`bridge.recipes.gemma.gemma3`#

Module Contents#

Functions#

`gemma3_1b_pretrain_config`	Return a pre-training config for Gemma3 1B.
`_adjust_gemma3_vocab_size`	Adjust vocab size for Gemma3 (model vocab < tokenizer vocab).
`gemma3_1b_sft_config`	Return a full SFT config for Gemma3 1B.
`gemma3_1b_peft_config`	Return a PEFT config for Gemma3 1B.

Data#

`SEQUENCE_LENGTH_32K`
`SEQUENCE_LENGTH_128K`

API#

bridge.recipes.gemma.gemma3.SEQUENCE_LENGTH_32K: int#: 32768

bridge.recipes.gemma.gemma3.SEQUENCE_LENGTH_128K: int#: 131072

bridge.recipes.gemma.gemma3.gemma3_1b_pretrain_config() → megatron.bridge.training.config.ConfigContainer#

Return a pre-training config for Gemma3 1B.

Default parallelism: TP=1, PP=1, seq_length=32K

bridge.recipes.gemma.gemma3._adjust_gemma3_vocab_size(model_cfg, hf_path: str)#

Adjust vocab size for Gemma3 (model vocab < tokenizer vocab).

Note: This requires HuggingFace authentication for Gemma3 models. If the tokenizer cannot be loaded, the vocab size adjustment is skipped.

bridge.recipes.gemma.gemma3.gemma3_1b_sft_config() → megatron.bridge.training.config.ConfigContainer#

Return a full SFT config for Gemma3 1B.

Default parallelism: TP=1, PP=1

Returns:: ConfigContainer with all settings pre-configured for Gemma3 1B SFT.

bridge.recipes.gemma.gemma3.gemma3_1b_peft_config( peft_scheme: str | megatron.bridge.peft.base.PEFT = 'lora', ) → megatron.bridge.training.config.ConfigContainer#

Return a PEFT config for Gemma3 1B.