o }oi)5@sdddlmZmZddlmZddlZddlZddl m Z ddl m Z ddl mZddlmZmZddlmZddlmZdd lmZmZdd lmZdd lmZdd lmZm Z m!Z!dd l"m#Z#ddl$m%Z%ddl&m'Z'ddl(m)Z)ddl*m+Z+dZ,ej-j.e,ddej/ej0fddZ1          d;de2de2deej3dee2d e2d!e4d"e2d#e2d$e2d%ee5ej/e dej/ej6fd&d'Z7ej-j.ee,d(dd)dddefd*ee8d+e8d"e2d#e2d,e4d-edej9fd.d/Z:d0ej9dej9fd1d2Z;ej-j.ee,d(  )   3   d>> model_config = model() >>> print(model_config) )config)runConfigr r rrZ/home/ubuntu/.local/lib/python3.10/site-packages/nemo/collections/llm/recipes/llama2_7b.pymodel)sr F{tensor_parallelismpipeline_parallelismpipeline_parallelism_typevirtual_pipeline_parallelismcontext_parallelismsequence_parallelism num_nodesnum_gpus_per_node max_steps callbacksc Csbtjtj||||||dddtjtddddddd } tjtjdd| |ddd||t| d d d } | S) ad Configure the NeMo Lightning Trainer for Llama2 7B model. This function sets up the distributed training strategy and other training parameters. Args: tensor_parallelism (int): Degree of tensor model parallelism. pipeline_parallelism (int): Degree of pipeline model parallelism. pipeline_parallelism_type (Optional[torch.dtype]): Data type for pipeline parallelism. virtual_pipeline_parallelism (Optional[int]): Size of virtual pipeline parallelism. context_parallelism (int): Degree of context parallelism. sequence_parallelism (bool): Whether to use sequence parallelism. num_nodes (int): Number of compute nodes to use. num_gpus_per_node (int): Number of GPUs per node. max_steps (int): Maximum number of training steps. callbacks (Optional[list[run.Config[Callback]]]): List of callback configurations. Returns: run.Config[nl.Trainer]: Configuration for the NeMo Lightning Trainer. Examples: CLI usage: $ nemo llm pretrain trainer=llama2_7b ... Python API usage: >>> trainer_config = trainer(num_nodes=2, num_gpus_per_node=8) >>> print(trainer_config) Note: For more information on distributed training strategies, refer to the NeMo documentation on multi-GPU and multi-node training. Tcheck_for_nan_in_gradgrad_reduce_in_fp32overlap_grad_reduceoverlap_param_gatheraverage_in_collective) tensor_model_parallel_sizepipeline_model_parallel_sizepipeline_dtype$virtual_pipeline_model_parallel_sizecontext_parallel_sizesequence_parallelgradient_as_bucket_viewckpt_async_saveckpt_parallel_loadddpgpur"2 Fi) acceleratoraccumulate_grad_batchesr.deviceslimit_test_batcheslimit_val_batcheslog_every_n_stepsr-r+pluginsstrategyuse_distributed_samplerval_check_interval)rrnlMegatronStrategyrTrainerr) r%r&r'r(r)r*r+r,r-r.rJtrainerrrrrP<sJ,rP)targetrdefaultdirrperformance_modefnc Csbtj|tt||ttgdtjtddddt||t|ddt dd t d }|r/t |}|S) a Create a pre-training recipe for Llama2 7B model. This function sets up a complete configuration for pre-training, including model, trainer, data, logging, optimization, and resumption settings. Args: dir (Optional[str]): Directory for saving logs and checkpoints. name (str): Name of the pre-training run. num_nodes (int): Number of compute nodes to use. num_gpus_per_node (int): Number of GPUs per node. performance_mode (bool): If true, enables optimizations for maximum performance. fn (Callable): The pre-training function to use. Returns: run.Partial: Partial configuration for pre-training. Examples: CLI usage: $ nemo llm pretrain --factory llama2_7b $ nemo llm pretrain --factory "llama2_7b(num_nodes=2, name='my_pretrain')" Python API usage: >>> recipe = pretrain_recipe(name="llama2_7b_pretrain", num_nodes=2) >>> print(recipe) )r+r,r.ir") seq_lengthglobal_batch_sizemicro_batch_sizer)rSrrga2U0*3?)max_lr)r rPdatalogoptimresume) rPartialr rPrrr rrrr"pretrain_performance_optimizations)rSrr+r,rTrUreciperrrpretrain_recipes # rbracCs|jjtjtdd|S)a Create a performance-optimized pre-training recipe for Llama2 7B model. This method enables performance optimizations that may not be suitable for all use cases. It