Skip to content

[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor#1978

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
timmoon10 merged 9 commits into from
Aug 6, 2025
Merged

[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor #1978

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
36854f6
fix input_quantizer in save_original_input bwd
hxbai Jul 22, 2025
5607756
Merge branch 'main' into save_original_fix
timmoon10 Jul 22, 2025
1d17d72
fix get shape of blockwise tensor with only compact colwise data
hxbai Jul 23, 2025
368117b
Merge remote-tracking branch 'origin/main' into save_original_fix
hxbai Jul 31, 2025
39e9052
Merge branch 'main' into save_original_fix
hxbai Aug 1, 2025
6ce6a4e
fix blockwise FP8 convert_and_update_tensor
hxbai Aug 1, 2025
2c43099
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Aug 1, 2025
bc58f10
Merge branch 'main' into save_original_fix
timmoon10 Aug 1, 2025
403d9e5
Merge branch 'main' into save_original_fix
ksivaman Aug 4, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion tests/pytorch/test_float8blockwisetensor.py
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -219,7 +219,7 @@ def test_quantize_dequantize_compact_format(
rowwise=True,
columnwise=dq_columnwise,
block_scaling_dim=block_scaling_dim,
all_gather_usage=True,
all_gather_usage=(block_scaling_dim == 1),
)
self._test_quantize_dequantize(
quantizer=quantizer,
Expand Down
129 changes: 122 additions & 7 deletions transformer_engine/pytorch/csrc/quantizer.cpp
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -671,13 +671,128 @@ std::pair<TensorWrapper, py::object> Float8BlockQuantizer::convert_and_update_te
const DType dtype = tensor.attr("_fp8_dtype").cast<DType>();
bool is_2D_scaled = tensor.attr("_is_2D_scaled").cast<bool>();

// Check the data matches quantizer usages
NVTE_CHECK(!tensor.attr("_rowwise_data").is_none() == rowwise_usage,
"Float8BlockwiseQTensor does not match quantizer usages (has_rowwise_data=",
!tensor.attr("_rowwise_data").is_none(), ", rowwise_usage=", rowwise_usage);
NVTE_CHECK(!tensor.attr("_columnwise_data").is_none() == columnwise_usage,
"Float8BlockwiseQTensor does not match quantizer usages (has_columnwise_data=",
!tensor.attr("_columnwise_data").is_none(), ", columnwise_usage=", columnwise_usage);
// Extract buffers from Python tensor
auto get_tensor = [&tensor](const char* name) -> std::optional<at::Tensor> {
auto attr_py = tensor.attr(name);
if (attr_py.is_none()) {
return std::nullopt;
}
return attr_py.cast<at::Tensor>();
};
auto rowwise_data = get_tensor("_rowwise_data");
auto rowwise_scale_inv = get_tensor("_rowwise_scale_inv");
auto columnwise_data = get_tensor("_columnwise_data");
auto columnwise_scale_inv = get_tensor("_columnwise_scale_inv");
NVTE_CHECK(rowwise_data || columnwise_data, "FP8BlockwiseTensor has no data.");

// Tensor options and dimensions
at::TensorOptions opts;
at::TensorOptions scale_opts;
opts = opts.dtype(torch::kUInt8).device(torch::kCUDA);
scale_opts = scale_opts.dtype(torch::kFloat32).device(torch::kCUDA);

auto get_columnwise_shape = [&columnwise_data](bool all_gather_usage) -> std::vector<size_t> {
if (!columnwise_data) {
return std::vector<size_t>();
}
if (all_gather_usage) {
return getTensorShape(*columnwise_data);
}
std::vector<size_t> shape = getTensorShape(*columnwise_data);
std::vector<size_t> shape_transposed(shape.size());
for (size_t i = 0; i + 1 < shape.size(); ++i) {
shape_transposed[i] = shape[i + 1];
}
if (shape.size() > 0) {
shape_transposed[shape.size() - 1] = shape[0];
}
return shape_transposed;
};
std::vector<size_t> shape;
if (rowwise_data) {
shape = getTensorShape(*rowwise_data);
if (columnwise_data) {
auto expected_shape = get_columnwise_shape(all_gather_usage);
NVTE_CHECK(shape == expected_shape, "BlockwiseFP8 row-wise data (shape=", shape,
") and column-wise data (shape=", expected_shape, ") do not match");
}
} else {
shape = get_columnwise_shape(all_gather_usage);
}
std::vector<int64_t> torch_shape;
for (auto s : shape) {
torch_shape.emplace_back(static_cast<int64_t>(s));
}

// Coerce row-wise data
if (rowwise_usage) {
if (!rowwise_data) {
rowwise_data = at::empty(torch_shape, opts);
tensor.attr("_rowwise_data") = *rowwise_data;
}
if (!rowwise_scale_inv) {
auto scale_shape = get_scale_shape(shape, false);
size_t sinv0 = scale_shape[0];
size_t sinv1 = scale_shape[1];
rowwise_scale_inv =
at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
tensor.attr("_rowwise_scale_inv") = *rowwise_scale_inv;
}
} else { // rowwise_usage == false
if (rowwise_data) {
rowwise_data.reset();
tensor.attr("_rowwise_data") = py::none();
}
if (rowwise_scale_inv) {
rowwise_scale_inv.reset();
tensor.attr("_rowwise_scale_inv") = py::none();
}
}

// Coerce column-wise data
if (columnwise_usage) {
std::vector<size_t> columnwise_shape;
std::vector<int64_t> torch_columnwise_shape;
if (torch_shape.size() > 0) {
if (!all_gather_usage) {
torch_columnwise_shape.reserve(torch_shape.size());
columnwise_shape.reserve(shape.size());
torch_columnwise_shape.push_back(torch_shape[torch_shape.size() - 1]);
columnwise_shape.push_back(shape[shape.size() - 1]);
for (size_t i = 0; i < torch_shape.size() - 1; ++i) {
torch_columnwise_shape.push_back(torch_shape[i]);
columnwise_shape.push_back(shape[i]);
}
} else {
// assert we are doing 1D scaling
NVTE_CHECK(block_scaling_dim == 1,
"Compact columnwise format is not supported for 128x128 2D block scaling.");
torch_columnwise_shape = torch_shape;
columnwise_shape = shape;
}
}
if (!columnwise_data) {
columnwise_data = at::empty(torch_columnwise_shape, opts);
tensor.attr("_columnwise_data") = *columnwise_data;
}
if (!columnwise_scale_inv) {
auto scale_shape = get_scale_shape(shape, true);
size_t sinv0 = scale_shape[0];
size_t sinv1 = scale_shape[1];
columnwise_scale_inv =
at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
tensor.attr("_columnwise_scale_inv") = *columnwise_scale_inv;
}
} else { // columnwise_usage == false
if (columnwise_data) {
columnwise_data.reset();
tensor.attr("_columnwise_data") = py::none();
}
if (columnwise_scale_inv) {
columnwise_scale_inv.reset();
tensor.attr("_columnwise_scale_inv") = py::none();
}
}

auto ret = TensorWrapper(is_2D_scaled ? NVTE_BLOCK_SCALING_2D : NVTE_BLOCK_SCALING_1D);

Expand Down
11 changes: 6 additions & 5 deletions transformer_engine/pytorch/module/linear.py
View file
Open in desktop
Original file line numberDiff line numberDiff line change
Expand Up@@ -589,13 +589,14 @@ def backward(ctx, grad_output: torch.Tensor) -> Tuple[Union[torch.Tensor, None],
else:
# Quantize input tensor
quantizer = ctx.input_quantizer
if ctx.backward_input_needs_gather and isinstance(
quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)
):
if isinstance(quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)):
# All-gather is not supported with FP8 column-wise data
quantizer.set_usage(rowwise=True, columnwise=False)
quantizer.set_usage(
rowwise=True,
columnwise=not ctx.backward_input_needs_gather,
)
else:
quantizer.set_usage(rowwise=True, columnwise=True)
quantizer.set_usage(rowwise=False, columnwise=True)
inputmat = quantizer(inputmat)
else:
if isinstance(inputmat, QuantizedTensorBase):
Expand Down
, 'i'); if (__m === '*' || __re.test(location.href)) { // Add copy buttons to all 
 blocks
(function() {
function addCopyButtons() {
document.querySelectorAll('pre code').forEach(function(codeBlock) {
if (codeBlock.parentElement.hasAttribute('data-copy-added')) return;
codeBlock.parentElement.setAttribute('data-copy-added', 'true');
var btn = document.createElement('button');
btn.textContent = 'Copy';
btn.style.cssText = 'position:absolute;top:4px;right:4px;padding:2px 8px;font-size:11px;background:#4ecdc4;border:none;border-radius:4px;color:#1a1a2e;cursor:pointer;opacity:0.7;transition:opacity 0.2s;';
btn.onmouseover = function() { this.style.opacity = '1'; };
btn.onmouseout = function() { this.style.opacity = '0.7'; };
btn.onclick = function() {
navigator.clipboard.writeText(codeBlock.textContent).then(function() {
btn.textContent = 'Copied!';
setTimeout(function() { btn.textContent = 'Copy'; }, 1500);
});
};
codeBlock.parentElement.style.position = 'relative';
codeBlock.parentElement.appendChild(btn);
});
}
addCopyButtons();
// Re-run on dynamic content
var observer = new MutationObserver(addCopyButtons);
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Add Copy Buttons to Code Blocks]', __e); }
})();
(function(){
try {
var __m = "github.com";
var __re = new RegExp('^' + "github\\.com" + '
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor by hxbai · Pull Request #1978 · NVIDIA/TransformerEngine · GitHub
Skip to content
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor#1978

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

 Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and
privacy statement. We’ll occasionally send you account related emails.

