Upload folder using huggingface_hub

.gitattributes

@@ -34,3 +34,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 tokenizer.json filter=lfs diff=lfs merge=lfs -text
+nemo/weights/.metadata filter=lfs diff=lfs merge=lfs -text
+nemo/weights/__0_0.distcp filter=lfs diff=lfs merge=lfs -text
+nemo/weights/__0_1.distcp filter=lfs diff=lfs merge=lfs -text

nemo/context/io.json

@@ -0,0 +1 @@
+{"root": {"type": "ref", "key": "trainer_context_1"}, "objects": {"tuple_1": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [["Index(index=0)", "tensor_model_parallel_size"], ["Index(index=1)", "pipeline_model_parallel_size"], ["Index(index=2)", "virtual_pipeline_model_parallel_size"], ["Index(index=3)", "sequence_parallel"], ["Index(index=4)", "context_parallel_size"], ["Index(index=5)", "expert_model_parallel_size"], ["Index(index=6)", "expert_tensor_parallel_size"], ["Index(index=7)", "moe_extended_tp"], ["Index(index=8)", "use_te_rng_tracker"], ["Index(index=9)", "pipeline_dtype"], ["Index(index=10)", "microbatch_group_size_per_vp_stage"], ["Index(index=11)", "num_layers_in_first_pipeline_stage"], ["Index(index=12)", "num_layers_in_last_pipeline_stage"], ["Index(index=13)", "account_for_embedding_in_pipeline_split"], ["Index(index=14)", "account_for_loss_in_pipeline_split"]], "metadata": null}, "dict_1": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "dict_2": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "buildable_traverser_metadata_1": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}, "items": [["Attr(name='fn_or_cls')", {"type": "pyref", "module": "nemo.collections.llm.gpt.model.ssm", "name": "NemotronHConfig47B"}], ["Attr(name='argument_names')", {"type": "ref", "key": "tuple_1"}], ["Attr(name='argument_tags')", {"type": "ref", "key": "dict_1"}], ["Attr(name='argument_history')", {"type": "ref", "key": "dict_2"}]], "metadata": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}}, "nemotron_hconfig47_b_1": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "Config"}, "items": [["Attr(name='tensor_model_parallel_size')", {"type": "leaf", "value": 1, "paths": ["<root>.model.config.tensor_model_parallel_size"]}], ["Attr(name='pipeline_model_parallel_size')", {"type": "leaf", "value": 1, "paths": ["<root>.model.config.pipeline_model_parallel_size"]}], ["Attr(name='virtual_pipeline_model_parallel_size')", {"type": "leaf", "value": null, "paths": ["<root>.model.config.virtual_pipeline_model_parallel_size"]}], ["Attr(name='sequence_parallel')", {"type": "leaf", "value": false, "paths": ["<root>.model.config.sequence_parallel"]}], ["Attr(name='context_parallel_size')", {"type": "leaf", "value": 1, "paths": ["<root>.model.config.context_parallel_size"]}], ["Attr(name='expert_model_parallel_size')", {"type": "leaf", "value": 1, "paths": ["<root>.model.config.expert_model_parallel_size"]}], ["Attr(name='expert_tensor_parallel_size')", {"type": "leaf", "value": null, "paths": ["<root>.model.config.expert_tensor_parallel_size"]}], ["Attr(name='moe_extended_tp')", {"type": "leaf", "value": false, "paths": ["<root>.model.config.moe_extended_tp"]}], ["Attr(name='use_te_rng_tracker')", {"type": "leaf", "value": false, "paths": ["<root>.model.config.use_te_rng_tracker"]}], ["Attr(name='pipeline_dtype')", {"type": "leaf", "value": null, "paths": ["<root>.model.config.pipeline_dtype"]}], ["Attr(name='microbatch_group_size_per_vp_stage')", {"type": "leaf", "value": 