Add w8a8 quantized matmul torchxla wrapper #9290
vanbasten23
merged
15 commits into
master
from
xiowei/add_w8a8_quantized_matmul_torchxla_wrapper
added the tests and the quantized matmul kernel. All test passed.
b6f5d4b7
added tuned table and unit test
f1c96aa2
fix the test
3609fa73
fix a test
b0d3aa43
convert scalar type to f32
6adfdd9b
fix linter
ad184f42
add vmem_limit_bytes to static_argnames
f699c0cf
removed int4_weight
317c7282
fix comments
6836baee
vanbasten23
changed the base branch from
master
to
xiowei/add_w8a8_quantized_matmul_kernel_v5e
114 days ago
fix comments
a48e36d3
add non-dynamo wrapper. Failing with vmem_limit is a tracer.
9c04c2bc
non-dynamo case passed.
b4323593
all tests pass.
2f9af5c2
rebased
277d4b7d
yaochengji
approved these changes
on 2025-06-04
vanbasten23
changed the base branch from
xiowei/add_w8a8_quantized_matmul_kernel_v5e
to
master
113 days ago
Merge branch 'master' into xiowei/add_w8a8_quantized_matmul_torchxla_…
8637b50a
Login to write a write a comment.
Login via GitHub