refactor(recipes): normalize h100 peft recipe aliases

Signed-off-by: yaoyu-33 <yaoyu.094@gmail.com>

Author: yaoyu-33

docs/fern/versions/nightly/pages/recipe-usage.mdx

@@ -12,7 +12,7 @@ This guide will cover the next steps to make use of a training recipe, including
 
 ## Choosing a recipe or a new config
 
-Start from an exported recipe when the model family and workflow already exist in `megatron.bridge.recipes`. Hardware-specific H100 recipes such as `llama3_8b_pretrain_2gpu_h100_bf16_config`, `llama32_1b_sft_1gpu_h100_bf16_config`, and `qwen3_8b_lora_1gpu_h100_bf16_config` are the canonical recipe names and share the same `<model>_<task>_<num>gpu_<gpu>_<dtype>_config` naming strategy as performance recipes. The older recipe names remain available as compatibility wrappers. Override those defaults for your dataset, checkpoint paths, run length, parallelism, or precision before creating a new recipe.
+Start from an exported recipe when the model family and workflow already exist in `megatron.bridge.recipes`. Hardware-specific H100 recipes such as `llama3_8b_pretrain_2gpu_h100_bf16_config`, `llama32_1b_sft_1gpu_h100_bf16_config`, and `qwen3_8b_peft_1gpu_h100_bf16_config` are the canonical recipe names and share the same `<model>_<task>_<num>gpu_<gpu>_<dtype>_config` naming strategy as performance recipes. The older recipe names remain available as compatibility wrappers. Override those defaults for your dataset, checkpoint paths, run length, parallelism, or precision before creating a new recipe.
 
 Create a new recipe or config when the base model architecture is not represented by an existing model provider, the checkpoint conversion needs a new bridge, the forward step or dataset provider is model-specific, or you need a reusable configuration that will be shared across jobs. If the Hugging Face model is already supported by `AutoBridge`, you usually only need to start from the closest recipe and override the model provider or `hf_path`.
 

docs/fern/versions/nightly/pages/training/data-preparation.mdx

@@ -115,7 +115,7 @@ The generic launcher provides preset Hugging Face text datasets through `--datas
 
 ```bash
 uv run python -m torch.distributed.run --nproc_per_node=1 scripts/training/run_recipe.py \
-    --recipe llama32_1b_lora_1gpu_h100_bf16_config \
+    --recipe llama32_1b_peft_1gpu_h100_bf16_config \
     --dataset llm-finetune \
     dataset.dataset_name=gsm8k \
     checkpoint.pretrained_checkpoint=/checkpoints/base_model
@@ -129,7 +129,7 @@ For Energon/WebDataset data, create tar shards plus `.nv-meta` metadata and pass
 
 ```bash
 uv run python -m torch.distributed.run --nproc_per_node=1 scripts/training/run_recipe.py \
-    --recipe qwen3_vl_8b_lora_1gpu_h100_bf16_energon_config \
+    --recipe qwen3_vl_8b_peft_1gpu_h100_bf16_energon_config \
     --dataset vlm-energon \
     --step_func qwen3_vl_step \
     dataset.path=/data/vlm_energon \
@@ -144,7 +144,7 @@ For preloaded VLM JSON or JSONL, use records with `messages` or `conversations`
 
 ```bash
 uv run python -m torch.distributed.run --nproc_per_node=1 scripts/training/run_recipe.py \
-    --recipe qwen3_vl_8b_lora_1gpu_h100_bf16_config \
+    --recipe qwen3_vl_8b_peft_1gpu_h100_bf16_config \
     --dataset vlm-preloaded \
     --step_func qwen3_vl_step \
     dataset.train_data_path=/data/vlm/train.jsonl \

docs/recipe-usage.md

@@ -12,7 +12,7 @@ This guide will cover the next steps to make use of a training recipe, including
 
 ## Choosing a recipe or a new config
 
-Start from an exported recipe when the model family and workflow already exist in `megatron.bridge.recipes`. Hardware-specific H100 recipes such as `llama3_8b_pretrain_2gpu_h100_bf16_config`, `llama32_1b_sft_1gpu_h100_bf16_config`, and `qwen3_8b_lora_1gpu_h100_bf16_config` are the canonical recipe names and share the same `<model>_<task>_<num>gpu_<gpu>_<dtype>_config` naming strategy as performance recipes. The older recipe names remain available as compatibility wrappers. Override those defaults for your dataset, checkpoint paths, run length, parallelism, or precision before creating a new recipe.
+Start from an exported recipe when the model family and workflow already exist in `megatron.bridge.recipes`. Hardware-specific H100 recipes such as `llama3_8b_pretrain_2gpu_h100_bf16_config`, `llama32_1b_sft_1gpu_h100_bf16_config`, and `qwen3_8b_peft_1gpu_h100_bf16_config` are the canonical recipe names and share the same `<model>_<task>_<num>gpu_<gpu>_<dtype>_config` naming strategy as performance recipes. The older recipe names remain available as compatibility wrappers. Override those defaults for your dataset, checkpoint paths, run length, parallelism, or precision before creating a new recipe.
 
 Create a new recipe or config when the base model architecture is not represented by an existing model provider, the checkpoint conversion needs a new bridge, the forward step or dataset provider is model-specific, or you need a reusable configuration that will be shared across jobs. If the Hugging Face model is already supported by `AutoBridge`, you usually only need to start from the closest recipe and override the model provider or `hf_path`.
 

docs/training/data-preparation.md

@@ -116,7 +116,7 @@ The generic launcher provides preset Hugging Face text datasets through `--datas
 