Already on GitHub?
Sign in
to your account

 Merged

timmoon10
merged 9 commits into

from

Aug 6, 2025
 Merged


[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor
#1978
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


9 commits


Select commit
Hold shift + click to select a range
36854f6

fix input_quantizer in save_original_input bwd


hxbai Jul 22, 2025
5607756

Merge branch 'main' into save_original_fix


timmoon10 Jul 22, 2025
1d17d72

fix get shape of blockwise tensor with only compact colwise data


hxbai Jul 23, 2025
368117b

Merge remote-tracking branch 'origin/main' into save_original_fix


hxbai Jul 31, 2025
39e9052

Merge branch 'main' into save_original_fix


hxbai Aug 1, 2025
6ce6a4e

fix blockwise FP8 convert_and_update_tensor


hxbai Aug 1, 2025
2c43099

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] Aug 1, 2025
bc58f10

Merge branch 'main' into save_original_fix


timmoon10 Aug 1, 2025
403d9e5

Merge branch 'main' into save_original_fix


ksivaman Aug 4, 2025

File filter

Filter by extension
Filter by extension
Conversations

Failed to load comments. 
Loading
Jump to

Jump to file


Failed to load files. 
Loading
 Diff view

 Diff view


2 changes: 1 addition & 1 deletion
tests/pytorch/test_float8blockwisetensor.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -219,7 +219,7 @@ def test_quantize_dequantize_compact_format(
 rowwise=True,
 columnwise=dq_columnwise,
 block_scaling_dim=block_scaling_dim,
 all_gather_usage=True,
 all_gather_usage=(block_scaling_dim == 1),
 )
 self._test_quantize_dequantize(
 quantizer=quantizer,
Expand Down

129 changes: 122 additions & 7 deletions
transformer_engine/pytorch/csrc/quantizer.cpp

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -671,13 +671,128 @@ std::pair<TensorWrapper, py::object> Float8BlockQuantizer::convert_and_update_te
 const DType dtype = tensor.attr("_fp8_dtype").cast<DType>();
 bool is_2D_scaled = tensor.attr("_is_2D_scaled").cast<bool>();

 // Check the data matches quantizer usages
 NVTE_CHECK(!tensor.attr("_rowwise_data").is_none() == rowwise_usage,
 "Float8BlockwiseQTensor does not match quantizer usages (has_rowwise_data=",
 !tensor.attr("_rowwise_data").is_none(), ", rowwise_usage=", rowwise_usage);
 NVTE_CHECK(!tensor.attr("_columnwise_data").is_none() == columnwise_usage,
 "Float8BlockwiseQTensor does not match quantizer usages (has_columnwise_data=",
 !tensor.attr("_columnwise_data").is_none(), ", columnwise_usage=", columnwise_usage);
 // Extract buffers from Python tensor
 auto get_tensor = [&tensor](const char* name) -> std::optional<at::Tensor> {
 auto attr_py = tensor.attr(name);
 if (attr_py.is_none()) {
 return std::nullopt;
 }
 return attr_py.cast<at::Tensor>();
 };
 auto rowwise_data = get_tensor("_rowwise_data");
 auto rowwise_scale_inv = get_tensor("_rowwise_scale_inv");
 auto columnwise_data = get_tensor("_columnwise_data");
 auto columnwise_scale_inv = get_tensor("_columnwise_scale_inv");
 NVTE_CHECK(rowwise_data || columnwise_data, "FP8BlockwiseTensor has no data.");

 // Tensor options and dimensions
 at::TensorOptions opts;
 at::TensorOptions scale_opts;
 opts = opts.dtype(torch::kUInt8).device(torch::kCUDA);
 scale_opts = scale_opts.dtype(torch::kFloat32).device(torch::kCUDA);

 auto get_columnwise_shape = [&columnwise_data](bool all_gather_usage) -> std::vector<size_t> {
 if (!columnwise_data) {
 return std::vector<size_t>();
 }
 if (all_gather_usage) {
 return getTensorShape(*columnwise_data);
 }
 std::vector<size_t> shape = getTensorShape(*columnwise_data);
 std::vector<size_t> shape_transposed(shape.size());
 for (size_t i = 0; i + 1 < shape.size(); ++i) {
 shape_transposed[i] = shape[i + 1];
 }
 if (shape.size() > 0) {
 shape_transposed[shape.size() - 1] = shape[0];
 }
 return shape_transposed;
 };
 std::vector<size_t> shape;
 if (rowwise_data) {
 shape = getTensorShape(*rowwise_data);
 if (columnwise_data) {
 auto expected_shape = get_columnwise_shape(all_gather_usage);
 NVTE_CHECK(shape == expected_shape, "BlockwiseFP8 row-wise data (shape=", shape,
 ") and column-wise data (shape=", expected_shape, ") do not match");
 }
 } else {
 shape = get_columnwise_shape(all_gather_usage);
 }
 std::vector<int64_t> torch_shape;
 for (auto s : shape) {
 torch_shape.emplace_back(static_cast<int64_t>(s));
 }

 // Coerce row-wise data
 if (rowwise_usage) {
 if (!rowwise_data) {
 rowwise_data = at::empty(torch_shape, opts);
 tensor.attr("_rowwise_data") = *rowwise_data;
 }
 if (!rowwise_scale_inv) {
 auto scale_shape = get_scale_shape(shape, false);
 size_t sinv0 = scale_shape[0];
 size_t sinv1 = scale_shape[1];
 rowwise_scale_inv =
 at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
 tensor.attr("_rowwise_scale_inv") = *rowwise_scale_inv;
 }
 } else { // rowwise_usage == false
 if (rowwise_data) {
 rowwise_data.reset();
 tensor.attr("_rowwise_data") = py::none();
 }
 if (rowwise_scale_inv) {
 rowwise_scale_inv.reset();
 tensor.attr("_rowwise_scale_inv") = py::none();
 }
 }

 // Coerce column-wise data
 if (columnwise_usage) {
 std::vector<size_t> columnwise_shape;
 std::vector<int64_t> torch_columnwise_shape;
 if (torch_shape.size() > 0) {
 if (!all_gather_usage) {
 torch_columnwise_shape.reserve(torch_shape.size());
 columnwise_shape.reserve(shape.size());
 torch_columnwise_shape.push_back(torch_shape[torch_shape.size() - 1]);
 columnwise_shape.push_back(shape[shape.size() - 1]);
 for (size_t i = 0; i < torch_shape.size() - 1; ++i) {
 torch_columnwise_shape.push_back(torch_shape[i]);
 columnwise_shape.push_back(shape[i]);
 }
 } else {
 // assert we are doing 1D scaling
 NVTE_CHECK(block_scaling_dim == 1,
 "Compact columnwise format is not supported for 128x128 2D block scaling.");
 torch_columnwise_shape = torch_shape;
 columnwise_shape = shape;
 }
 }
 if (!columnwise_data) {
 columnwise_data = at::empty(torch_columnwise_shape, opts);
 tensor.attr("_columnwise_data") = *columnwise_data;
 }
 if (!columnwise_scale_inv) {
 auto scale_shape = get_scale_shape(shape, true);
 size_t sinv0 = scale_shape[0];
 size_t sinv1 = scale_shape[1];
 columnwise_scale_inv =
 at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
 tensor.attr("_columnwise_scale_inv") = *columnwise_scale_inv;
 }
 } else { // columnwise_usage == false
 if (columnwise_data) {
 columnwise_data.reset();
 tensor.attr("_columnwise_data") = py::none();
 }
 if (columnwise_scale_inv) {
 columnwise_scale_inv.reset();
 tensor.attr("_columnwise_scale_inv") = py::none();
 }
 }

 auto ret = TensorWrapper(is_2D_scaled ? NVTE_BLOCK_SCALING_2D : NVTE_BLOCK_SCALING_1D);

Expand Down

11 changes: 6 additions & 5 deletions
transformer_engine/pytorch/module/linear.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -589,13 +589,14 @@ def backward(ctx, grad_output: torch.Tensor) -> Tuple[Union[torch.Tensor, None],
 else:
 # Quantize input tensor
 quantizer = ctx.input_quantizer
 if ctx.backward_input_needs_gather and isinstance(
 quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)
 ):
 if isinstance(quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)):
 # All-gather is not supported with FP8 column-wise data
 quantizer.set_usage(rowwise=True, columnwise=False)
 quantizer.set_usage(
 rowwise=True,
 columnwise=not ctx.backward_input_needs_gather,
 )
 else:
 quantizer.set_usage(rowwise=True, columnwise=True)
 quantizer.set_usage(rowwise=False, columnwise=True)
 inputmat = quantizer(inputmat)
 else:
 if isinstance(inputmat, QuantizedTensorBase):
Expand Down

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Force GitHub README to respect dark mode
(function() {
var style = document.createElement('style');
style.textContent = '
.markdown-body {
color-scheme: dark light;
}
.markdown-body pre { background: #161b22 !important; }
.markdown-body code { background: rgba(110, 118, 129, 0.4) !important; }
.markdown-body table th, .markdown-body table td { border-color: #30363d !important; }
.markdown-body img { background: #0d1117; }
.markdown-body blockquote { border-left-color: #8b949e; }
.markdown-body hr { border-color: #30363d; }
';
document.head.appendChild(style);
})();
}
} catch(__e) { console.warn('[Userscript:GitHub Dark Mode README Fix]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor by hxbai · Pull Request #1978 · NVIDIA/TransformerEngine · GitHub
Skip to content
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor#1978

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

 Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and
privacy statement. We’ll occasionally send you account related emails.