1, "paths": ["<root>.model.config.microbatch_group_size_per_vp_stage"]}], ["Attr(name='num_layers_in_first_pipeline_stage')", {"type": "leaf", "value": null, "paths": ["<root>.model.config.num_layers_in_first_pipeline_stage"]}], ["Attr(name='num_layers_in_last_pipeline_stage')", {"type": "leaf", "value": null, "paths": ["<root>.model.config.num_layers_in_last_pipeline_stage"]}], ["Attr(name='account_for_embedding_in_pipeline_split')", {"type": "leaf", "value": false, "paths": ["<root>.model.config.account_for_embedding_in_pipeline_split"]}], ["Attr(name='account_for_loss_in_pipeline_split')", {"type": "leaf", "value": false, "paths": ["<root>.model.config.account_for_loss_in_pipeline_split"]}]], "metadata": {"type": "ref", "key": "buildable_traverser_metadata_1"}, "paths": ["<root>.model.config"]}, "tuple_2": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [["Index(index=0)", "optimizer"], ["Index(index=1)", "lr"], ["Index(index=2)", "min_lr"], ["Index(index=3)", "decoupled_lr"], ["Index(index=4)", "decoupled_min_lr"], ["Index(index=5)", "weight_decay"], ["Index(index=6)", "fp16"], ["Index(index=7)", "bf16"], ["Index(index=8)", "params_dtype"], ["Index(index=9)", "use_precision_aware_optimizer"], ["Index(index=10)", "main_grads_dtype"], ["Index(index=11)", "main_params_dtype"], ["Index(index=12)", "exp_avg_dtype"], ["Index(index=13)", "exp_avg_sq_dtype"], ["Index(index=14)", "loss_scale"], ["Index(index=15)", "initial_loss_scale"], ["Index(index=16)", "min_loss_scale"], ["Index(index=17)", "loss_scale_window"], ["Index(index=18)", "hysteresis"], ["Index(index=19)", "adam_beta1"], ["Index(index=20)", "adam_beta2"], ["Index(index=21)", "adam_eps"], ["Index(index=22)", "sgd_momentum"], ["Index(index=23)", "use_distributed_optimizer"], ["Index(index=24)", "overlap_param_gather_with_optimizer_step"], ["Index(index=25)", "optimizer_cpu_offload"], ["Index(index=26)", "optimizer_offload_fraction"], ["Index(index=27)", "use_torch_optimizer_for_cpu_offload"], ["Index(index=28)", "overlap_cpu_optimizer_d2h_h2d"], ["Index(index=29)", "pin_cpu_grads"], ["Index(index=30)", "pin_cpu_params"], ["Index(index=31)", "clip_grad"], ["Index(index=32)", "log_num_zeros_in_grad"], ["Index(index=33)", "barrier_with_L1_time"], ["Index(index=34)", "timers"], ["Index(index=35)", "config_logger_dir"]], "metadata": null}, "dict_3": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "dict_4": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "buildable_traverser_metadata_2": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}, "items": [["Attr(name='fn_or_cls')", {"type": "pyref", "module": "megatron.core.optimizer.optimizer_config", "name": "OptimizerConfig"}], ["Attr(name='argument_names')", {"type": "ref", "key": "tuple_2"}], ["Attr(name='argument_tags')", {"type": "ref", "key": "dict_3"}], ["Attr(name='argument_history')", {"type": "ref", "key": "dict_4"}]], "metadata": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}}, "optimizer_config_1": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "Config"}, "items": [["Attr(name='optimizer')", {"type": "leaf", "value": "adam", "paths": ["<root>.model.optim.config.optimizer"]}], ["Attr(name='lr')", {"type": "leaf", "value": 0.0001, "paths": ["<root>.model.optim.config.lr"]}], ["Attr(name='min_lr')", {"type": "leaf", "value": null, "paths": ["<root>.model.optim.config.min_lr"]}], ["Attr(name='decoupled_lr')", {"type": "leaf", "value": null, "paths": ["<root>.model.optim.config.decoupled_lr"]}], ["Attr(name='decoupled_min_lr')", {"type": "leaf", "value": null, "paths": ["<root>.model.optim.config.decoupled_min_lr"]}], ["Attr(name='weight_decay')", {"type": "leaf", "value": 