58:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix the block size for some cuda kernels. (#1767)

Browse Source
This commit is contained in:
Laurent Mazare
2024-02-27 14:08:33 +01:00
committed by GitHub
parent 32544a2ad6
commit 6400e1b0a0
2 changed files with 15 additions and 45 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
candle-core/tests/quantized_tests.rs
View File

@ -178,10 +178,6 @@ test_device!(
);
fn quantize_q4_0(device: &Device) -> Result<()> {
// TODO Enable this later when we enable cuda.
if device.is_cuda() {
return Ok(());
}
let src = (0..32 * 4).map(|v| v as f32).collect::<Vec<_>>();
let src = Tensor::from_slice(&src, (32 * 4,), device)?;
@ -209,10 +205,6 @@ fn quantize_q4_0(device: &Device) -> Result<()> {
}
fn quantize_q4_1(device: &Device) -> Result<()> {
// TODO Enable this later when we enable cuda.
if device.is_cuda() {
return Ok(());
}
let src = (0..32 * 4).map(|v| v as f32).collect::<Vec<_>>();
let src = Tensor::from_slice(&src, (32 * 4,), device)?;
let quant = quantized::QTensor::quantize(&src, GgmlDType::Q4_1)?;
@ -373,10 +365,6 @@ fn ggml_quantization_error_test(dtype: GgmlDType, device: &Device, max_error: f3
}
fn quantize_q2k(device: &Device) -> Result<()> {
// TODO Enable this later when we enable cuda.
if device.is_cuda() {
return Ok(());
}
let dtype = GgmlDType::Q2K;
let src = get_test_vector2(0.5, 1024, device)?;
@ -411,10 +399,6 @@ fn quantize_q2k(device: &Device) -> Result<()> {
}
fn quantize_q3k(device: &Device) -> Result<()> {
// TODO Enable this later when we enable cuda.
if device.is_cuda() {
return Ok(());
}
let dtype = GgmlDType::Q3K;
let src = get_test_vector2(0.5, 1024, device)?;
let quant = quantized::QTensor::quantize(&src, dtype)?;
@ -448,10 +432,6 @@ fn quantize_q3k(device: &Device) -> Result<()> {
}
fn quantize_q4k(device: &Device) -> Result<()> {
// TODO Enable this later when we enable cuda.
if device.is_cuda() {
return Ok(());
}
let dtype = GgmlDType::Q4K;
let src = get_test_vector2(0.5, 1024, device)?;
let quant = quantized::QTensor::quantize(&src, dtype)?;
@ -485,10 +465,6 @@ fn quantize_q4k(device: &Device) -> Result<()> {
}
fn quantize_q5k(device: &Device) -> Result<()> {
// TODO Enable this later when we enable cuda.
if device.is_cuda() {
return Ok(());
}
let dtype = GgmlDType::Q5K;
let src = get_test_vector2(0.5, 1024, device)?;
let quant = quantized::QTensor::quantize(&src, dtype)?;
@ -522,10 +498,6 @@ fn quantize_q5k(device: &Device) -> Result<()> {
}
fn quantize_q6k(device: &Device) -> Result<()> {
// TODO Enable this later when we enable cuda.
if device.is_cuda() {
return Ok(());
}
let dtype = GgmlDType::Q6K;
let src = get_test_vector2(0.5, 1024, device)?;
let quant = quantized::QTensor::quantize(&src, dtype)?;
@ -559,10 +531,6 @@ fn quantize_q6k(device: &Device) -> Result<()> {
}
fn quantize_q8k(device: &Device) -> Result<()> {
// TODO Enable this later when we enable cuda.
if device.is_cuda() {
return Ok(());
}
let dtype = GgmlDType::Q8K;
let src = get_test_vector2(0.5, 1024, device)?;
let quant = quantized::QTensor::quantize(&src, dtype)?;