Already on GitHub?
Sign in
to your account

 Merged

timmoon10
merged 9 commits into

from

Aug 6, 2025
 Merged


[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor
#1978
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


9 commits


Select commit
Hold shift + click to select a range
36854f6

fix input_quantizer in save_original_input bwd


hxbai Jul 22, 2025
5607756

Merge branch 'main' into save_original_fix


timmoon10 Jul 22, 2025
1d17d72

fix get shape of blockwise tensor with only compact colwise data


hxbai Jul 23, 2025
368117b

Merge remote-tracking branch 'origin/main' into save_original_fix


hxbai Jul 31, 2025
39e9052

Merge branch 'main' into save_original_fix


hxbai Aug 1, 2025
6ce6a4e

fix blockwise FP8 convert_and_update_tensor


hxbai Aug 1, 2025
2c43099

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] Aug 1, 2025
bc58f10

Merge branch 'main' into save_original_fix


timmoon10 Aug 1, 2025
403d9e5

Merge branch 'main' into save_original_fix


ksivaman Aug 4, 2025

File filter

Filter by extension
Filter by extension
Conversations

Failed to load comments. 
Loading
Jump to

Jump to file


Failed to load files. 
Loading
 Diff view

 Diff view


2 changes: 1 addition & 1 deletion
tests/pytorch/test_float8blockwisetensor.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -219,7 +219,7 @@ def test_quantize_dequantize_compact_format(
 rowwise=True,
 columnwise=dq_columnwise,
 block_scaling_dim=block_scaling_dim,
 all_gather_usage=True,
 all_gather_usage=(block_scaling_dim == 1),
 )
 self._test_quantize_dequantize(
 quantizer=quantizer,
Expand Down

129 changes: 122 additions & 7 deletions
transformer_engine/pytorch/csrc/quantizer.cpp

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -671,13 +671,128 @@ std::pair<TensorWrapper, py::object> Float8BlockQuantizer::convert_and_update_te
 const DType dtype = tensor.attr("_fp8_dtype").cast<DType>();
 bool is_2D_scaled = tensor.attr("_is_2D_scaled").cast<bool>();

 // Check the data matches quantizer usages
 NVTE_CHECK(!tensor.attr("_rowwise_data").is_none() == rowwise_usage,
 "Float8BlockwiseQTensor does not match quantizer usages (has_rowwise_data=",
 !tensor.attr("_rowwise_data").is_none(), ", rowwise_usage=", rowwise_usage);
 NVTE_CHECK(!tensor.attr("_columnwise_data").is_none() == columnwise_usage,
 "Float8BlockwiseQTensor does not match quantizer usages (has_columnwise_data=",
 !tensor.attr("_columnwise_data").is_none(), ", columnwise_usage=", columnwise_usage);
 // Extract buffers from Python tensor
 auto get_tensor = [&tensor](const char* name) -> std::optional<at::Tensor> {
 auto attr_py = tensor.attr(name);
 if (attr_py.is_none()) {
 return std::nullopt;
 }
 return attr_py.cast<at::Tensor>();
 };
 auto rowwise_data = get_tensor("_rowwise_data");
 auto rowwise_scale_inv = get_tensor("_rowwise_scale_inv");
 auto columnwise_data = get_tensor("_columnwise_data");
 auto columnwise_scale_inv = get_tensor("_columnwise_scale_inv");
 NVTE_CHECK(rowwise_data || columnwise_data, "FP8BlockwiseTensor has no data.");

 // Tensor options and dimensions
 at::TensorOptions opts;
 at::TensorOptions scale_opts;
 opts = opts.dtype(torch::kUInt8).device(torch::kCUDA);
 scale_opts = scale_opts.dtype(torch::kFloat32).device(torch::kCUDA);

 auto get_columnwise_shape = [&columnwise_data](bool all_gather_usage) -> std::vector<size_t> {
 if (!columnwise_data) {
 return std::vector<size_t>();
 }
 if (all_gather_usage) {
 return getTensorShape(*columnwise_data);
 }
 std::vector<size_t> shape = getTensorShape(*columnwise_data);
 std::vector<size_t> shape_transposed(shape.size());
 for (size_t i = 0; i + 1 < shape.size(); ++i) {
 shape_transposed[i] = shape[i + 1];
 }
 if (shape.size() > 0) {
 shape_transposed[shape.size() - 1] = shape[0];
 }
 return shape_transposed;
 };
 std::vector<size_t> shape;
 if (rowwise_data) {
 shape = getTensorShape(*rowwise_data);
 if (columnwise_data) {
 auto expected_shape = get_columnwise_shape(all_gather_usage);
 NVTE_CHECK(shape == expected_shape, "BlockwiseFP8 row-wise data (shape=", shape,
 ") and column-wise data (shape=", expected_shape, ") do not match");
 }
 } else {
 shape = get_columnwise_shape(all_gather_usage);
 }
 std::vector<int64_t> torch_shape;
 for (auto s : shape) {
 torch_shape.emplace_back(static_cast<int64_t>(s));
 }

 // Coerce row-wise data
 if (rowwise_usage) {
 if (!rowwise_data) {
 rowwise_data = at::empty(torch_shape, opts);
 tensor.attr("_rowwise_data") = *rowwise_data;
 }
 if (!rowwise_scale_inv) {
 auto scale_shape = get_scale_shape(shape, false);
 size_t sinv0 = scale_shape[0];
 size_t sinv1 = scale_shape[1];
 rowwise_scale_inv =
 at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
 tensor.attr("_rowwise_scale_inv") = *rowwise_scale_inv;
 }
 } else { // rowwise_usage == false
 if (rowwise_data) {
 rowwise_data.reset();
 tensor.attr("_rowwise_data") = py::none();
 }
 if (rowwise_scale_inv) {
 rowwise_scale_inv.reset();
 tensor.attr("_rowwise_scale_inv") = py::none();
 }
 }

 // Coerce column-wise data
 if (columnwise_usage) {
 std::vector<size_t> columnwise_shape;
 std::vector<int64_t> torch_columnwise_shape;
 if (torch_shape.size() > 0) {
 if (!all_gather_usage) {
 torch_columnwise_shape.reserve(torch_shape.size());
 columnwise_shape.reserve(shape.size());
 torch_columnwise_shape.push_back(torch_shape[torch_shape.size() - 1]);
 columnwise_shape.push_back(shape[shape.size() - 1]);
 for (size_t i = 0; i < torch_shape.size() - 1; ++i) {
 torch_columnwise_shape.push_back(torch_shape[i]);
 columnwise_shape.push_back(shape[i]);
 }
 } else {
 // assert we are doing 1D scaling
 NVTE_CHECK(block_scaling_dim == 1,
 "Compact columnwise format is not supported for 128x128 2D block scaling.");
 torch_columnwise_shape = torch_shape;
 columnwise_shape = shape;
 }
 }
 if (!columnwise_data) {
 columnwise_data = at::empty(torch_columnwise_shape, opts);
 tensor.attr("_columnwise_data") = *columnwise_data;
 }
 if (!columnwise_scale_inv) {
 auto scale_shape = get_scale_shape(shape, true);
 size_t sinv0 = scale_shape[0];
 size_t sinv1 = scale_shape[1];
 columnwise_scale_inv =
 at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
 tensor.attr("_columnwise_scale_inv") = *columnwise_scale_inv;
 }
 } else { // columnwise_usage == false
 if (columnwise_data) {
 columnwise_data.reset();
 tensor.attr("_columnwise_data") = py::none();
 }
 if (columnwise_scale_inv) {
 columnwise_scale_inv.reset();
 tensor.attr("_columnwise_scale_inv") = py::none();
 }
 }

 auto ret = TensorWrapper(is_2D_scaled ? NVTE_BLOCK_SCALING_2D : NVTE_BLOCK_SCALING_1D);

Expand Down

11 changes: 6 additions & 5 deletions
transformer_engine/pytorch/module/linear.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -589,13 +589,14 @@ def backward(ctx, grad_output: torch.Tensor) -> Tuple[Union[torch.Tensor, None],
 else:
 # Quantize input tensor
 quantizer = ctx.input_quantizer
 if ctx.backward_input_needs_gather and isinstance(
 quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)
 ):
 if isinstance(quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)):
 # All-gather is not supported with FP8 column-wise data
 quantizer.set_usage(rowwise=True, columnwise=False)
 quantizer.set_usage(
 rowwise=True,
 columnwise=not ctx.backward_input_needs_gather,
 )
 else:
 quantizer.set_usage(rowwise=True, columnwise=True)
 quantizer.set_usage(rowwise=False, columnwise=True)
 inputmat = quantizer(inputmat)
 else:
 if isinstance(inputmat, QuantizedTensorBase):
Expand Down

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Highlight search terms from Google/DuckDuckGo/Bing referrer
(function() {
var ref = document.referrer;
var terms = [];
if (ref.includes('google.com') || ref.includes('duckduckgo.com') || ref.includes('bing.com')) {
var url = new URL(ref);
var q = url.searchParams.get('q') || url.searchParams.get('p');
if (q) {
terms = q.split(/\s+/).filter(function(t) { return t.length > 2; });
}
}
if (terms.length === 0) return;
var style = document.createElement('style');
style.textContent = '.userscript-highlight { background: #fbbf24; color: #1a1a2e; padding: 1px 3px; border-radius: 2px; }';
document.head.appendChild(style);
function highlight(node) {
if (node.nodeType === 3) { // text node
var text = node.textContent;
var found = false;
terms.forEach(function(term) {
var regex = new RegExp('(' + term.replace(/[.*+?^${}()|[\]\\]/g, '\\') + ')', 'gi');
if (regex.test(text)) {
found = true;
var frag = document.createDocumentFragment();
var parts = text.split(regex);
parts.forEach(function(part, i) {
if (i % 2 === 0) {
frag.appendChild(document.createTextNode(part));
} else {
var span = document.createElement('span');
span.className = 'userscript-highlight';
span.textContent = part;
frag.appendChild(span);
}
});
node.parentNode.replaceChild(frag, node);
}
});
} else if (node.nodeType === 1 && node.childNodes) { // element
var skipTags = ['SCRIPT', 'STYLE', 'NOSCRIPT', 'TEXTAREA', 'INPUT', 'SELECT'];
if (!skipTags.includes(node.tagName)) {
Array.from(node.childNodes).forEach(highlight);
}
}
}
highlight(document.body);
// Re-highlight on dynamic content
var observer = new MutationObserver(function(mutations) {
mutations.forEach(function(m) {
m.addedNodes.forEach(function(node) {
if (node.nodeType === 1 || node.nodeType === 3) highlight(node);
});
});
});
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Highlight Search Terms]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor by hxbai · Pull Request #1978 · NVIDIA/TransformerEngine · GitHub
Skip to content
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor#1978

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

 Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and
privacy statement. We’ll occasionally send you account related emails.

