bridge.recipes.nemotron_vl.h100.nemotron_nano_v2_vl#

Nemotron Nano V2 VL finetuning recipes with parameterless defaults.

This module provides SFT and PEFT configurations for Nemotron Nano V2 VL 12B.

Module Contents#

Functions#

_nemotron_vl_target_modules

Return adapter target modules for the Nemotron VL PEFT recipe.

_build_nemotron_vl_peft

Build the fixed Nemotron VL LoRA recipe.

nemotron_nano_v2_vl_12b_sft_4gpu_h100_bf16_config

Return a full SFT config for Nemotron Nano V2 VL 12B.

nemotron_nano_v2_vl_12b_peft_2gpu_h100_bf16_config

Return a PEFT config for Nemotron Nano V2 VL 12B.

Data#

_DEFAULT_HF_MODEL_PATH

_ALL_COMPONENT_LORA_TARGET_MODULES

__all__

API#

bridge.recipes.nemotron_vl.h100.nemotron_nano_v2_vl._DEFAULT_HF_MODEL_PATH#

‘nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16’

bridge.recipes.nemotron_vl.h100.nemotron_nano_v2_vl._ALL_COMPONENT_LORA_TARGET_MODULES#

[‘linear_qkv’, ‘linear_proj’, ‘linear_fc1’, ‘linear_fc2’]

bridge.recipes.nemotron_vl.h100.nemotron_nano_v2_vl._nemotron_vl_target_modules() → list[str]#

Return adapter target modules for the Nemotron VL PEFT recipe.

bridge.recipes.nemotron_vl.h100.nemotron_nano_v2_vl._build_nemotron_vl_peft() → megatron.bridge.peft.lora.VLMLoRA#

Build the fixed Nemotron VL LoRA recipe.

bridge.recipes.nemotron_vl.h100.nemotron_nano_v2_vl.nemotron_nano_v2_vl_12b_sft_4gpu_h100_bf16_config() → megatron.bridge.training.config.ConfigContainer#

Return a full SFT config for Nemotron Nano V2 VL 12B.

Default configuration: 4 GPUs

  • TP=4, PP=1

  • LR=1e-5 (finetune default)

  • Sequence length: 4096

bridge.recipes.nemotron_vl.h100.nemotron_nano_v2_vl.nemotron_nano_v2_vl_12b_peft_2gpu_h100_bf16_config(
peft_scheme: str | megatron.bridge.peft.base.PEFT = 'lora',
) → megatron.bridge.training.config.ConfigContainer#

Return a PEFT config for Nemotron Nano V2 VL 12B.

Default configuration: 2 GPUs

  • TP=2, PP=1

  • LR=5e-5 (PEFT)

  • Sequence length: 4096

Parameters:

peft_scheme – PEFT scheme - “lora”, “dora”, or a custom PEFT instance. Note: Default uses VLMLoRA targeting all model components.

bridge.recipes.nemotron_vl.h100.nemotron_nano_v2_vl.__all__#

[‘nemotron_nano_v2_vl_12b_peft_2gpu_h100_bf16_config’, ‘nemotron_nano_v2_vl_12b_sft_4gpu_h100_bf16_c…