bridge.recipes.gemma3_vl.gemma3_vl#

Gemma3-VL finetuning recipes with parameterless API.

This module provides SFT and PEFT configurations for Gemma3-VL models (4B, 12B, 27B).

Module Contents#

Functions#

gemma3_vl_4b_sft_config

Return a full SFT config for Gemma3-VL 4B Instruct.

gemma3_vl_12b_sft_config

Return a full SFT config for Gemma3-VL 12B Instruct.

gemma3_vl_27b_sft_config

Return a full SFT config for Gemma3-VL 27B Instruct.

gemma3_vl_4b_peft_config

Return a PEFT config for Gemma3-VL 4B Instruct.

gemma3_vl_12b_peft_config

Return a PEFT config for Gemma3-VL 12B Instruct.

gemma3_vl_27b_peft_config

Return a PEFT config for Gemma3-VL 27B Instruct.

API#

bridge.recipes.gemma3_vl.gemma3_vl.gemma3_vl_4b_sft_config() → megatron.bridge.training.config.ConfigContainer#

Return a full SFT config for Gemma3-VL 4B Instruct.

Default configuration: 1 node, 8 GPUs

  • TP=1, PP=1

  • LR=5e-6 (full SFT)

  • Sequence length: 4096

bridge.recipes.gemma3_vl.gemma3_vl.gemma3_vl_12b_sft_config() → megatron.bridge.training.config.ConfigContainer#

Return a full SFT config for Gemma3-VL 12B Instruct.

Default configuration: 1 node, 8 GPUs

  • TP=4, PP=1

  • LR=5e-6 (full SFT)

  • Sequence length: 4096

bridge.recipes.gemma3_vl.gemma3_vl.gemma3_vl_27b_sft_config() → megatron.bridge.training.config.ConfigContainer#

Return a full SFT config for Gemma3-VL 27B Instruct.

Default configuration: 2 nodes, 16 GPUs total

  • TP=8, PP=2

  • LR=5e-6 (full SFT)

  • Sequence length: 4096

bridge.recipes.gemma3_vl.gemma3_vl.gemma3_vl_4b_peft_config(
peft_scheme: str | megatron.bridge.peft.base.PEFT = 'lora',
) → megatron.bridge.training.config.ConfigContainer#

Return a PEFT config for Gemma3-VL 4B Instruct.

Default configuration: 1 node, 8 GPUs

  • TP=1, PP=1

  • LR=1e-4 (PEFT)

  • Sequence length: 4096

Parameters:

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

bridge.recipes.gemma3_vl.gemma3_vl.gemma3_vl_12b_peft_config(
peft_scheme: str | megatron.bridge.peft.base.PEFT = 'lora',
) → megatron.bridge.training.config.ConfigContainer#

Return a PEFT config for Gemma3-VL 12B Instruct.

Default configuration: 1 node, 8 GPUs

  • TP=1, PP=1 (lower than SFT for PEFT)

  • LR=1e-4 (PEFT)

  • Sequence length: 4096

Parameters:

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

bridge.recipes.gemma3_vl.gemma3_vl.gemma3_vl_27b_peft_config(
peft_scheme: str | megatron.bridge.peft.base.PEFT = 'lora',
) → megatron.bridge.training.config.ConfigContainer#

Return a PEFT config for Gemma3-VL 27B Instruct.

Default configuration: 1 node, 8 GPUs

  • TP=4, PP=1 (lower than SFT for PEFT)

  • LR=1e-4 (PEFT)

  • Sequence length: 4096

Parameters:

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