builds upon the standard pre-training recipe and adds additional performance enhancements. Args: recipe (run.Partial): Base pre-train recipe to which performance optimizations will be added Returns: run.Partial: Partial configuration for performance-optimized pre-training. Note: Use this method with caution and only when you need maximum performance. It may not be suitable for all hardware configurations or use cases. Ftp_comm_overlap)rPr.appendrrr)rarrrr`sr`lora peft_schemerWpacked_sequencec Cs|dur|}|dur|rdnd}ttd|||||}|dus%|dkr0d|jj_d|jj_n/|dvrXt t ||_ d |j _ d |j _d |jj_d |jj_d |jj_ntd |||jj_||j_|ryddi|j_t j t|d|j_|rt||}|S)a Create a fine-tuning recipe for Llama2 7B model. This function sets up a complete configuration for fine-tuning, including model, trainer, data, logging, optimization, and resumption settings. The recipe uses LoRA (Low-Rank Adaptation) for efficient fine-tuning, unless peft_scheme is set to None. Args: dir (Optional[str]): Directory for saving logs and checkpoints. name (str): Name of the fine-tuning run. num_nodes (int): Number of compute nodes to use. num_gpus_per_node (int): Number of GPUs per node. peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora'/'dora'/'none'/None. seq_length (int): Maximum number of tokens per microbatch. packed_sequence (Optional[bool]): If true, fine-tuning sequences will be packed into batches up to the given maximum seq_length for better efficiency. By default, this value equals performance_mode. performance_mode (bool): If true, enables optimizations for maximum performance. Returns: run.Partial: Partial configuration for fine-tuning. Examples: CLI usage: $ nemo llm finetune --factory llama2_7b Python API usage: >>> recipe = finetune_recipe(name="llama2_7b_finetune", num_nodes=2) >>> print(recipe) Note: This recipe uses the SQuAD dataset for fine-tuning. NrVizmeta-llama/Llama-2-7b-hfnoner!gh㈵>)rfdorar#Fg-C6?zUnrecognized peft scheme: pad_to_max_lengthT)packed_sequence_size)rr lowerrPrJr5r]rlrrrr peftdimalphause_distributed_optimizercross_entropy_loss_fusion ValueErrorrWr[dataset_kwargsr packed_sequence_specs"finetune_performance_optimizations) rSrr+r,rgrWrhrTrarrrfinetune_recipes4.          rycCsd|jj_t|jdsg|j_|dus|dkr9d|jj_tj t dddddd|jj_ |jj tj t ddnd g|j_|jj t t|jj t td d |S) a Modify the given recipe to optimize settings for performance. This method enables performance optimizations that may not be suitable for all use cases. Intended to build upon the standard fine-tuning recipe. Args: recipe (run.Partial): Base fine-tuning recipe to which performance optimizations will be added peft_scheme (Optional[str]): Name of the peft scheme to use for fine-tuning. Allowed values: 'lora'/'dora'/'none'/None. Returns: run.Partial: Partial configuration for performance-optimized fine-tuning. Note: Use this method with caution and only when you need maximum performance. It may not be suitable for all hardware configurations or use cases. r"r.NriFTr/rc linear_qkvd)rPrJr5hasattrr.rnrIr1rrrr>rerrptarget_modulesrr)rargrrrrx7s:     rx) r!r"NNr"Fr"r#r$N)NrRr"r#rfNNF)>typingrrlightning.pytorchpytorchplnemo_runrtorch$lightning.pytorch.callbacks.callbackrmegatron.core.distributedrnemorrMnemo.collections.llm.apirr"nemo.collections.llm.gpt.data.mockr -nemo.collections.llm.gpt.data.packed_sequencer $nemo.collections.llm.gpt.model.llamar r nemo.collections.llm.peftr -nemo.collections.llm.recipes.finetune_defaultr(nemo.collections.llm.recipes.log.defaultrrr'nemo.collections.llm.recipes.optim.adamr6nemo.collections.llm.recipes.precision.mixed_precisionr3nemo.lightning.pytorch.callbacks.garbage_collectionr6nemo.lightning.pytorch.callbacks.megatron_comm_overlaprnemo.utils.exp_managerrNAMEclifactoryrLightningModuler intdtypeboollistrOrPstrr_rbr`ryrxrrrrs                 U 6 T