Skip to content

[JAX] Fix test_layer to support fused attention and adjust test encoder tolerance to account for minor diff #2563

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
jberchtold-nvidia merged 1 commit into from
Jan 6, 2026
Merged

[JAX] Fix test_layer to support fused attention and adjust test encoder tolerance to account for minor diff #2563

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
8 changes: 4 additions & 4 deletions examples/jax/encoder/test_model_parallel_encoder.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -503,7 +503,7 @@ def test_te_delayed_scaling_fp8(self):
self.args.use_fp8 = True
self.args.fp8_recipe = "DelayedScaling"
actual = train_and_evaluate(self.args)
assert actual[0] < 0.361 and actual[1] > 0.84
assert actual[0] < 0.362 and actual[1] > 0.84

@unittest.skipIf(not is_mxfp8_supported, mxfp8_reason)
def test_te_mxfp8(self):
Expand Down Expand Up @@ -535,7 +535,7 @@ def test_te_delayed_scaling_fp8_with_sp(self):
self.args.use_fp8 = True
self.args.fp8_recipe = "DelayedScaling"
actual = train_and_evaluate(self.args)
assert actual[0] < 0.361 and actual[1] > 0.84
assert actual[0] < 0.362 and actual[1] > 0.84

@unittest.skipIf(not is_mxfp8_supported, mxfp8_reason)
def test_te_mxfp8_with_sp(self):
Expand Down Expand Up @@ -569,7 +569,7 @@ def test_te_delayed_scaling_fp8_shardy(self):
self.args.use_fp8 = True
self.args.fp8_recipe = "DelayedScaling"
actual = train_and_evaluate(self.args)
assert actual[0] < 0.361 and actual[1] > 0.84
assert actual[0] < 0.362 and actual[1] > 0.84

@unittest.skipIf(not is_fp8_supported, fp8_reason)
def test_te_delayed_scaling_fp8_with_sp_shardy(self):
Expand All @@ -579,7 +579,7 @@ def test_te_delayed_scaling_fp8_with_sp_shardy(self):
self.args.use_fp8 = True
self.args.fp8_recipe = "DelayedScaling"
actual = train_and_evaluate(self.args)
assert actual[0] < 0.361 and actual[1] > 0.84
assert actual[0] < 0.362 and actual[1] > 0.84

@unittest.skipIf(not is_mxfp8_supported, mxfp8_reason)
def test_te_mxfp8_shardy(self):
Expand Down
9 changes: 9 additions & 0 deletions tests/jax/test_layer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -430,6 +430,9 @@ class EncoderRunner(BaseRunner):
"attention/DotProductAttention_0/_UnfusedDotProductAttention_0/softmax_offset": (
"attention/DotProductAttention_0/softmax_offset"
),
"attention/DotProductAttention_0/_FusedDotProductAttention_0/softmax_offset": (
"attention/DotProductAttention_0/softmax_offset"
),
"mlp/wi_kernel": "mlp/wi/kernel",
"mlp/wi_bias": "mlp/wi/bias",
"mlp/wo_kernel": "mlp/wo/kernel",
Expand Down Expand Up @@ -478,13 +481,19 @@ class DecoderRunner(BaseRunner):
"encoder_decoder_attention/DotProductAttention_0/_UnfusedDotProductAttention_0/softmax_offset": (
"encoder_decoder_attention/DotProductAttention_0/softmax_offset"
),
"encoder_decoder_attention/DotProductAttention_0/_FusedDotProductAttention_0/softmax_offset": (
"encoder_decoder_attention/DotProductAttention_0/softmax_offset"
),
"self_attention/qkv/scale": "pre_self_attention_layer_norm/scale",
"self_attention/qkv/ln_bias": "pre_self_attention_layer_norm/ln_bias",
"self_attention/query/scale": "pre_self_attention_layer_norm/scale",
"self_attention/query/ln_bias": "pre_self_attention_layer_norm/ln_bias",
"self_attention/DotProductAttention_0/_UnfusedDotProductAttention_0/softmax_offset": (
"self_attention/DotProductAttention_0/softmax_offset"
),
"self_attention/DotProductAttention_0/_FusedDotProductAttention_0/softmax_offset": (
"self_attention/DotProductAttention_0/softmax_offset"
),
"mlp/wi_kernel": "mlp/wi/kernel",
"mlp/wi_bias": "mlp/wi/bias",
"mlp/wo_kernel": "mlp/wo/kernel",
Expand Down
Loading