add default llama2 config converter (#563)

Co-authored-by: Duy Phung <[email protected]>

examples/llama_nemo/convert_llama_to_nemo.py

@@ -106,7 +106,7 @@ def get_mlp_weight(model_state_dict, layer_mapping, key, tp_idx):
 
     # dummy config
 
-    megatron_cfg_path = Path(__file__).parent / "megatron_7b_llama.yaml"
+    megatron_cfg_path = Path(__file__).parent / "megatron_llama_cfg.yaml"
 
     megatron_cfg = OmegaConf.load(megatron_cfg_path)
     megatron_cfg.name = f"megatron_{args.name}"

examples/llama_nemo/megatron_llama_cfg.yaml

@@ -0,0 +1,166 @@
+name: megatron_llama2
+restore_from_path: null  # used when starting from a .nemo file
+
+trainer:
+  devices: 8
+  num_nodes: 4
+  accelerator: gpu
+  precision: bf16
+  logger: False # logger provided by exp_manager
+  enable_checkpointing: False
+  replace_sampler_ddp: False
+  max_epochs: -1 # PTL default. In practice, max_steps will be reached first.
+  max_steps: 200 # consumed_samples = global_step * micro_batch_size * data_parallel_size * accumulate_grad_batches
+  log_every_n_steps: 1
+  val_check_interval: 20
+  # check_val_every_n_epoch: null
+  limit_val_batches: 2
+  limit_test_batches: 0
+  accumulate_grad_batches: 1 # do not modify, grad acc is automatic for training megatron models
+  gradient_clip_val: 1.0
+  benchmark: False
+
+exp_manager:
+  # set this to save checkpoints
+  explicit_log_dir: ppo_sentiments_logs
+  exp_dir: null
+  name: megatron_gpt_7b_ppo_sentiments
+  create_tensorboard_logger: False
+  create_wandb_logger: False
+  wandb_logger_kwargs:
+    project: trlxnemo
+    name: megatron_gpt_7b_ppo_sentiments
+  resume_if_exists: False
+  resume_ignore_no_checkpoint: True
+  # set this to save checkpoints
+  create_checkpoint_callback: False
+  checkpoint_callback_params:
+    monitor: reduced_train_loss
+    save_top_k: 1
+    mode: min
+    always_save_nemo: False # saves nemo file during validation, not implemented for model parallel
+    save_nemo_on_train_end: True # not recommended when training large models on clusters with short time limits
+    filename: 'megatron_gpt-{reduced_train_loss:.2f}-{step}-{consumed_samples}'
+    model_parallel_size: ${multiply:${model.tensor_model_parallel_size}, ${model.pipeline_model_parallel_size}}
+  log_step_timing: True
+  step_timing_kwargs:
+    sync_cuda: True
+    buffer_size: 5
+
+model:
+  padded_vocab_size: 32000
+  micro_batch_size: 1
+  global_batch_size: 32
+  tensor_model_parallel_size: 4
+  pipeline_model_parallel_size: 1
+  resume_from_checkpoint: null # manually set the checkpoint file to load from
+  # model architecture
+  encoder_seq_length: 2048
+  max_position_embeddings: 2048
+  num_layers: 32
+  hidden_size: 4096
+  ffn_hidden_size: 11008 # Transformer FFN hidden size. Usually 4 * hidden_size.
+  num_attention_heads: 32
+  init_method_std: 0.02 # Standard deviation of the zero mean normal distribution used for weight initialization.')
+  use_scaled_init_method: True # use scaled residuals initialization
+  hidden_dropout: 0. # Dropout probability for hidden state transformer.
+  attention_dropout: 0. # Dropout probability for attention
+  ffn_dropout: 0.0 # Dropout probability in the feed-forward layer.
+  kv_channels: null # Projection weights dimension in multi-head attention. Set to hidden_size // num_attention_heads if null
+  apply_query_key_layer_scaling: False # scale Q * K^T by 1 / layer-number.
+  layernorm_epsilon: 1e-6
+  transformer_block_type: pre_ln
+  do_layer_norm_weight_decay: False # True means weight decay on all params
+  make_vocab_size_divisible_by: 1 # Pad the vocab size to be divisible by this value for computation efficiency.
+  pre_process: True # add embedding
+  post_process: True # add pooler
+  persist_layer_norm: True # Use of persistent fused layer norm kernel.
