Skip to content

Commit a42250d

Browse files
committed
Polish model-specific recipe YAML formatting
- Drop unnecessary single quotes around `$import` keys in the generator-produced nvfp4-kv_fp8_cast recipes; `$` is not special in YAML and the hand-written recipes already use the unquoted form. - Bump SPDX copyright year from 2024 to 2026 on the new YAML files to match the repository's LICENSE_HEADER. - Insert a blank line between the `imports:` block and the following `metadata:` section to match the hand-written recipe style. Signed-off-by: Shengliang Xu <shengliangx@nvidia.com>
1 parent 44970f8 commit a42250d

7 files changed

Lines changed: 17 additions & 15 deletions

File tree

modelopt_recipes/huggingface/gemma/ptq/int8_sq-kv_fp8_cast.yaml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
1+
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
22
# SPDX-License-Identifier: Apache-2.0
33
#
44
# Licensed under the Apache License, Version 2.0 (the "License");

modelopt_recipes/huggingface/gemma/ptq/w4a8_awq-kv_fp8_cast.yaml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
1+
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
22
# SPDX-License-Identifier: Apache-2.0
33
#
44
# Licensed under the Apache License, Version 2.0 (the "License");

modelopt_recipes/huggingface/mpt/ptq/w4a8_awq-kv_fp8_cast.yaml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
1+
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
22
# SPDX-License-Identifier: Apache-2.0
33
#
44
# Licensed under the Apache License, Version 2.0 (the "License");

modelopt_recipes/huggingface/nemotron_vl/ptq/disabled_quantizers.yaml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
1+
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
22
# SPDX-License-Identifier: Apache-2.0
33
#
44
# Licensed under the Apache License, Version 2.0 (the "License");

modelopt_recipes/huggingface/nemotron_vl/ptq/nvfp4-kv_fp8_cast.yaml

Lines changed: 6 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
1+
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
22
# SPDX-License-Identifier: Apache-2.0
33
#
44
# Licensed under the Apache License, Version 2.0 (the "License");
@@ -22,14 +22,15 @@ imports:
2222
w4a4_nvfp4_nvfp4: configs/ptq/units/w4a4_nvfp4_nvfp4
2323
disabled_quantizers: huggingface/nemotron_vl/ptq/disabled_quantizers
2424
kv_fp8_cast: configs/ptq/units/kv_fp8_cast
25+
2526
metadata:
2627
recipe_type: ptq
2728
description: 'Nemotron VL PTQ recipe (nvfp4): same numerics as the general nvfp4 preset, applied to the decoder (text generation) component only (vision/encoder
2829
branches are skipped).'
2930
quantize:
3031
algorithm: max
3132
quant_cfg:
32-
- '$import': base_disable_all
33-
- '$import': w4a4_nvfp4_nvfp4
34-
- '$import': kv_fp8_cast
35-
- '$import': disabled_quantizers
33+
- $import: base_disable_all
34+
- $import: w4a4_nvfp4_nvfp4
35+
- $import: kv_fp8_cast
36+
- $import: disabled_quantizers

modelopt_recipes/huggingface/phi4mm/ptq/disabled_quantizers.yaml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
1+
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
22
# SPDX-License-Identifier: Apache-2.0
33
#
44
# Licensed under the Apache License, Version 2.0 (the "License");

modelopt_recipes/huggingface/phi4mm/ptq/nvfp4-kv_fp8_cast.yaml

Lines changed: 6 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
1-
# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
1+
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
22
# SPDX-License-Identifier: Apache-2.0
33
#
44
# Licensed under the Apache License, Version 2.0 (the "License");
@@ -22,14 +22,15 @@ imports:
2222
w4a4_nvfp4_nvfp4: configs/ptq/units/w4a4_nvfp4_nvfp4
2323
disabled_quantizers: huggingface/phi4mm/ptq/disabled_quantizers
2424
kv_fp8_cast: configs/ptq/units/kv_fp8_cast
25+
2526
metadata:
2627
recipe_type: ptq
2728
description: 'Phi-4-Multimodal PTQ recipe (nvfp4): same numerics as the general nvfp4 preset, applied to the language model only (speech, audio, image,
2829
and vision branches are skipped).'
2930
quantize:
3031
algorithm: max
3132
quant_cfg:
32-
- '$import': base_disable_all
33-
- '$import': w4a4_nvfp4_nvfp4
34-
- '$import': kv_fp8_cast
35-
- '$import': disabled_quantizers
33+
- $import: base_disable_all
34+
- $import: w4a4_nvfp4_nvfp4
35+
- $import: kv_fp8_cast
36+
- $import: disabled_quantizers

0 commit comments

Comments
 (0)