 ```bash
 uv run python -m torch.distributed.run --nproc_per_node=1 scripts/training/run_recipe.py \
-    --recipe llama32_1b_lora_1gpu_h100_bf16_config \
+    --recipe llama32_1b_peft_1gpu_h100_bf16_config \
     --dataset llm-finetune \
     dataset.dataset_name=gsm8k \
     checkpoint.pretrained_checkpoint=/checkpoints/base_model
@@ -130,7 +130,7 @@ For Energon/WebDataset data, create tar shards plus `.nv-meta` metadata and pass
 
 ```bash
 uv run python -m torch.distributed.run --nproc_per_node=1 scripts/training/run_recipe.py \
-    --recipe qwen3_vl_8b_lora_1gpu_h100_bf16_energon_config \
+    --recipe qwen3_vl_8b_peft_1gpu_h100_bf16_energon_config \
     --dataset vlm-energon \
     --step_func qwen3_vl_step \
     dataset.path=/data/vlm_energon \
@@ -145,7 +145,7 @@ For preloaded VLM JSON or JSONL, use records with `messages` or `conversations`
 
 ```bash
 uv run python -m torch.distributed.run --nproc_per_node=1 scripts/training/run_recipe.py \
-    --recipe qwen3_vl_8b_lora_1gpu_h100_bf16_config \
+    --recipe qwen3_vl_8b_peft_1gpu_h100_bf16_config \
     --dataset vlm-preloaded \
     --step_func qwen3_vl_step \
     dataset.train_data_path=/data/vlm/train.jsonl \

src/megatron/bridge/recipes/deepseek/deepseek_v2.py

@@ -17,33 +17,22 @@
 from __future__ import annotations
 
 from megatron.bridge.recipes.deepseek.h100 import deepseek_v2 as _h100_module
-from megatron.bridge.recipes.utils.aliases import call_h100_recipe
 from megatron.bridge.training.config import ConfigContainer
 
 
 AutoBridge = _h100_module.AutoBridge
 
 
-def deepseek_v2_lite_pretrain_config(*args: object, **kwargs: object) -> ConfigContainer:
+def deepseek_v2_lite_pretrain_config() -> ConfigContainer:
     """Compatibility alias for ``deepseek_v2_lite_pretrain_8gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.deepseek_v2_lite_pretrain_8gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.deepseek_v2_lite_pretrain_8gpu_h100_bf16_config()
 
 
-def deepseek_v2_pretrain_config(*args: object, **kwargs: object) -> ConfigContainer:
+def deepseek_v2_pretrain_config() -> ConfigContainer:
     """Compatibility alias for ``deepseek_v2_pretrain_128gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.deepseek_v2_pretrain_128gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.deepseek_v2_pretrain_128gpu_h100_bf16_config()
 
 
 __all__ = [

src/megatron/bridge/recipes/deepseek/deepseek_v3.py

@@ -21,33 +21,22 @@
     _build_standalone_mtp_layout,
     set_deepseek_v3_pipeline_model_parallel_layout,
 )
-from megatron.bridge.recipes.utils.aliases import call_h100_recipe
 from megatron.bridge.training.config import ConfigContainer
 