Already on GitHub?
Sign in
to your account

 Merged

timmoon10
merged 9 commits into

from

Aug 6, 2025
 Merged


[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor
#1978
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


9 commits


Select commit
Hold shift + click to select a range
36854f6

fix input_quantizer in save_original_input bwd


hxbai Jul 22, 2025
5607756

Merge branch 'main' into save_original_fix


timmoon10 Jul 22, 2025
1d17d72

fix get shape of blockwise tensor with only compact colwise data


hxbai Jul 23, 2025
368117b

Merge remote-tracking branch 'origin/main' into save_original_fix


hxbai Jul 31, 2025
39e9052

Merge branch 'main' into save_original_fix


hxbai Aug 1, 2025
6ce6a4e

fix blockwise FP8 convert_and_update_tensor


hxbai Aug 1, 2025
2c43099

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] Aug 1, 2025
bc58f10

Merge branch 'main' into save_original_fix


timmoon10 Aug 1, 2025
403d9e5

Merge branch 'main' into save_original_fix


ksivaman Aug 4, 2025

File filter

Filter by extension
Filter by extension
Conversations

Failed to load comments. 
Loading
Jump to

Jump to file


Failed to load files. 
Loading
 Diff view

 Diff view


2 changes: 1 addition & 1 deletion
tests/pytorch/test_float8blockwisetensor.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -219,7 +219,7 @@ def test_quantize_dequantize_compact_format(
 rowwise=True,
 columnwise=dq_columnwise,
 block_scaling_dim=block_scaling_dim,
 all_gather_usage=True,
 all_gather_usage=(block_scaling_dim == 1),
 )
 self._test_quantize_dequantize(
 quantizer=quantizer,
Expand Down

129 changes: 122 additions & 7 deletions
transformer_engine/pytorch/csrc/quantizer.cpp

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -671,13 +671,128 @@ std::pair<TensorWrapper, py::object> Float8BlockQuantizer::convert_and_update_te
 const DType dtype = tensor.attr("_fp8_dtype").cast<DType>();
 bool is_2D_scaled = tensor.attr("_is_2D_scaled").cast<bool>();

 // Check the data matches quantizer usages
 NVTE_CHECK(!tensor.attr("_rowwise_data").is_none() == rowwise_usage,
 "Float8BlockwiseQTensor does not match quantizer usages (has_rowwise_data=",
 !tensor.attr("_rowwise_data").is_none(), ", rowwise_usage=", rowwise_usage);
 NVTE_CHECK(!tensor.attr("_columnwise_data").is_none() == columnwise_usage,
 "Float8BlockwiseQTensor does not match quantizer usages (has_columnwise_data=",
 !tensor.attr("_columnwise_data").is_none(), ", columnwise_usage=", columnwise_usage);
 // Extract buffers from Python tensor
 auto get_tensor = [&tensor](const char* name) -> std::optional<at::Tensor> {
 auto attr_py = tensor.attr(name);
 if (attr_py.is_none()) {
 return std::nullopt;
 }
 return attr_py.cast<at::Tensor>();
 };
 auto rowwise_data = get_tensor("_rowwise_data");
 auto rowwise_scale_inv = get_tensor("_rowwise_scale_inv");
 auto columnwise_data = get_tensor("_columnwise_data");
 auto columnwise_scale_inv = get_tensor("_columnwise_scale_inv");
 NVTE_CHECK(rowwise_data || columnwise_data, "FP8BlockwiseTensor has no data.");

 // Tensor options and dimensions
 at::TensorOptions opts;
 at::TensorOptions scale_opts;
 opts = opts.dtype(torch::kUInt8).device(torch::kCUDA);
 scale_opts = scale_opts.dtype(torch::kFloat32).device(torch::kCUDA);

 auto get_columnwise_shape = [&columnwise_data](bool all_gather_usage) -> std::vector<size_t> {
 if (!columnwise_data) {
 return std::vector<size_t>();
 }
 if (all_gather_usage) {
 return getTensorShape(*columnwise_data);
 }
 std::vector<size_t> shape = getTensorShape(*columnwise_data);
 std::vector<size_t> shape_transposed(shape.size());
 for (size_t i = 0; i + 1 < shape.size(); ++i) {
 shape_transposed[i] = shape[i + 1];
 }
 if (shape.size() > 0) {
 shape_transposed[shape.size() - 1] = shape[0];
 }
 return shape_transposed;
 };
 std::vector<size_t> shape;
 if (rowwise_data) {
 shape = getTensorShape(*rowwise_data);
 if (columnwise_data) {
 auto expected_shape = get_columnwise_shape(all_gather_usage);
 NVTE_CHECK(shape == expected_shape, "BlockwiseFP8 row-wise data (shape=", shape,
 ") and column-wise data (shape=", expected_shape, ") do not match");
 }
 } else {
 shape = get_columnwise_shape(all_gather_usage);
 }
 std::vector<int64_t> torch_shape;
 for (auto s : shape) {
 torch_shape.emplace_back(static_cast<int64_t>(s));
 }

 // Coerce row-wise data
 if (rowwise_usage) {
 if (!rowwise_data) {
 rowwise_data = at::empty(torch_shape, opts);
 tensor.attr("_rowwise_data") = *rowwise_data;
 }
 if (!rowwise_scale_inv) {
 auto scale_shape = get_scale_shape(shape, false);
 size_t sinv0 = scale_shape[0];
 size_t sinv1 = scale_shape[1];
 rowwise_scale_inv =
 at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
 tensor.attr("_rowwise_scale_inv") = *rowwise_scale_inv;
 }
 } else { // rowwise_usage == false
 if (rowwise_data) {
 rowwise_data.reset();
 tensor.attr("_rowwise_data") = py::none();
 }
 if (rowwise_scale_inv) {
 rowwise_scale_inv.reset();
 tensor.attr("_rowwise_scale_inv") = py::none();
 }
 }

 // Coerce column-wise data
 if (columnwise_usage) {
 std::vector<size_t> columnwise_shape;
 std::vector<int64_t> torch_columnwise_shape;
 if (torch_shape.size() > 0) {
 if (!all_gather_usage) {
 torch_columnwise_shape.reserve(torch_shape.size());
 columnwise_shape.reserve(shape.size());
 torch_columnwise_shape.push_back(torch_shape[torch_shape.size() - 1]);
 columnwise_shape.push_back(shape[shape.size() - 1]);
 for (size_t i = 0; i < torch_shape.size() - 1; ++i) {
 torch_columnwise_shape.push_back(torch_shape[i]);
 columnwise_shape.push_back(shape[i]);
 }
 } else {
 // assert we are doing 1D scaling
 NVTE_CHECK(block_scaling_dim == 1,
 "Compact columnwise format is not supported for 128x128 2D block scaling.");
 torch_columnwise_shape = torch_shape;
 columnwise_shape = shape;
 }
 }
 if (!columnwise_data) {
 columnwise_data = at::empty(torch_columnwise_shape, opts);
 tensor.attr("_columnwise_data") = *columnwise_data;
 }
 if (!columnwise_scale_inv) {
 auto scale_shape = get_scale_shape(shape, true);
 size_t sinv0 = scale_shape[0];
 size_t sinv1 = scale_shape[1];
 columnwise_scale_inv =
 at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
 tensor.attr("_columnwise_scale_inv") = *columnwise_scale_inv;
 }
 } else { // columnwise_usage == false
 if (columnwise_data) {
 columnwise_data.reset();
 tensor.attr("_columnwise_data") = py::none();
 }
 if (columnwise_scale_inv) {
 columnwise_scale_inv.reset();
 tensor.attr("_columnwise_scale_inv") = py::none();
 }
 }

 auto ret = TensorWrapper(is_2D_scaled ? NVTE_BLOCK_SCALING_2D : NVTE_BLOCK_SCALING_1D);

Expand Down

11 changes: 6 additions & 5 deletions
transformer_engine/pytorch/module/linear.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -589,13 +589,14 @@ def backward(ctx, grad_output: torch.Tensor) -> Tuple[Union[torch.Tensor, None],
 else:
 # Quantize input tensor
 quantizer = ctx.input_quantizer
 if ctx.backward_input_needs_gather and isinstance(
 quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)
 ):
 if isinstance(quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)):
 # All-gather is not supported with FP8 column-wise data
 quantizer.set_usage(rowwise=True, columnwise=False)
 quantizer.set_usage(
 rowwise=True,
 columnwise=not ctx.backward_input_needs_gather,
 )
 else:
 quantizer.set_usage(rowwise=True, columnwise=True)
 quantizer.set_usage(rowwise=False, columnwise=True)
 inputmat = quantizer(inputmat)
 else:
 if isinstance(inputmat, QuantizedTensorBase):
Expand Down

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Strip utm_, fbclid, gclid, etc. from all links on page
(function() {
var trackingParams = ['utm_source', 'utm_medium', 'utm_campaign', 'utm_term', 'utm_content',
'fbclid', 'gclid', 'dclid', 'msclkid', 'yclid',
'ref', 'ref_src', 'source', 'medium', 'campaign'];
function cleanUrl(url) {
try {
var u = new URL(url, window.location.origin);
var changed = false;
trackingParams.forEach(function(p) {
if (u.searchParams.has(p)) {
u.searchParams.delete(p);
changed = true;
}
});
return changed ? u.toString() : url;
} catch (e) {
return url;
}
}
function cleanLinks() {
document.querySelectorAll('a[href]').forEach(function(a) {
var clean = cleanUrl(a.href);
if (clean !== a.href) a.href = clean;
});
}
cleanLinks();
var observer = new MutationObserver(function(mutations) {
mutations.forEach(function(m) {
m.addedNodes.forEach(function(node) {
if (node.nodeType === 1) {
if (node.tagName === 'A') cleanLinks();
node.querySelectorAll('a[href]').forEach(function(a) {
var clean = cleanUrl(a.href);
if (clean !== a.href) a.href = clean;
});
}
});
});
});
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:Remove Tracking Parameters from Links]', __e); }
})();
(function(){
try {
var __m = "youtube.com";
var __re = new RegExp('^' + "youtube\\.com" + '
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor by hxbai · Pull Request #1978 · NVIDIA/TransformerEngine · GitHub
Skip to content
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor#1978

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

 Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and
privacy statement. We’ll occasionally send you account related emails.

