Skip to content

Fix module-config tensor init: shape-bounded randint high, int-dtype randint, float-dtype xavier, bfloat16 default, int64 for index tensors #228

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
moriohara merged 6 commits into from
Aug 10, 2026
Merged

Fix module-config tensor init: shape-bounded randint high, int-dtype randint, float-dtype xavier, bfloat16 default, int64 for index tensors #228

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
2cd4e84
use xavier for initialization
kiszk Jul 22, 2026
e52c660
update YAML files
kiszk Jul 22, 2026
e870247
Fix handling of position_embeddings
kiszk Jul 22, 2026
ab353de
Fix comments
kiszk Jul 22, 2026
bdc3604
Merge branch 'main' into fix/module-config-tensor-init-defaults
moriohara Jul 23, 2026
d8f9bc6
Merge branch 'main' into fix/module-config-tensor-init-defaults
moriohara Aug 10, 2026
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
76 changes: 30 additions & 46 deletions tests/configs/module_tests/Ministral-3-14B-Instruct-2512.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,15 +46,15 @@ test_suite_config:
device: spyre
init: randint
init_args:
high: 10000
high: 1
inputs_embeds:
tensor:
shape: [1, 128, 4096]
stride: null
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randn
init: xavier
- args: []
kwargs:
attention_mask:
Expand All @@ -66,15 +66,15 @@ test_suite_config:
device: spyre
init: randint
init_args:
high: 10000
high: 1
inputs_embeds:
tensor:
shape: [1, 1, 4096]
stride: null
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randn
init: xavier
- name: Ministral3RotaryEmbedding_483e0b9c
module_path: transformers.models.ministral3.modeling_ministral3.Ministral3RotaryEmbedding
description: 'Module: transformers.models.ministral3.modeling_ministral3.Ministral3RotaryEmbedding'
Expand All @@ -97,7 +97,7 @@ test_suite_config:
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randn
init: xavier
kwargs:
position_ids:
tensor:
Expand All @@ -108,15 +108,15 @@ test_suite_config:
device: spyre
init: randint
init_args:
high: 10000
high: 1
- args:
- tensor:
shape: [1, 1, 4096]
stride: null
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randn
init: xavier
kwargs:
position_ids:
tensor:
Expand All @@ -127,7 +127,7 @@ test_suite_config:
device: spyre
init: randint
init_args:
high: 10000
high: 1
- name: Ministral3DecoderLayer_layer0
module_path: transformers.models.ministral3.modeling_ministral3.Ministral3DecoderLayer
description: 'Module: transformers.models.ministral3.modeling_ministral3.Ministral3DecoderLayer'
Expand All @@ -151,7 +151,7 @@ test_suite_config:
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randn
init: xavier
kwargs:
position_ids:
tensor:
Expand All @@ -162,33 +162,29 @@ test_suite_config:
device: spyre
init: randint
init_args:
high: 10000
high: 1
position_embeddings:
tensor_list:
- shape: [1, 128, 128]
stride: null
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randint
init_args:
high: 10000
init: xavier
- shape: [1, 128, 128]
stride: null
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randint
init_args:
high: 10000
init: xavier
- args:
- tensor:
shape: [1, 1, 4096]
stride: null
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randn
init: xavier
kwargs:
position_ids:
tensor:
Expand All @@ -199,25 +195,21 @@ test_suite_config:
device: spyre
init: randint
init_args:
high: 10000
high: 1
position_embeddings:
tensor_list:
- shape: [1, 1, 128]
stride: null
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randint
init_args:
high: 10000
init: xavier
- shape: [1, 1, 128]
stride: null
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randint
init_args:
high: 10000
init: xavier
- name: Ministral3RMSNorm_4096
module_path: transformers.models.ministral3.modeling_ministral3.Ministral3RMSNorm
description: 'Module: transformers.models.ministral3.modeling_ministral3.Ministral3RMSNorm'
Expand All @@ -233,7 +225,7 @@ test_suite_config:
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randn
init: xavier
kwargs: {}
- args:
- tensor:
Expand All @@ -242,7 +234,7 @@ test_suite_config:
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randn
init: xavier
kwargs: {}
- name: Ministral3Attention_layer0
module_path: transformers.models.ministral3.modeling_ministral3.Ministral3Attention
Expand All @@ -269,7 +261,7 @@ test_suite_config:
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randn
init: xavier
position_ids:
tensor:
shape: [1, 128]
Expand All @@ -279,25 +271,21 @@ test_suite_config:
device: spyre
init: randint
init_args:
high: 10000
high: 1
position_embeddings:
tensor_list:
- shape: [1, 128, 128]
stride: null
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randint
init_args:
high: 10000
init: xavier
- shape: [1, 128, 128]
stride: null
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randint
init_args:
high: 10000
init: xavier
- args: []
kwargs:
hidden_states:
Expand All @@ -307,7 +295,7 @@ test_suite_config:
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randn
init: xavier
position_ids:
tensor:
shape: [1, 1]
Expand All @@ -317,25 +305,21 @@ test_suite_config:
device: spyre
init: randint
init_args:
high: 10000
high: 1
position_embeddings:
tensor_list:
- shape: [1, 1, 128]
stride: null
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randint
init_args:
high: 10000
init: xavier
- shape: [1, 1, 128]
stride: null
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randint
init_args:
high: 10000
init: xavier
- name: Ministral3MLP_eb806b7d
module_path: transformers.models.ministral3.modeling_ministral3.Ministral3MLP
description: 'Module: transformers.models.ministral3.modeling_ministral3.Ministral3MLP'
Expand All @@ -358,7 +342,7 @@ test_suite_config:
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randn
init: xavier
kwargs: {}
- args:
- tensor:
Expand All @@ -367,7 +351,7 @@ test_suite_config:
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randn
init: xavier
kwargs: {}
- name: SiLUActivation_d2f532e9
module_path: transformers.activations.SiLUActivation
Expand All @@ -383,7 +367,7 @@ test_suite_config:
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randn
init: xavier
kwargs: {}
- args:
- tensor:
Expand All @@ -392,7 +376,7 @@ test_suite_config:
storage_offset: 0
dtype: torch.bfloat16
device: spyre
init: randn
init: xavier
kwargs: {}
- path: ${TORCH_DEVICE_ROOT}/tests/test_modules_custom.py
unlisted_test_mode: skip
Expand Down
Loading
Loading