bridge.recipes.gpt_oss.gpt_oss

Module Contents

Functions

_enable_gpt_oss_hopper_fp8_current_scaling	Enable Hopper FP8 current scaling for GPT-OSS recipes.
gpt_oss_20b_pretrain_config	Return a pre-training config for GPT-OSS 20B variant.
gpt_oss_120b_pretrain_config	Return a pre-training config for GPT-OSS 120B variant.
gpt_oss_20b_pretrain_fp8_current_scaling_config	Return a pre-training config for GPT-OSS 20B with Hopper FP8 current scaling.
gpt_oss_20b_sft_config	Return a full SFT config for GPT-OSS 20B.
gpt_oss_120b_sft_config	Return a full SFT config for GPT-OSS 120B.
gpt_oss_20b_sft_fp8_current_scaling_config	Return a full SFT config for GPT-OSS 20B with Hopper FP8 current scaling.
gpt_oss_20b_peft_config	Return a PEFT config for GPT-OSS 20B.
gpt_oss_120b_peft_config	Return a PEFT config for GPT-OSS 120B.
gpt_oss_20b_peft_fp8_current_scaling_config	Return a PEFT config for GPT-OSS 20B with Hopper FP8 current scaling.
_enable_gpt_oss_blackwell_mxfp8	Enable Blackwell MXFP8 for GPT-OSS recipes.
gpt_oss_20b_pretrain_mxfp8_config	Return a pre-training config for GPT-OSS 20B with Blackwell MXFP8.
gpt_oss_20b_sft_mxfp8_config	Return a full SFT config for GPT-OSS 20B with Blackwell MXFP8.
gpt_oss_20b_peft_mxfp8_config	Return a PEFT config for GPT-OSS 20B with Blackwell MXFP8.

API

bridge.recipes.gpt_oss.gpt_oss._enable_gpt_oss_hopper_fp8_current_scaling(

cfg: megatron.bridge.training.config.ConfigContainer,

) → megatron.bridge.training.config.ConfigContainer

Enable Hopper FP8 current scaling for GPT-OSS recipes.

bridge.recipes.gpt_oss.gpt_oss.gpt_oss_20b_pretrain_config() → megatron.bridge.training.config.ConfigContainer

Return a pre-training config for GPT-OSS 20B variant.

Recommended parallelism: TP=2, PP=4, EP=4

bridge.recipes.gpt_oss.gpt_oss.gpt_oss_120b_pretrain_config() → megatron.bridge.training.config.ConfigContainer

Return a pre-training config for GPT-OSS 120B variant.

Recommended parallelism: TP=2, PP=4, EP=16

bridge.recipes.gpt_oss.gpt_oss.gpt_oss_20b_pretrain_fp8_current_scaling_config() → megatron.bridge.training.config.ConfigContainer

Return a pre-training config for GPT-OSS 20B with Hopper FP8 current scaling.

bridge.recipes.gpt_oss.gpt_oss.gpt_oss_20b_sft_config() → megatron.bridge.training.config.ConfigContainer

Return a full SFT config for GPT-OSS 20B.

Default parallelism: TP=1, PP=1, EP=8

Returns:

ConfigContainer with all settings pre-configured for GPT-OSS 20B SFT.

bridge.recipes.gpt_oss.gpt_oss.gpt_oss_120b_sft_config() → megatron.bridge.training.config.ConfigContainer

Return a full SFT config for GPT-OSS 120B.

Default parallelism: TP=1, PP=4, EP=8

Returns:

ConfigContainer with all settings pre-configured for GPT-OSS 120B SFT.

bridge.recipes.gpt_oss.gpt_oss.gpt_oss_20b_sft_fp8_current_scaling_config() → megatron.bridge.training.config.ConfigContainer

Return a full SFT config for GPT-OSS 20B with Hopper FP8 current scaling.

bridge.recipes.gpt_oss.gpt_oss.gpt_oss_20b_peft_config(

peft_scheme: str | megatron.bridge.peft.base.PEFT = 'lora',

) → megatron.bridge.training.config.ConfigContainer

Return a PEFT config for GPT-OSS 20B.

Default parallelism: TP=1, PP=1, EP=1

Parameters:

peft_scheme – PEFT scheme - “lora”, “dora”, or a custom PEFT instance.

Returns:

ConfigContainer with all settings pre-configured for GPT-OSS 20B PEFT.

bridge.recipes.gpt_oss.gpt_oss.gpt_oss_120b_peft_config(

peft_scheme: str | megatron.bridge.peft.base.PEFT = 'lora',

) → megatron.bridge.training.config.ConfigContainer

Return a PEFT config for GPT-OSS 120B.

Default parallelism: TP=1, PP=1, EP=8

Parameters:

peft_scheme – PEFT scheme - “lora”, “dora”, or a custom PEFT instance.

Returns:

ConfigContainer with all settings pre-configured for GPT-OSS 120B PEFT.

bridge.recipes.gpt_oss.gpt_oss.gpt_oss_20b_peft_fp8_current_scaling_config(

peft_scheme: str | megatron.bridge.peft.base.PEFT = 'lora',

) → megatron.bridge.training.config.ConfigContainer

Return a PEFT config for GPT-OSS 20B with Hopper FP8 current scaling.

bridge.recipes.gpt_oss.gpt_oss._enable_gpt_oss_blackwell_mxfp8(

cfg: megatron.bridge.training.config.ConfigContainer,

) → megatron.bridge.training.config.ConfigContainer

Enable Blackwell MXFP8 for GPT-OSS recipes.

bridge.recipes.gpt_oss.gpt_oss.gpt_oss_20b_pretrain_mxfp8_config() → megatron.bridge.training.config.ConfigContainer

Return a pre-training config for GPT-OSS 20B with Blackwell MXFP8.

bridge.recipes.gpt_oss.gpt_oss.gpt_oss_20b_sft_mxfp8_config() → megatron.bridge.training.config.ConfigContainer

Return a full SFT config for GPT-OSS 20B with Blackwell MXFP8.

bridge.recipes.gpt_oss.gpt_oss.gpt_oss_20b_peft_mxfp8_config(

peft_scheme: str | megatron.bridge.peft.base.PEFT = 'lora',

) → megatron.bridge.training.config.ConfigContainer

Return a PEFT config for GPT-OSS 20B with Blackwell MXFP8.