 
 AutoBridge = _h100_module.AutoBridge
 
 
-def deepseek_v3_pretrain_config(*args: object, **kwargs: object) -> ConfigContainer:
+def deepseek_v3_pretrain_config() -> ConfigContainer:
     """Compatibility alias for ``deepseek_v3_pretrain_1024gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.deepseek_v3_pretrain_1024gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.deepseek_v3_pretrain_1024gpu_h100_bf16_config()
 
 
-def deepseek_v3_pretrain_config_32nodes(*args: object, **kwargs: object) -> ConfigContainer:
+def deepseek_v3_pretrain_config_32nodes() -> ConfigContainer:
     """Compatibility alias for ``deepseek_v3_pretrain_256gpu_h100_bf16_32nodes_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.deepseek_v3_pretrain_256gpu_h100_bf16_32nodes_config,
-        *args,
-        **kwargs,
-    )
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.deepseek_v3_pretrain_256gpu_h100_bf16_32nodes_config()
 
 
 __all__ = [

src/megatron/bridge/recipes/deepseek/deepseek_v4.py

@@ -21,66 +21,40 @@
     DEEPSEEK_V4_FLASH_HF_PATH,
     _deepseek_v4_mxfp8_quant_recipe,
 )
-from megatron.bridge.recipes.utils.aliases import call_h100_recipe
 from megatron.bridge.training.config import ConfigContainer
 
 
 AutoBridge = _h100_module.AutoBridge
 
 
-def deepseek_v4_flash_no_mtp_sft_config(*args: object, **kwargs: object) -> ConfigContainer:
+def deepseek_v4_flash_no_mtp_sft_config() -> ConfigContainer:
     """Compatibility alias for ``deepseek_v4_flash_no_mtp_sft_32gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.deepseek_v4_flash_no_mtp_sft_32gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.deepseek_v4_flash_no_mtp_sft_32gpu_h100_bf16_config()
 
 
-def deepseek_v4_flash_pretrain_config(*args: object, **kwargs: object) -> ConfigContainer:
+def deepseek_v4_flash_pretrain_config() -> ConfigContainer:
     """Compatibility alias for ``deepseek_v4_flash_pretrain_32gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.deepseek_v4_flash_pretrain_32gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.deepseek_v4_flash_pretrain_32gpu_h100_bf16_config()
 
 
-def deepseek_v4_flash_pretrain_muon_config(*args: object, **kwargs: object) -> ConfigContainer:
+def deepseek_v4_flash_pretrain_muon_config() -> ConfigContainer:
     """Compatibility alias for ``deepseek_v4_flash_pretrain_32gpu_h100_bf16_muon_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.deepseek_v4_flash_pretrain_32gpu_h100_bf16_muon_config,
-        *args,
-        **kwargs,
-    )
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.deepseek_v4_flash_pretrain_32gpu_h100_bf16_muon_config()
 
 
-def deepseek_v4_flash_pretrain_mxfp8_config(*args: object, **kwargs: object) -> ConfigContainer:
+def deepseek_v4_flash_pretrain_mxfp8_config() -> ConfigContainer:
     """Compatibility alias for ``deepseek_v4_flash_pretrain_32gpu_h100_fp8mx_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.deepseek_v4_flash_pretrain_32gpu_h100_fp8mx_config,
-        *args,
-        **kwargs,
-    )
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.deepseek_v4_flash_pretrain_32gpu_h100_fp8mx_config()
 
 
-def deepseek_v4_flash_sft_config(*args: object, **kwargs: object) -> ConfigContainer:
+def deepseek_v4_flash_sft_config() -> ConfigContainer:
     """Compatibility alias for ``deepseek_v4_flash_sft_32gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.deepseek_v4_flash_sft_32gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.deepseek_v4_flash_sft_32gpu_h100_bf16_config()
 
 
 __all__ = [

src/megatron/bridge/recipes/gemma/gemma2.py

@@ -16,114 +16,69 @@
 
 from __future__ import annotations
 
+from megatron.bridge.peft.base import PEFT
 from megatron.bridge.recipes.gemma.h100 import gemma2 as _h100_module
 from megatron.bridge.recipes.gemma.h100.gemma2 import (
     _adjust_gemma2_vocab_size,
 )
-from megatron.bridge.recipes.utils.aliases import call_h100_recipe
 from megatron.bridge.training.config import ConfigContainer
 