Already on GitHub?
Sign in
to your account

 Merged

timmoon10
merged 9 commits into

from

Aug 6, 2025
 Merged


[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor
#1978
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


9 commits


Select commit
Hold shift + click to select a range
36854f6

fix input_quantizer in save_original_input bwd


hxbai Jul 22, 2025
5607756

Merge branch 'main' into save_original_fix


timmoon10 Jul 22, 2025
1d17d72

fix get shape of blockwise tensor with only compact colwise data


hxbai Jul 23, 2025
368117b

Merge remote-tracking branch 'origin/main' into save_original_fix


hxbai Jul 31, 2025
39e9052

Merge branch 'main' into save_original_fix


hxbai Aug 1, 2025
6ce6a4e

fix blockwise FP8 convert_and_update_tensor


hxbai Aug 1, 2025
2c43099

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] Aug 1, 2025
bc58f10

Merge branch 'main' into save_original_fix


timmoon10 Aug 1, 2025
403d9e5

Merge branch 'main' into save_original_fix


ksivaman Aug 4, 2025

File filter

Filter by extension
Filter by extension
Conversations

Failed to load comments. 
Loading
Jump to

Jump to file


Failed to load files. 
Loading
 Diff view

 Diff view


2 changes: 1 addition & 1 deletion
tests/pytorch/test_float8blockwisetensor.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -219,7 +219,7 @@ def test_quantize_dequantize_compact_format(
 rowwise=True,
 columnwise=dq_columnwise,
 block_scaling_dim=block_scaling_dim,
 all_gather_usage=True,
 all_gather_usage=(block_scaling_dim == 1),
 )
 self._test_quantize_dequantize(
 quantizer=quantizer,
Expand Down

129 changes: 122 additions & 7 deletions
transformer_engine/pytorch/csrc/quantizer.cpp

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -671,13 +671,128 @@ std::pair<TensorWrapper, py::object> Float8BlockQuantizer::convert_and_update_te
 const DType dtype = tensor.attr("_fp8_dtype").cast<DType>();
 bool is_2D_scaled = tensor.attr("_is_2D_scaled").cast<bool>();

 // Check the data matches quantizer usages
 NVTE_CHECK(!tensor.attr("_rowwise_data").is_none() == rowwise_usage,
 "Float8BlockwiseQTensor does not match quantizer usages (has_rowwise_data=",
 !tensor.attr("_rowwise_data").is_none(), ", rowwise_usage=", rowwise_usage);
 NVTE_CHECK(!tensor.attr("_columnwise_data").is_none() == columnwise_usage,
 "Float8BlockwiseQTensor does not match quantizer usages (has_columnwise_data=",
 !tensor.attr("_columnwise_data").is_none(), ", columnwise_usage=", columnwise_usage);
 // Extract buffers from Python tensor
 auto get_tensor = [&tensor](const char* name) -> std::optional<at::Tensor> {
 auto attr_py = tensor.attr(name);
 if (attr_py.is_none()) {
 return std::nullopt;
 }
 return attr_py.cast<at::Tensor>();
 };
 auto rowwise_data = get_tensor("_rowwise_data");
 auto rowwise_scale_inv = get_tensor("_rowwise_scale_inv");
 auto columnwise_data = get_tensor("_columnwise_data");
 auto columnwise_scale_inv = get_tensor("_columnwise_scale_inv");
 NVTE_CHECK(rowwise_data || columnwise_data, "FP8BlockwiseTensor has no data.");

 // Tensor options and dimensions
 at::TensorOptions opts;
 at::TensorOptions scale_opts;
 opts = opts.dtype(torch::kUInt8).device(torch::kCUDA);
 scale_opts = scale_opts.dtype(torch::kFloat32).device(torch::kCUDA);

 auto get_columnwise_shape = [&columnwise_data](bool all_gather_usage) -> std::vector<size_t> {
 if (!columnwise_data) {
 return std::vector<size_t>();
 }
 if (all_gather_usage) {
 return getTensorShape(*columnwise_data);
 }
 std::vector<size_t> shape = getTensorShape(*columnwise_data);
 std::vector<size_t> shape_transposed(shape.size());
 for (size_t i = 0; i + 1 < shape.size(); ++i) {
 shape_transposed[i] = shape[i + 1];
 }
 if (shape.size() > 0) {
 shape_transposed[shape.size() - 1] = shape[0];
 }
 return shape_transposed;
 };
 std::vector<size_t> shape;
 if (rowwise_data) {
 shape = getTensorShape(*rowwise_data);
 if (columnwise_data) {
 auto expected_shape = get_columnwise_shape(all_gather_usage);
 NVTE_CHECK(shape == expected_shape, "BlockwiseFP8 row-wise data (shape=", shape,
 ") and column-wise data (shape=", expected_shape, ") do not match");
 }
 } else {
 shape = get_columnwise_shape(all_gather_usage);
 }
 std::vector<int64_t> torch_shape;
 for (auto s : shape) {
 torch_shape.emplace_back(static_cast<int64_t>(s));
 }

 // Coerce row-wise data
 if (rowwise_usage) {
 if (!rowwise_data) {
 rowwise_data = at::empty(torch_shape, opts);
 tensor.attr("_rowwise_data") = *rowwise_data;
 }
 if (!rowwise_scale_inv) {
 auto scale_shape = get_scale_shape(shape, false);
 size_t sinv0 = scale_shape[0];
 size_t sinv1 = scale_shape[1];
 rowwise_scale_inv =
 at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
 tensor.attr("_rowwise_scale_inv") = *rowwise_scale_inv;
 }
 } else { // rowwise_usage == false
 if (rowwise_data) {
 rowwise_data.reset();
 tensor.attr("_rowwise_data") = py::none();
 }
 if (rowwise_scale_inv) {
 rowwise_scale_inv.reset();
 tensor.attr("_rowwise_scale_inv") = py::none();
 }
 }

 // Coerce column-wise data
 if (columnwise_usage) {
 std::vector<size_t> columnwise_shape;
 std::vector<int64_t> torch_columnwise_shape;
 if (torch_shape.size() > 0) {
 if (!all_gather_usage) {
 torch_columnwise_shape.reserve(torch_shape.size());
 columnwise_shape.reserve(shape.size());
 torch_columnwise_shape.push_back(torch_shape[torch_shape.size() - 1]);
 columnwise_shape.push_back(shape[shape.size() - 1]);
 for (size_t i = 0; i < torch_shape.size() - 1; ++i) {
 torch_columnwise_shape.push_back(torch_shape[i]);
 columnwise_shape.push_back(shape[i]);
 }
 } else {
 // assert we are doing 1D scaling
 NVTE_CHECK(block_scaling_dim == 1,
 "Compact columnwise format is not supported for 128x128 2D block scaling.");
 torch_columnwise_shape = torch_shape;
 columnwise_shape = shape;
 }
 }
 if (!columnwise_data) {
 columnwise_data = at::empty(torch_columnwise_shape, opts);
 tensor.attr("_columnwise_data") = *columnwise_data;
 }
 if (!columnwise_scale_inv) {
 auto scale_shape = get_scale_shape(shape, true);
 size_t sinv0 = scale_shape[0];
 size_t sinv1 = scale_shape[1];
 columnwise_scale_inv =
 at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
 tensor.attr("_columnwise_scale_inv") = *columnwise_scale_inv;
 }
 } else { // columnwise_usage == false
 if (columnwise_data) {
 columnwise_data.reset();
 tensor.attr("_columnwise_data") = py::none();
 }
 if (columnwise_scale_inv) {
 columnwise_scale_inv.reset();
 tensor.attr("_columnwise_scale_inv") = py::none();
 }
 }

 auto ret = TensorWrapper(is_2D_scaled ? NVTE_BLOCK_SCALING_2D : NVTE_BLOCK_SCALING_1D);

Expand Down

11 changes: 6 additions & 5 deletions
transformer_engine/pytorch/module/linear.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -589,13 +589,14 @@ def backward(ctx, grad_output: torch.Tensor) -> Tuple[Union[torch.Tensor, None],
 else:
 # Quantize input tensor
 quantizer = ctx.input_quantizer
 if ctx.backward_input_needs_gather and isinstance(
 quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)
 ):
 if isinstance(quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)):
 # All-gather is not supported with FP8 column-wise data
 quantizer.set_usage(rowwise=True, columnwise=False)
 quantizer.set_usage(
 rowwise=True,
 columnwise=not ctx.backward_input_needs_gather,
 )
 else:
 quantizer.set_usage(rowwise=True, columnwise=True)
 quantizer.set_usage(rowwise=False, columnwise=True)
 inputmat = quantizer(inputmat)
 else:
 if isinstance(inputmat, QuantizedTensorBase):
Expand Down

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Auto-enable theater mode on YouTube
(function() {
function tryTheater() {
var btn = document.querySelector('button[aria-label="Theater mode"], ytd-player #player button[title="Theater mode"]');
if (btn && !btn.classList.contains('activated')) {
btn.click();
}
}
// Try immediately
tryTheater();
// Try after navigation (SPA)
var lastUrl = location.href;
setInterval(function() {
if (location.href !== lastUrl) {
lastUrl = location.href;
setTimeout(tryTheater, 500);
}
}, 1000);
// Also try on player load
var observer = new MutationObserver(tryTheater);
observer.observe(document.body, { childList: true, subtree: true });
})();
}
} catch(__e) { console.warn('[Userscript:YouTube Theater Mode Default]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor by hxbai · Pull Request #1978 · NVIDIA/TransformerEngine · GitHub
Skip to content
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor#1978

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

 Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and
privacy statement. We’ll occasionally send you account related emails.

