Add GPU profiler#1997

Closed

jainapurva wants to merge 31 commits intomainfrom

gh/jainapurva/25/head

Commits on Apr 4, 2025

Commits on Apr 8, 2025

Reintroduce has_weight_zeros as a template param

metascroy
authored and
jainapurva
committed
Claen up op interface

metascroy
authored and
jainapurva
committed
quantized matmul

kimishpatel
authored and
jainapurva
committed
Allow builds on less than sm75 raise runtime failure (#1999)

drisspg
authored and
jainapurva
committed
Skip galore test if not cuda (#2003)

jerryzh168
authored and
jainapurva
committed
Fix experimental CI (#2005)

metascroy
authored and
jainapurva
committed
Add fp32xint8 matmul

kimishpatel
authored and
jainapurva
committed
Add quantized q @ k test for intented used in quantized attention

kimishpatel
authored and
jainapurva
committed
Update version.txt (#2009)

jerryzh168
authored and
jainapurva
committed
Initial prototype of differentiable _scaled_grouped_mm function (#1969)

danielvegamyhre
authored and
jainapurva
committed
Add quantized attn_scores @ v test for intented used in quantized attention

kimishpatel
authored and
jainapurva
committed
add fallback kernel and interface

kimishpatel
authored and
jainapurva
committed
Add fallback kernel and interface for rhs only quantized matmul

kimishpatel
authored and
jainapurva
committed
Add KleidiAI gemm kernels (#2000)

metascroy
authored and
jainapurva
committed
Update float8nocompile test code to use new float8 matmul function (#2013)

danielvegamyhre
authored and
jainapurva
committed
Remove float8nocompile CI (#1976)

danielvegamyhre
authored and
jainapurva
committed
Update clean_release_notes.py (#2014)

jerryzh168
authored and
jainapurva
committed
Match QAT prepare and convert numerics exactly (#1964)

andrewor14
authored and
jainapurva
committed
Skip failing tests for rowwise-scaled (#2022)

drisspg
authored and
jainapurva
committed
Update torchao.prototype.parq and add 4-bit Llama 3.2 1B benchmark (#2017)

lisjin
authored and
jainapurva
committed
Use quantized gemm only on aarch64

kimishpatel
authored and
jainapurva
committed
Adds utility to replace Q/DQ ops with torchao quantized linear ops (#1967)

metascroy
authored and
jainapurva
committed
Fix slice and padding for TensorCoreTiledLayout (#2015)

jerryzh168
authored and
jainapurva
committed
Fix Int4WeightEmbeddingQATQuantizer.convert path (#2024)

andrewor14
authored and
jainapurva
committed
Add gguf q4_k quantization (#2001)

jerryzh168
authored and
jainapurva
committed
torch/ao

gmagogsfm
authored and
jainapurva
committed
Adds Q/DQ layout support for embedding quantization with IntxWeightOnlyConfig (#1972)

metascroy
authored and
jainapurva
committed
Merge remote-tracking branch 'origin/main' into base-fix
jainapurva
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add GPU profiler#1997

Add GPU profiler#1997
jainapurva wants to merge 31 commits intomainfrom
gh/jainapurva/25/head

Commits on Apr 4, 2025

Add profiler and Perfetto UI link with comprehensive tests (#1984, #1992)

More models

Update

Commits on Apr 8, 2025

Reintroduce has_weight_zeros as a template param

Claen up op interface

quantized matmul

Allow builds on less than sm75 raise runtime failure (#1999)

Skip galore test if not cuda (#2003)

Fix experimental CI (#2005)

Add fp32xint8 matmul

Add quantized q @ k test for intented used in quantized attention

Update version.txt (#2009)

Initial prototype of differentiable _scaled_grouped_mm function (#1969)

Add quantized attn_scores @ v test for intented used in quantized attention

add fallback kernel and interface

Add fallback kernel and interface for rhs only quantized matmul

Add KleidiAI gemm kernels (#2000)

Update float8nocompile test code to use new float8 matmul function (#2013)

Remove float8nocompile CI (#1976)

Update clean_release_notes.py (#2014)

Match QAT prepare and convert numerics exactly (#1964)

Skip failing tests for rowwise-scaled (#2022)

Update torchao.prototype.parq and add 4-bit Llama 3.2 1B benchmark (#2017)

Use quantized gemm only on aarch64

Adds utility to replace Q/DQ ops with torchao quantized linear ops (#1967)

Fix slice and padding for TensorCoreTiledLayout (#2015)

Fix Int4WeightEmbeddingQATQuantizer.convert path (#2024)

Add gguf q4_k quantization (#2001)

torch/ao

Adds Q/DQ layout support for embedding quantization with IntxWeightOnlyConfig (#1972)

Merge remote-tracking branch 'origin/main' into base-fix