bridge.recipes.qwen.qwen3

Module Contents

Functions

qwen3_600m_pretrain_config	Return a pre-training config for Qwen3 0.6B.
qwen3_1p7b_pretrain_config	Return a pre-training config for Qwen3 1.7B.
qwen3_4b_pretrain_config	Return a pre-training config for Qwen3 4B.
qwen3_8b_pretrain_config	Return a pre-training config for Qwen3 8B.
qwen3_14b_pretrain_config	Return a pre-training config for Qwen3 14B.
qwen3_32b_pretrain_config	Return a pre-training config for Qwen3 32B.
qwen3_600m_sft_config	Return a full SFT config for Qwen3 600M.
qwen3_1p7b_sft_config	Return a full SFT config for Qwen3 1.7B.
qwen3_4b_sft_config	Return a full SFT config for Qwen3 4B.
qwen3_8b_sft_config	Return a full SFT config for Qwen3 8B.
qwen3_14b_sft_config	Return a full SFT config for Qwen3 14B.
qwen3_32b_sft_config	Return a full SFT config for Qwen3 32B.
qwen3_600m_peft_config	Return a PEFT config for Qwen3 600M.
qwen3_1p7b_peft_config	Return a PEFT config for Qwen3 1.7B.
qwen3_4b_peft_config	Return a PEFT config for Qwen3 4B.
qwen3_8b_peft_config	Return a PEFT config for Qwen3 8B.
qwen3_14b_peft_config	Return a PEFT config for Qwen3 14B.
qwen3_32b_peft_config	Return a PEFT config for Qwen3 32B.

API

bridge.recipes.qwen.qwen3.qwen3_600m_pretrain_config() → megatron.bridge.training.config.ConfigContainer

Return a pre-training config for Qwen3 0.6B.

Recommended parallelism: TP=1, PP=1 (fits on a single GPU).

bridge.recipes.qwen.qwen3.qwen3_1p7b_pretrain_config() → megatron.bridge.training.config.ConfigContainer

Return a pre-training config for Qwen3 1.7B.

Recommended parallelism: TP=1, PP=1 (fits on a single GPU).

bridge.recipes.qwen.qwen3.qwen3_4b_pretrain_config() → megatron.bridge.training.config.ConfigContainer

Return a pre-training config for Qwen3 4B.

Recommended parallelism: TP=2, PP=1.

bridge.recipes.qwen.qwen3.qwen3_8b_pretrain_config() → megatron.bridge.training.config.ConfigContainer

Return a pre-training config for Qwen3 8B.

Recommended parallelism: TP=4, PP=1.

bridge.recipes.qwen.qwen3.qwen3_14b_pretrain_config() → megatron.bridge.training.config.ConfigContainer

Return a pre-training config for Qwen3 14B.

Recommended parallelism: TP=8, PP=1.

bridge.recipes.qwen.qwen3.qwen3_32b_pretrain_config() → megatron.bridge.training.config.ConfigContainer

Return a pre-training config for Qwen3 32B.

Recommended parallelism: TP=8, PP=2 with recompute enabled for memory optimization.

bridge.recipes.qwen.qwen3.qwen3_600m_sft_config() → megatron.bridge.training.config.ConfigContainer

Return a full SFT config for Qwen3 600M.

Recommended parallelism: TP=1, PP=1 (1 node, 8 GPUs)

bridge.recipes.qwen.qwen3.qwen3_1p7b_sft_config() → megatron.bridge.training.config.ConfigContainer

Return a full SFT config for Qwen3 1.7B.

Recommended parallelism: TP=1, PP=1 (1 node, 8 GPUs)

bridge.recipes.qwen.qwen3.qwen3_4b_sft_config() → megatron.bridge.training.config.ConfigContainer

Return a full SFT config for Qwen3 4B.

Recommended parallelism: TP=2, PP=1 (1 node, 8 GPUs)

bridge.recipes.qwen.qwen3.qwen3_8b_sft_config() → megatron.bridge.training.config.ConfigContainer

Return a full SFT config for Qwen3 8B.

Recommended parallelism: TP=4, PP=1 (1 node, 8 GPUs)

bridge.recipes.qwen.qwen3.qwen3_14b_sft_config() → megatron.bridge.training.config.ConfigContainer

Return a full SFT config for Qwen3 14B.

Recommended parallelism: TP=8, PP=1 (1 node, 8 GPUs)

bridge.recipes.qwen.qwen3.qwen3_32b_sft_config() → megatron.bridge.training.config.ConfigContainer

Return a full SFT config for Qwen3 32B.

Recommended parallelism: TP=8, PP=2 (2 nodes, 16 GPUs total) Includes recompute for memory optimization.

bridge.recipes.qwen.qwen3.qwen3_600m_peft_config(

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

) → megatron.bridge.training.config.ConfigContainer

Return a PEFT config for Qwen3 600M.

Parameters:

peft_scheme – PEFT scheme - ‘lora’, ‘dora’, or a PEFT instance. Default: ‘lora’

Recommended parallelism: TP=1, PP=1 (1 node, 8 GPUs)

bridge.recipes.qwen.qwen3.qwen3_1p7b_peft_config(

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

) → megatron.bridge.training.config.ConfigContainer

Return a PEFT config for Qwen3 1.7B.

Parameters:

peft_scheme – PEFT scheme - ‘lora’, ‘dora’, or a PEFT instance. Default: ‘lora’

Recommended parallelism: TP=1, PP=1 (1 node, 8 GPUs)

bridge.recipes.qwen.qwen3.qwen3_4b_peft_config(

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

) → megatron.bridge.training.config.ConfigContainer

Return a PEFT config for Qwen3 4B.

Parameters:

peft_scheme – PEFT scheme - ‘lora’, ‘dora’, or a PEFT instance. Default: ‘lora’

Recommended parallelism: TP=1, PP=1 (1 node, 8 GPUs)

bridge.recipes.qwen.qwen3.qwen3_8b_peft_config(

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

) → megatron.bridge.training.config.ConfigContainer

Return a PEFT config for Qwen3 8B.

Parameters:

peft_scheme – PEFT scheme - ‘lora’, ‘dora’, or a PEFT instance. Default: ‘lora’

Recommended parallelism: TP=1, PP=1 (1 node, 8 GPUs)

bridge.recipes.qwen.qwen3.qwen3_14b_peft_config(

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

) → megatron.bridge.training.config.ConfigContainer

Return a PEFT config for Qwen3 14B.

Parameters:

peft_scheme – PEFT scheme - ‘lora’, ‘dora’, or a PEFT instance. Default: ‘lora’

Recommended parallelism: TP=1, PP=1 (1 node, 8 GPUs)

bridge.recipes.qwen.qwen3.qwen3_32b_peft_config(

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

) → megatron.bridge.training.config.ConfigContainer

Return a PEFT config for Qwen3 32B.

Parameters:

peft_scheme – PEFT scheme - ‘lora’, ‘dora’, or a PEFT instance. Default: ‘lora’

Recommended parallelism: TP=1, PP=1 (1 node, 8 GPUs) Includes recompute for memory optimization.