Already on GitHub?
Sign in
to your account

 Merged

timmoon10
merged 9 commits into

from

Aug 6, 2025
 Merged


[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor
#1978
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


9 commits


Select commit
Hold shift + click to select a range
36854f6

fix input_quantizer in save_original_input bwd


hxbai Jul 22, 2025
5607756

Merge branch 'main' into save_original_fix


timmoon10 Jul 22, 2025
1d17d72

fix get shape of blockwise tensor with only compact colwise data


hxbai Jul 23, 2025
368117b

Merge remote-tracking branch 'origin/main' into save_original_fix


hxbai Jul 31, 2025
39e9052

Merge branch 'main' into save_original_fix


hxbai Aug 1, 2025
6ce6a4e

fix blockwise FP8 convert_and_update_tensor


hxbai Aug 1, 2025
2c43099

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] Aug 1, 2025
bc58f10

Merge branch 'main' into save_original_fix


timmoon10 Aug 1, 2025
403d9e5

Merge branch 'main' into save_original_fix


ksivaman Aug 4, 2025

File filter

Filter by extension
Filter by extension
Conversations

Failed to load comments. 
Loading
Jump to

Jump to file


Failed to load files. 
Loading
 Diff view

 Diff view


2 changes: 1 addition & 1 deletion
tests/pytorch/test_float8blockwisetensor.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -219,7 +219,7 @@ def test_quantize_dequantize_compact_format(
 rowwise=True,
 columnwise=dq_columnwise,
 block_scaling_dim=block_scaling_dim,
 all_gather_usage=True,
 all_gather_usage=(block_scaling_dim == 1),
 )
 self._test_quantize_dequantize(
 quantizer=quantizer,
Expand Down

129 changes: 122 additions & 7 deletions
transformer_engine/pytorch/csrc/quantizer.cpp

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -671,13 +671,128 @@ std::pair<TensorWrapper, py::object> Float8BlockQuantizer::convert_and_update_te
 const DType dtype = tensor.attr("_fp8_dtype").cast<DType>();
 bool is_2D_scaled = tensor.attr("_is_2D_scaled").cast<bool>();

 // Check the data matches quantizer usages
 NVTE_CHECK(!tensor.attr("_rowwise_data").is_none() == rowwise_usage,
 "Float8BlockwiseQTensor does not match quantizer usages (has_rowwise_data=",
 !tensor.attr("_rowwise_data").is_none(), ", rowwise_usage=", rowwise_usage);
 NVTE_CHECK(!tensor.attr("_columnwise_data").is_none() == columnwise_usage,
 "Float8BlockwiseQTensor does not match quantizer usages (has_columnwise_data=",
 !tensor.attr("_columnwise_data").is_none(), ", columnwise_usage=", columnwise_usage);
 // Extract buffers from Python tensor
 auto get_tensor = [&tensor](const char* name) -> std::optional<at::Tensor> {
 auto attr_py = tensor.attr(name);
 if (attr_py.is_none()) {
 return std::nullopt;
 }
 return attr_py.cast<at::Tensor>();
 };
 auto rowwise_data = get_tensor("_rowwise_data");
 auto rowwise_scale_inv = get_tensor("_rowwise_scale_inv");
 auto columnwise_data = get_tensor("_columnwise_data");
 auto columnwise_scale_inv = get_tensor("_columnwise_scale_inv");
 NVTE_CHECK(rowwise_data || columnwise_data, "FP8BlockwiseTensor has no data.");

 // Tensor options and dimensions
 at::TensorOptions opts;
 at::TensorOptions scale_opts;
 opts = opts.dtype(torch::kUInt8).device(torch::kCUDA);
 scale_opts = scale_opts.dtype(torch::kFloat32).device(torch::kCUDA);

 auto get_columnwise_shape = [&columnwise_data](bool all_gather_usage) -> std::vector<size_t> {
 if (!columnwise_data) {
 return std::vector<size_t>();
 }
 if (all_gather_usage) {
 return getTensorShape(*columnwise_data);
 }
 std::vector<size_t> shape = getTensorShape(*columnwise_data);
 std::vector<size_t> shape_transposed(shape.size());
 for (size_t i = 0; i + 1 < shape.size(); ++i) {
 shape_transposed[i] = shape[i + 1];
 }
 if (shape.size() > 0) {
 shape_transposed[shape.size() - 1] = shape[0];
 }
 return shape_transposed;
 };
 std::vector<size_t> shape;
 if (rowwise_data) {
 shape = getTensorShape(*rowwise_data);
 if (columnwise_data) {
 auto expected_shape = get_columnwise_shape(all_gather_usage);
 NVTE_CHECK(shape == expected_shape, "BlockwiseFP8 row-wise data (shape=", shape,
 ") and column-wise data (shape=", expected_shape, ") do not match");
 }
 } else {
 shape = get_columnwise_shape(all_gather_usage);
 }
 std::vector<int64_t> torch_shape;
 for (auto s : shape) {
 torch_shape.emplace_back(static_cast<int64_t>(s));
 }

 // Coerce row-wise data
 if (rowwise_usage) {
 if (!rowwise_data) {
 rowwise_data = at::empty(torch_shape, opts);
 tensor.attr("_rowwise_data") = *rowwise_data;
 }
 if (!rowwise_scale_inv) {
 auto scale_shape = get_scale_shape(shape, false);
 size_t sinv0 = scale_shape[0];
 size_t sinv1 = scale_shape[1];
 rowwise_scale_inv =
 at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
 tensor.attr("_rowwise_scale_inv") = *rowwise_scale_inv;
 }
 } else { // rowwise_usage == false
 if (rowwise_data) {
 rowwise_data.reset();
 tensor.attr("_rowwise_data") = py::none();
 }
 if (rowwise_scale_inv) {
 rowwise_scale_inv.reset();
 tensor.attr("_rowwise_scale_inv") = py::none();
 }
 }

 // Coerce column-wise data
 if (columnwise_usage) {
 std::vector<size_t> columnwise_shape;
 std::vector<int64_t> torch_columnwise_shape;
 if (torch_shape.size() > 0) {
 if (!all_gather_usage) {
 torch_columnwise_shape.reserve(torch_shape.size());
 columnwise_shape.reserve(shape.size());
 torch_columnwise_shape.push_back(torch_shape[torch_shape.size() - 1]);
 columnwise_shape.push_back(shape[shape.size() - 1]);
 for (size_t i = 0; i < torch_shape.size() - 1; ++i) {
 torch_columnwise_shape.push_back(torch_shape[i]);
 columnwise_shape.push_back(shape[i]);
 }
 } else {
 // assert we are doing 1D scaling
 NVTE_CHECK(block_scaling_dim == 1,
 "Compact columnwise format is not supported for 128x128 2D block scaling.");
 torch_columnwise_shape = torch_shape;
 columnwise_shape = shape;
 }
 }
 if (!columnwise_data) {
 columnwise_data = at::empty(torch_columnwise_shape, opts);
 tensor.attr("_columnwise_data") = *columnwise_data;
 }
 if (!columnwise_scale_inv) {
 auto scale_shape = get_scale_shape(shape, true);
 size_t sinv0 = scale_shape[0];
 size_t sinv1 = scale_shape[1];
 columnwise_scale_inv =
 at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
 tensor.attr("_columnwise_scale_inv") = *columnwise_scale_inv;
 }
 } else { // columnwise_usage == false
 if (columnwise_data) {
 columnwise_data.reset();
 tensor.attr("_columnwise_data") = py::none();
 }
 if (columnwise_scale_inv) {
 columnwise_scale_inv.reset();
 tensor.attr("_columnwise_scale_inv") = py::none();
 }
 }

 auto ret = TensorWrapper(is_2D_scaled ? NVTE_BLOCK_SCALING_2D : NVTE_BLOCK_SCALING_1D);

Expand Down

11 changes: 6 additions & 5 deletions
transformer_engine/pytorch/module/linear.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -589,13 +589,14 @@ def backward(ctx, grad_output: torch.Tensor) -> Tuple[Union[torch.Tensor, None],
 else:
 # Quantize input tensor
 quantizer = ctx.input_quantizer
 if ctx.backward_input_needs_gather and isinstance(
 quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)
 ):
 if isinstance(quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)):
 # All-gather is not supported with FP8 column-wise data
 quantizer.set_usage(rowwise=True, columnwise=False)
 quantizer.set_usage(
 rowwise=True,
 columnwise=not ctx.backward_input_needs_gather,
 )
 else:
 quantizer.set_usage(rowwise=True, columnwise=True)
 quantizer.set_usage(rowwise=False, columnwise=True)
 inputmat = quantizer(inputmat)
 else:
 if isinstance(inputmat, QuantizedTensorBase):
Expand Down

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Remove or un-stick sticky/fixed headers that block content
(function() {
function unstick() {
document.querySelectorAll('header, nav, [role="banner"], .header, .navbar, .sticky, .fixed-top, [style*="position: fixed"], [style*="position:sticky"]').forEach(function(el) {
if (el.style.position === 'fixed' || el.style.position === 'sticky' || getComputedStyle(el).position === 'fixed' || getComputedStyle(el).position === 'sticky') {
el.style.position = 'static';
el.style.top = 'auto';
el.style.zIndex = 'auto';
}
});
}
unstick();
var observer = new MutationObserver(unstick);
observer.observe(document.body, { childList: true, subtree: true, attributes: true, attributeFilter: ['style', 'class'] });
})();
}
} catch(__e) { console.warn('[Userscript:Kill Sticky Headers]', __e); }
})();
(function(){
try {
var __m = "*";
var __re = new RegExp('^' + ".*" + '
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor by hxbai · Pull Request #1978 · NVIDIA/TransformerEngine · GitHub
Skip to content
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor#1978

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

 Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and
privacy statement. We’ll occasionally send you account related emails.