+  # Need to set this to remove bias terms in LN in NeMo
+  normalization: rmsnorm
+  bias: False # Whether to use bias terms in all weight matrices.
+  bias_activation_fusion: False # Use a kernel that fuses the bias addition from weight matrices with the subsequent activation function.
+  bias_dropout_add_fusion: False # Use a kernel that fuses the bias addition, dropout and residual connection addition.
+  activation: 'swiglu' # Options ['gelu', 'geglu', 'swiglu', 'reglu', 'squared-relu', 'fast-geglu', 'fast-swiglu', 'fast-reglu']
+  headscale: False # Whether to learn extra parameters that scale the output of the each self-attention head.
+  rotary_percentage: 1
+  megatron_legacy: True
+  position_embedding_type: 'rope' # Options ['rotary', 'learned', 'rope']
+  share_embeddings_and_output_weights: false
+
+  ## Activation Checkpointing
+  activations_checkpoint_granularity:  'selective' #'selective'  # 'selective' or 'full'
+  activations_checkpoint_method: 'uniform' # 'uniform', 'block', not used with 'selective'
+  activations_checkpoint_num_layers: null # not used with 'selective'
+
+  ## Sequence Parallelism
+  sequence_parallel: True
+
+  # tokenizer:
+  #   library: 'megatron'
+  #   type: 'GPT2BPETokenizer'
+  #   model: null
+  #   vocab_file: null
+  #   merge_file: null
+  #   delimiter: null # only used for tabular tokenizer
+  #   sentencepiece_legacy: false # Legacy=True allows you to add special tokens to sentencepiece tokenizers.
+  tokenizer:
+    library: 'huggingface'
+    type: 'NousResearch/Llama-2-7b-hf'
+    model: null
+    vocab_file: null
+    merge_file: null
+    delimiter: null # only used for tabular tokenizer
+    sentencepiece_legacy: false # Legacy=True allows you to add special tokens to sentencepiece tokenizers.
+
+
+  # precision
+  native_amp_init_scale: 4294967296 # 2 ** 32
+  native_amp_growth_interval: 1000
+  hysteresis: 2 # Gradient scale hysteresis
+  fp32_residual_connection: False # Move residual connections to fp32
+  fp16_lm_cross_entropy: False # Move the cross entropy unreduced loss calculation for lm head to fp16
+
+  # Megatron O2-style half-precision
+  megatron_amp_O2: True # Enable O2-level automatic mixed precision using main parameters
+  grad_allreduce_chunk_size_mb: 125
+  sync_batch_comm: False
+  # miscellaneous
+  seed: 1234
+  use_cpu_initialization: False # Init weights on the CPU (slow for large models)
+  onnx_safe: False # Use work-arounds for known problems with Torch ONNX exporter.
+  apex_transformer_log_level: 30 # Python logging level displays logs with severity greater than or equal to this
+  gradient_as_bucket_view: True # PyTorch DDP argument. Allocate gradients in a contiguous bucket to save memory (less fragmentation and buffer memory)
+
+  data:
+    data_prefix:
+        - dataset: hh
+    index_mapping_dir: null # path to save index mapping .npy files, by default will save in the same location as data_prefix
+    data_impl: mmap
+    splits_string: 900,50,50
+    seq_length: ${model.encoder_seq_length}
+    skip_warmup: True
+    num_workers: 2
+    dataloader_type: cyclic
+    reset_position_ids: False # Reset position ids after end-of-document token
+    reset_attention_mask: False # Reset attention mask after end-of-document token
+    eod_mask_loss: False # Mask loss for the end of document tokens
+
+  # Nsys profiling options
+  nsys_profile:
+    enabled: False
+    start_step: 10  # Global batch to start profiling
+    end_step: 10 # Global batch to end profiling
+    ranks: [0, 4, 8, 12] # Global rank IDs to profile
+    gen_shape: False # Generate model and kernel details including input shapes
+
+  optim:
+    name: distributed_fused_adam
+    lr: 1.0e-6
+    weight_decay: 0.0 #1.0e-6
+    betas:
+    - 0.9
+    - 0.95
+    sched:
+      name: CosineAnnealing
+      max_steps: 200
+      min_lr: 1.0e-6