 
 AutoBridge = _h100_module.AutoBridge
 
 
-def gemma2_27b_peft_config(*args: object, **kwargs: object) -> ConfigContainer:
-    """Compatibility alias for ``gemma2_27b_lora_4gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.gemma2_27b_lora_4gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
+def gemma2_27b_peft_config(peft_scheme: str | PEFT = "lora") -> ConfigContainer:
+    """Compatibility alias for ``gemma2_27b_peft_4gpu_h100_bf16_config``."""
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.gemma2_27b_peft_4gpu_h100_bf16_config(peft_scheme=peft_scheme)
 
 
-def gemma2_27b_pretrain_config(*args: object, **kwargs: object) -> ConfigContainer:
+def gemma2_27b_pretrain_config() -> ConfigContainer:
     """Compatibility alias for ``gemma2_27b_pretrain_16gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.gemma2_27b_pretrain_16gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.gemma2_27b_pretrain_16gpu_h100_bf16_config()
 
 
-def gemma2_27b_sft_config(*args: object, **kwargs: object) -> ConfigContainer:
+def gemma2_27b_sft_config() -> ConfigContainer:
     """Compatibility alias for ``gemma2_27b_sft_16gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.gemma2_27b_sft_16gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
-
-
-def gemma2_2b_peft_config(*args: object, **kwargs: object) -> ConfigContainer:
-    """Compatibility alias for ``gemma2_2b_lora_1gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.gemma2_2b_lora_1gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
-
-
-def gemma2_2b_pretrain_config(*args: object, **kwargs: object) -> ConfigContainer:
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.gemma2_27b_sft_16gpu_h100_bf16_config()
+
+
+def gemma2_2b_peft_config(peft_scheme: str | PEFT = "lora") -> ConfigContainer:
+    """Compatibility alias for ``gemma2_2b_peft_1gpu_h100_bf16_config``."""
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.gemma2_2b_peft_1gpu_h100_bf16_config(peft_scheme=peft_scheme)
+
+
+def gemma2_2b_pretrain_config() -> ConfigContainer:
     """Compatibility alias for ``gemma2_2b_pretrain_2gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.gemma2_2b_pretrain_2gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.gemma2_2b_pretrain_2gpu_h100_bf16_config()
 
 
-def gemma2_2b_sft_config(*args: object, **kwargs: object) -> ConfigContainer:
+def gemma2_2b_sft_config() -> ConfigContainer:
     """Compatibility alias for ``gemma2_2b_sft_1gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.gemma2_2b_sft_1gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
-
-
-def gemma2_9b_peft_config(*args: object, **kwargs: object) -> ConfigContainer:
-    """Compatibility alias for ``gemma2_9b_lora_1gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.gemma2_9b_lora_1gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
-
-
-def gemma2_9b_pretrain_config(*args: object, **kwargs: object) -> ConfigContainer:
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.gemma2_2b_sft_1gpu_h100_bf16_config()
+
+
+def gemma2_9b_peft_config(peft_scheme: str | PEFT = "lora") -> ConfigContainer:
+    """Compatibility alias for ``gemma2_9b_peft_1gpu_h100_bf16_config``."""
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.gemma2_9b_peft_1gpu_h100_bf16_config(peft_scheme=peft_scheme)
+
+
+def gemma2_9b_pretrain_config() -> ConfigContainer:
     """Compatibility alias for ``gemma2_9b_pretrain_8gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.gemma2_9b_pretrain_8gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.gemma2_9b_pretrain_8gpu_h100_bf16_config()
 
 
-def gemma2_9b_sft_config(*args: object, **kwargs: object) -> ConfigContainer:
+def gemma2_9b_sft_config() -> ConfigContainer:
     """Compatibility alias for ``gemma2_9b_sft_4gpu_h100_bf16_config``."""
-    return call_h100_recipe(
-        globals(),
-        _h100_module,
-        _h100_module.gemma2_9b_sft_4gpu_h100_bf16_config,
-        *args,
-        **kwargs,
-    )
+    _h100_module.AutoBridge = AutoBridge
+    return _h100_module.gemma2_9b_sft_4gpu_h100_bf16_config()
 
 
 __all__ = [