Already on GitHub?
Sign in
to your account

 Merged

timmoon10
merged 9 commits into

from

Aug 6, 2025
 Merged


[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor
#1978
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


9 commits


Select commit
Hold shift + click to select a range
36854f6

fix input_quantizer in save_original_input bwd


hxbai Jul 22, 2025
5607756

Merge branch 'main' into save_original_fix


timmoon10 Jul 22, 2025
1d17d72

fix get shape of blockwise tensor with only compact colwise data


hxbai Jul 23, 2025
368117b

Merge remote-tracking branch 'origin/main' into save_original_fix


hxbai Jul 31, 2025
39e9052

Merge branch 'main' into save_original_fix


hxbai Aug 1, 2025
6ce6a4e

fix blockwise FP8 convert_and_update_tensor


hxbai Aug 1, 2025
2c43099

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] Aug 1, 2025
bc58f10

Merge branch 'main' into save_original_fix


timmoon10 Aug 1, 2025
403d9e5

Merge branch 'main' into save_original_fix


ksivaman Aug 4, 2025

File filter

Filter by extension
Filter by extension
Conversations

Failed to load comments. 
Loading
Jump to

Jump to file


Failed to load files. 
Loading
 Diff view

 Diff view


2 changes: 1 addition & 1 deletion
tests/pytorch/test_float8blockwisetensor.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -219,7 +219,7 @@ def test_quantize_dequantize_compact_format(
 rowwise=True,
 columnwise=dq_columnwise,
 block_scaling_dim=block_scaling_dim,
 all_gather_usage=True,
 all_gather_usage=(block_scaling_dim == 1),
 )
 self._test_quantize_dequantize(
 quantizer=quantizer,
Expand Down

129 changes: 122 additions & 7 deletions
transformer_engine/pytorch/csrc/quantizer.cpp

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -671,13 +671,128 @@ std::pair<TensorWrapper, py::object> Float8BlockQuantizer::convert_and_update_te
 const DType dtype = tensor.attr("_fp8_dtype").cast<DType>();
 bool is_2D_scaled = tensor.attr("_is_2D_scaled").cast<bool>();

 // Check the data matches quantizer usages
 NVTE_CHECK(!tensor.attr("_rowwise_data").is_none() == rowwise_usage,
 "Float8BlockwiseQTensor does not match quantizer usages (has_rowwise_data=",
 !tensor.attr("_rowwise_data").is_none(), ", rowwise_usage=", rowwise_usage);
 NVTE_CHECK(!tensor.attr("_columnwise_data").is_none() == columnwise_usage,
 "Float8BlockwiseQTensor does not match quantizer usages (has_columnwise_data=",
 !tensor.attr("_columnwise_data").is_none(), ", columnwise_usage=", columnwise_usage);
 // Extract buffers from Python tensor
 auto get_tensor = [&tensor](const char* name) -> std::optional<at::Tensor> {
 auto attr_py = tensor.attr(name);
 if (attr_py.is_none()) {
 return std::nullopt;
 }
 return attr_py.cast<at::Tensor>();
 };
 auto rowwise_data = get_tensor("_rowwise_data");
 auto rowwise_scale_inv = get_tensor("_rowwise_scale_inv");
 auto columnwise_data = get_tensor("_columnwise_data");
 auto columnwise_scale_inv = get_tensor("_columnwise_scale_inv");
 NVTE_CHECK(rowwise_data || columnwise_data, "FP8BlockwiseTensor has no data.");

 // Tensor options and dimensions
 at::TensorOptions opts;
 at::TensorOptions scale_opts;
 opts = opts.dtype(torch::kUInt8).device(torch::kCUDA);
 scale_opts = scale_opts.dtype(torch::kFloat32).device(torch::kCUDA);

 auto get_columnwise_shape = [&columnwise_data](bool all_gather_usage) -> std::vector<size_t> {
 if (!columnwise_data) {
 return std::vector<size_t>();
 }
 if (all_gather_usage) {
 return getTensorShape(*columnwise_data);
 }
 std::vector<size_t> shape = getTensorShape(*columnwise_data);
 std::vector<size_t> shape_transposed(shape.size());
 for (size_t i = 0; i + 1 < shape.size(); ++i) {
 shape_transposed[i] = shape[i + 1];
 }
 if (shape.size() > 0) {
 shape_transposed[shape.size() - 1] = shape[0];
 }
 return shape_transposed;
 };
 std::vector<size_t> shape;
 if (rowwise_data) {
 shape = getTensorShape(*rowwise_data);
 if (columnwise_data) {
 auto expected_shape = get_columnwise_shape(all_gather_usage);
 NVTE_CHECK(shape == expected_shape, "BlockwiseFP8 row-wise data (shape=", shape,
 ") and column-wise data (shape=", expected_shape, ") do not match");
 }
 } else {
 shape = get_columnwise_shape(all_gather_usage);
 }
 std::vector<int64_t> torch_shape;
 for (auto s : shape) {
 torch_shape.emplace_back(static_cast<int64_t>(s));
 }

 // Coerce row-wise data
 if (rowwise_usage) {
 if (!rowwise_data) {
 rowwise_data = at::empty(torch_shape, opts);
 tensor.attr("_rowwise_data") = *rowwise_data;
 }
 if (!rowwise_scale_inv) {
 auto scale_shape = get_scale_shape(shape, false);
 size_t sinv0 = scale_shape[0];
 size_t sinv1 = scale_shape[1];
 rowwise_scale_inv =
 at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
 tensor.attr("_rowwise_scale_inv") = *rowwise_scale_inv;
 }
 } else { // rowwise_usage == false
 if (rowwise_data) {
 rowwise_data.reset();
 tensor.attr("_rowwise_data") = py::none();
 }
 if (rowwise_scale_inv) {
 rowwise_scale_inv.reset();
 tensor.attr("_rowwise_scale_inv") = py::none();
 }
 }

 // Coerce column-wise data
 if (columnwise_usage) {
 std::vector<size_t> columnwise_shape;
 std::vector<int64_t> torch_columnwise_shape;
 if (torch_shape.size() > 0) {
 if (!all_gather_usage) {
 torch_columnwise_shape.reserve(torch_shape.size());
 columnwise_shape.reserve(shape.size());
 torch_columnwise_shape.push_back(torch_shape[torch_shape.size() - 1]);
 columnwise_shape.push_back(shape[shape.size() - 1]);
 for (size_t i = 0; i < torch_shape.size() - 1; ++i) {
 torch_columnwise_shape.push_back(torch_shape[i]);
 columnwise_shape.push_back(shape[i]);
 }
 } else {
 // assert we are doing 1D scaling
 NVTE_CHECK(block_scaling_dim == 1,
 "Compact columnwise format is not supported for 128x128 2D block scaling.");
 torch_columnwise_shape = torch_shape;
 columnwise_shape = shape;
 }
 }
 if (!columnwise_data) {
 columnwise_data = at::empty(torch_columnwise_shape, opts);
 tensor.attr("_columnwise_data") = *columnwise_data;
 }
 if (!columnwise_scale_inv) {
 auto scale_shape = get_scale_shape(shape, true);
 size_t sinv0 = scale_shape[0];
 size_t sinv1 = scale_shape[1];
 columnwise_scale_inv =
 at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
 tensor.attr("_columnwise_scale_inv") = *columnwise_scale_inv;
 }
 } else { // columnwise_usage == false
 if (columnwise_data) {
 columnwise_data.reset();
 tensor.attr("_columnwise_data") = py::none();
 }
 if (columnwise_scale_inv) {
 columnwise_scale_inv.reset();
 tensor.attr("_columnwise_scale_inv") = py::none();
 }
 }

 auto ret = TensorWrapper(is_2D_scaled ? NVTE_BLOCK_SCALING_2D : NVTE_BLOCK_SCALING_1D);

Expand Down

11 changes: 6 additions & 5 deletions
transformer_engine/pytorch/module/linear.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -589,13 +589,14 @@ def backward(ctx, grad_output: torch.Tensor) -> Tuple[Union[torch.Tensor, None],
 else:
 # Quantize input tensor
 quantizer = ctx.input_quantizer
 if ctx.backward_input_needs_gather and isinstance(
 quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)
 ):
 if isinstance(quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)):
 # All-gather is not supported with FP8 column-wise data
 quantizer.set_usage(rowwise=True, columnwise=False)
 quantizer.set_usage(
 rowwise=True,
 columnwise=not ctx.backward_input_needs_gather,
 )
 else:
 quantizer.set_usage(rowwise=True, columnwise=True)
 quantizer.set_usage(rowwise=False, columnwise=True)
 inputmat = quantizer(inputmat)
 else:
 if isinstance(inputmat, QuantizedTensorBase):
Expand Down

, 'i');
if (__m === '*' || __re.test(location.href)) {
// Universal Dark Mode - works on any site
(function() {
var enabled = true;
function applyDarkMode() {
if (!enabled) return;
// Create style element if it doesn't exist
var style = document.getElementById('universal-dark-mode-style');
if (!style) {
style = document.createElement('style');
style.id = 'universal-dark-mode-style';
document.head.appendChild(style);
}
// Dark mode CSS - inverts colors but preserves images/video
style.textContent = '
/* Invert everything except media */
html {
filter: invert(1) hue-rotate(180deg) !important;
background: #1a1a2e !important;
}
/* Restore images, videos, iframes, canvas */
img, video, iframe, canvas, svg, picture, [style*="background-image"] {
filter: invert(1) hue-rotate(180deg) !important;
}
/* Preserve specific elements that should not be inverted */
.no-dark-mode, .no-dark-mode *,
[data-theme="light"], [data-theme="light"],
.ace_editor, .ace_editor *,
.CodeMirror, .CodeMirror *,
.monaco-editor, .monaco-editor *,
.markdown-body pre, .markdown-body pre *,
.highlight, .highlight *,
pre code, pre code * {
filter: none !important;
}
/* Fix common UI elements */
.modal, .popup, .dropdown-menu, .tooltip, .popover {
filter: invert(1) hue-rotate(180deg) !important;
background: #2d2d44 !important;
border-color: #444 !important;
}
/* Scrollbars */
::-webkit-scrollbar { background: #1a1a2e !important; }
::-webkit-scrollbar-thumb { background: #444 !important; }
::-webkit-scrollbar-thumb:hover { background: #555 !important; }
/* Selection */
::selection { background: #4ecdc4 !important; color: #1a1a2e !important; }
::-moz-selection { background: #4ecdc4 !important; color: #1a1a2e !important; }
';
}
function removeDarkMode() {
var style = document.getElementById('universal-dark-mode-style');
if (style) style.remove();
}
// Toggle with Alt+Shift+D
document.addEventListener('keydown', function(e) {
if (e.altKey && e.shiftKey && e.key === 'D') {
e.preventDefault();
enabled = !enabled;
if (enabled) {
applyDarkMode();
console.log('[Universal Dark Mode] Enabled');
} else {
removeDarkMode();
console.log('[Universal Dark Mode] Disabled');
}
}
});
// Apply on load
applyDarkMode();
// Re-apply on dynamic content
var observer = new MutationObserver(function(mutations) {
if (enabled && !document.getElementById('universal-dark-mode-style')) {
applyDarkMode();
}
});
observer.observe(document.head, { childList: true });
console.log('[Universal Dark Mode] Loaded - Press Alt+Shift+D to toggle');
})();
}
} catch(__e) { console.warn('[Userscript:Universal Dark Mode]', __e); }
})();
})();
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor by hxbai · Pull Request #1978 · NVIDIA/TransformerEngine · GitHub
Skip to content
[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor#1978

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

 Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and
privacy statement. We’ll occasionally send you account related emails.

