bridge.recipes.qwen_vl.h100.qwen25_vl#

Qwen2.5-VL finetuning recipes with parameterless API.

This module provides SFT and PEFT configurations for Qwen2.5-VL models (3B, 7B, 32B, 72B).

Module Contents#

Functions#

qwen25_vl_3b_sft_1gpu_h100_bf16_config

Return a full SFT config for Qwen2.5-VL 3B Instruct.

qwen25_vl_7b_sft_2gpu_h100_bf16_config

Return a full SFT config for Qwen2.5-VL 7B Instruct.

qwen25_vl_32b_sft_16gpu_h100_bf16_config

Return a full SFT config for Qwen2.5-VL 32B Instruct.

qwen25_vl_72b_sft_32gpu_h100_bf16_config

Return a full SFT config for Qwen2.5-VL 72B Instruct.

qwen25_vl_3b_peft_1gpu_h100_bf16_config

Return a PEFT config for Qwen2.5-VL 3B Instruct.

qwen25_vl_7b_peft_1gpu_h100_bf16_config

Return a PEFT config for Qwen2.5-VL 7B Instruct.

qwen25_vl_32b_peft_1gpu_h100_bf16_config

Return a PEFT config for Qwen2.5-VL 32B Instruct.

qwen25_vl_72b_peft_1gpu_h100_bf16_config

Return a PEFT config for Qwen2.5-VL 72B Instruct.

Data#

__all__

API#

bridge.recipes.qwen_vl.h100.qwen25_vl.qwen25_vl_3b_sft_1gpu_h100_bf16_config() → megatron.bridge.training.config.ConfigContainer#

Return a full SFT config for Qwen2.5-VL 3B Instruct.

Default configuration: 1 GPU

  • TP=1, PP=1

  • LR=5e-6 (full SFT)

  • Sequence length: 4096

bridge.recipes.qwen_vl.h100.qwen25_vl.qwen25_vl_7b_sft_2gpu_h100_bf16_config() → megatron.bridge.training.config.ConfigContainer#

Return a full SFT config for Qwen2.5-VL 7B Instruct.

Default configuration: 2 GPUs

  • TP=2, PP=1

  • LR=5e-6 (full SFT)

  • Sequence length: 4096

bridge.recipes.qwen_vl.h100.qwen25_vl.qwen25_vl_32b_sft_16gpu_h100_bf16_config() → megatron.bridge.training.config.ConfigContainer#

Return a full SFT config for Qwen2.5-VL 32B Instruct.

Default configuration: 16 GPUs

  • TP=8, PP=2

  • LR=5e-6 (full SFT)

  • Sequence length: 4096

bridge.recipes.qwen_vl.h100.qwen25_vl.qwen25_vl_72b_sft_32gpu_h100_bf16_config() → megatron.bridge.training.config.ConfigContainer#

Return a full SFT config for Qwen2.5-VL 72B Instruct.

Default configuration: 32 GPUs

  • TP=8, PP=4

  • LR=5e-6 (full SFT)

  • Sequence length: 4096

bridge.recipes.qwen_vl.h100.qwen25_vl.qwen25_vl_3b_peft_1gpu_h100_bf16_config(
peft_scheme: str | megatron.bridge.peft.base.PEFT = 'lora',
) → megatron.bridge.training.config.ConfigContainer#

Return a PEFT config for Qwen2.5-VL 3B Instruct.

Default configuration: 1 GPU

  • 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.qwen_vl.h100.qwen25_vl.qwen25_vl_7b_peft_1gpu_h100_bf16_config(
peft_scheme: str | megatron.bridge.peft.base.PEFT = 'lora',
) → megatron.bridge.training.config.ConfigContainer#

Return a PEFT config for Qwen2.5-VL 7B Instruct.

Default configuration: 1 GPU

  • 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.qwen_vl.h100.qwen25_vl.qwen25_vl_32b_peft_1gpu_h100_bf16_config(
peft_scheme: str | megatron.bridge.peft.base.PEFT = 'lora',
) → megatron.bridge.training.config.ConfigContainer#

Return a PEFT config for Qwen2.5-VL 32B Instruct.

Default configuration: 1 GPU

  • 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.qwen_vl.h100.qwen25_vl.qwen25_vl_72b_peft_1gpu_h100_bf16_config(
peft_scheme: str | megatron.bridge.peft.base.PEFT = 'lora',
) → megatron.bridge.training.config.ConfigContainer#

Return a PEFT config for Qwen2.5-VL 72B Instruct.

Default configuration: 1 GPU

  • 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.qwen_vl.h100.qwen25_vl.__all__#

[‘qwen25_vl_32b_peft_1gpu_h100_bf16_config’, ‘qwen25_vl_32b_sft_16gpu_h100_bf16_config’, ‘qwen25_vl_…