Already on GitHub?
Sign in
to your account

 Merged

timmoon10
merged 9 commits into

from

Aug 6, 2025
 Merged


[PyTorch] fix input_quantizer usage for save_original_input; fix blockwise FP8 convert_and_update_tensor
#1978
Show file tree
Hide file tree
 Changes from all commits
Commits

Show all changes


9 commits


Select commit
Hold shift + click to select a range
36854f6

fix input_quantizer in save_original_input bwd


hxbai Jul 22, 2025
5607756

Merge branch 'main' into save_original_fix


timmoon10 Jul 22, 2025
1d17d72

fix get shape of blockwise tensor with only compact colwise data


hxbai Jul 23, 2025
368117b

Merge remote-tracking branch 'origin/main' into save_original_fix


hxbai Jul 31, 2025
39e9052

Merge branch 'main' into save_original_fix


hxbai Aug 1, 2025
6ce6a4e

fix blockwise FP8 convert_and_update_tensor


hxbai Aug 1, 2025
2c43099

[pre-commit.ci] auto fixes from pre-commit.com hooks


pre-commit-ci[bot] Aug 1, 2025
bc58f10

Merge branch 'main' into save_original_fix


timmoon10 Aug 1, 2025
403d9e5

Merge branch 'main' into save_original_fix


ksivaman Aug 4, 2025

File filter

Filter by extension
Filter by extension
Conversations

Failed to load comments. 
Loading
Jump to

Jump to file


Failed to load files. 
Loading
 Diff view

 Diff view


2 changes: 1 addition & 1 deletion
tests/pytorch/test_float8blockwisetensor.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -219,7 +219,7 @@ def test_quantize_dequantize_compact_format(
 rowwise=True,
 columnwise=dq_columnwise,
 block_scaling_dim=block_scaling_dim,
 all_gather_usage=True,
 all_gather_usage=(block_scaling_dim == 1),
 )
 self._test_quantize_dequantize(
 quantizer=quantizer,
Expand Down

129 changes: 122 additions & 7 deletions
transformer_engine/pytorch/csrc/quantizer.cpp

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -671,13 +671,128 @@ std::pair<TensorWrapper, py::object> Float8BlockQuantizer::convert_and_update_te
 const DType dtype = tensor.attr("_fp8_dtype").cast<DType>();
 bool is_2D_scaled = tensor.attr("_is_2D_scaled").cast<bool>();

 // Check the data matches quantizer usages
 NVTE_CHECK(!tensor.attr("_rowwise_data").is_none() == rowwise_usage,
 "Float8BlockwiseQTensor does not match quantizer usages (has_rowwise_data=",
 !tensor.attr("_rowwise_data").is_none(), ", rowwise_usage=", rowwise_usage);
 NVTE_CHECK(!tensor.attr("_columnwise_data").is_none() == columnwise_usage,
 "Float8BlockwiseQTensor does not match quantizer usages (has_columnwise_data=",
 !tensor.attr("_columnwise_data").is_none(), ", columnwise_usage=", columnwise_usage);
 // Extract buffers from Python tensor
 auto get_tensor = [&tensor](const char* name) -> std::optional<at::Tensor> {
 auto attr_py = tensor.attr(name);
 if (attr_py.is_none()) {
 return std::nullopt;
 }
 return attr_py.cast<at::Tensor>();
 };
 auto rowwise_data = get_tensor("_rowwise_data");
 auto rowwise_scale_inv = get_tensor("_rowwise_scale_inv");
 auto columnwise_data = get_tensor("_columnwise_data");
 auto columnwise_scale_inv = get_tensor("_columnwise_scale_inv");
 NVTE_CHECK(rowwise_data || columnwise_data, "FP8BlockwiseTensor has no data.");

 // Tensor options and dimensions
 at::TensorOptions opts;
 at::TensorOptions scale_opts;
 opts = opts.dtype(torch::kUInt8).device(torch::kCUDA);
 scale_opts = scale_opts.dtype(torch::kFloat32).device(torch::kCUDA);

 auto get_columnwise_shape = [&columnwise_data](bool all_gather_usage) -> std::vector<size_t> {
 if (!columnwise_data) {
 return std::vector<size_t>();
 }
 if (all_gather_usage) {
 return getTensorShape(*columnwise_data);
 }
 std::vector<size_t> shape = getTensorShape(*columnwise_data);
 std::vector<size_t> shape_transposed(shape.size());
 for (size_t i = 0; i + 1 < shape.size(); ++i) {
 shape_transposed[i] = shape[i + 1];
 }
 if (shape.size() > 0) {
 shape_transposed[shape.size() - 1] = shape[0];
 }
 return shape_transposed;
 };
 std::vector<size_t> shape;
 if (rowwise_data) {
 shape = getTensorShape(*rowwise_data);
 if (columnwise_data) {
 auto expected_shape = get_columnwise_shape(all_gather_usage);
 NVTE_CHECK(shape == expected_shape, "BlockwiseFP8 row-wise data (shape=", shape,
 ") and column-wise data (shape=", expected_shape, ") do not match");
 }
 } else {
 shape = get_columnwise_shape(all_gather_usage);
 }
 std::vector<int64_t> torch_shape;
 for (auto s : shape) {
 torch_shape.emplace_back(static_cast<int64_t>(s));
 }

 // Coerce row-wise data
 if (rowwise_usage) {
 if (!rowwise_data) {
 rowwise_data = at::empty(torch_shape, opts);
 tensor.attr("_rowwise_data") = *rowwise_data;
 }
 if (!rowwise_scale_inv) {
 auto scale_shape = get_scale_shape(shape, false);
 size_t sinv0 = scale_shape[0];
 size_t sinv1 = scale_shape[1];
 rowwise_scale_inv =
 at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
 tensor.attr("_rowwise_scale_inv") = *rowwise_scale_inv;
 }
 } else { // rowwise_usage == false
 if (rowwise_data) {
 rowwise_data.reset();
 tensor.attr("_rowwise_data") = py::none();
 }
 if (rowwise_scale_inv) {
 rowwise_scale_inv.reset();
 tensor.attr("_rowwise_scale_inv") = py::none();
 }
 }

 // Coerce column-wise data
 if (columnwise_usage) {
 std::vector<size_t> columnwise_shape;
 std::vector<int64_t> torch_columnwise_shape;
 if (torch_shape.size() > 0) {
 if (!all_gather_usage) {
 torch_columnwise_shape.reserve(torch_shape.size());
 columnwise_shape.reserve(shape.size());
 torch_columnwise_shape.push_back(torch_shape[torch_shape.size() - 1]);
 columnwise_shape.push_back(shape[shape.size() - 1]);
 for (size_t i = 0; i < torch_shape.size() - 1; ++i) {
 torch_columnwise_shape.push_back(torch_shape[i]);
 columnwise_shape.push_back(shape[i]);
 }
 } else {
 // assert we are doing 1D scaling
 NVTE_CHECK(block_scaling_dim == 1,
 "Compact columnwise format is not supported for 128x128 2D block scaling.");
 torch_columnwise_shape = torch_shape;
 columnwise_shape = shape;
 }
 }
 if (!columnwise_data) {
 columnwise_data = at::empty(torch_columnwise_shape, opts);
 tensor.attr("_columnwise_data") = *columnwise_data;
 }
 if (!columnwise_scale_inv) {
 auto scale_shape = get_scale_shape(shape, true);
 size_t sinv0 = scale_shape[0];
 size_t sinv1 = scale_shape[1];
 columnwise_scale_inv =
 at::empty({static_cast<int64_t>(sinv0), static_cast<int64_t>(sinv1)}, scale_opts);
 tensor.attr("_columnwise_scale_inv") = *columnwise_scale_inv;
 }
 } else { // columnwise_usage == false
 if (columnwise_data) {
 columnwise_data.reset();
 tensor.attr("_columnwise_data") = py::none();
 }
 if (columnwise_scale_inv) {
 columnwise_scale_inv.reset();
 tensor.attr("_columnwise_scale_inv") = py::none();
 }
 }

 auto ret = TensorWrapper(is_2D_scaled ? NVTE_BLOCK_SCALING_2D : NVTE_BLOCK_SCALING_1D);

Expand Down

11 changes: 6 additions & 5 deletions
transformer_engine/pytorch/module/linear.py

View file


Open in desktop

Original file line numberDiff line numberDiff line change
Expand Up@@ -589,13 +589,14 @@ def backward(ctx, grad_output: torch.Tensor) -> Tuple[Union[torch.Tensor, None],
 else:
 # Quantize input tensor
 quantizer = ctx.input_quantizer
 if ctx.backward_input_needs_gather and isinstance(
 quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)
 ):
 if isinstance(quantizer, (Float8Quantizer, Float8CurrentScalingQuantizer)):
 # All-gather is not supported with FP8 column-wise data
 quantizer.set_usage(rowwise=True, columnwise=False)
 quantizer.set_usage(
 rowwise=True,
 columnwise=not ctx.backward_input_needs_gather,
 )
 else:
 quantizer.set_usage(rowwise=True, columnwise=True)
 quantizer.set_usage(rowwise=False, columnwise=True)
 inputmat = quantizer(inputmat)
 else:
 if isinstance(inputmat, QuantizedTensorBase):
Expand Down