0.01, "paths": ["<root>.model.optim.config.weight_decay"]}], ["Attr(name='fp16')", {"type": "leaf", "value": false, "paths": ["<root>.model.optim.config.fp16"]}], ["Attr(name='bf16')", {"type": "leaf", "value": false, "paths": ["<root>.model.optim.config.bf16"]}], ["Attr(name='params_dtype')", {"type": "pyref", "module": "torch", "name": "float32", "paths": ["<root>.model.optim.config.params_dtype", "<root>.model.optim.config.main_grads_dtype", "<root>.model.optim.config.main_params_dtype", "<root>.model.optim.config.exp_avg_dtype", "<root>.model.optim.config.exp_avg_sq_dtype"]}], ["Attr(name='use_precision_aware_optimizer')", {"type": "leaf", "value": false, "paths": ["<root>.model.optim.config.use_precision_aware_optimizer"]}], ["Attr(name='main_grads_dtype')", {"type": "pyref", "module": "torch", "name": "float32", "paths": ["<root>.model.optim.config.params_dtype", "<root>.model.optim.config.main_grads_dtype", "<root>.model.optim.config.main_params_dtype", "<root>.model.optim.config.exp_avg_dtype", "<root>.model.optim.config.exp_avg_sq_dtype"]}], ["Attr(name='main_params_dtype')", {"type": "pyref", "module": "torch", "name": "float32", "paths": ["<root>.model.optim.config.params_dtype", "<root>.model.optim.config.main_grads_dtype", "<root>.model.optim.config.main_params_dtype", "<root>.model.optim.config.exp_avg_dtype", "<root>.model.optim.config.exp_avg_sq_dtype"]}], ["Attr(name='exp_avg_dtype')", {"type": "pyref", "module": "torch", "name": "float32", "paths": ["<root>.model.optim.config.params_dtype", "<root>.model.optim.config.main_grads_dtype", "<root>.model.optim.config.main_params_dtype", "<root>.model.optim.config.exp_avg_dtype", "<root>.model.optim.config.exp_avg_sq_dtype"]}], ["Attr(name='exp_avg_sq_dtype')", {"type": "pyref", "module": "torch", "name": "float32", "paths": ["<root>.model.optim.config.params_dtype", "<root>.model.optim.config.main_grads_dtype", "<root>.model.optim.config.main_params_dtype", "<root>.model.optim.config.exp_avg_dtype", "<root>.model.optim.config.exp_avg_sq_dtype"]}], ["Attr(name='loss_scale')", {"type": "leaf", "value": null, "paths": ["<root>.model.optim.config.loss_scale"]}], ["Attr(name='initial_loss_scale')", {"type": "leaf", "value": 4294967296, "paths": ["<root>.model.optim.config.initial_loss_scale"]}], ["Attr(name='min_loss_scale')", {"type": "leaf", "value": 1.0, "paths": ["<root>.model.optim.config.min_loss_scale"]}], ["Attr(name='loss_scale_window')", {"type": "leaf", "value": 1000, "paths": ["<root>.model.optim.config.loss_scale_window"]}], ["Attr(name='hysteresis')", {"type": "leaf", "value": 2, "paths": ["<root>.model.optim.config.hysteresis"]}], ["Attr(name='adam_beta1')", {"type": "leaf", "value": 0.9, "paths": ["<root>.model.optim.config.adam_beta1"]}], ["Attr(name='adam_beta2')", {"type": "leaf", "value": 0.999, "paths": ["<root>.model.optim.config.adam_beta2"]}], ["Attr(name='adam_eps')", {"type": "leaf", "value": 1e-08, "paths": ["<root>.model.optim.config.adam_eps"]}], ["Attr(name='sgd_momentum')", {"type": "leaf", "value": 0.9, "paths": ["<root>.model.optim.config.sgd_momentum"]}], ["Attr(name='use_distributed_optimizer')", {"type": "leaf", "value": true, "paths": ["<root>.model.optim.config.use_distributed_optimizer"]}], ["Attr(name='overlap_param_gather_with_optimizer_step')", {"type": "leaf", "value": false, "paths": ["<root>.model.optim.config.overlap_param_gather_with_optimizer_step"]}], ["Attr(name='optimizer_cpu_offload')", {"type": "leaf", "value": false, "paths": ["<root>.model.optim.config.optimizer_cpu_offload"]}], ["Attr(name='optimizer_offload_fraction')", {"type": "leaf", "value": 0.0, "paths": ["<root>.model.optim.config.optimizer_offload_fraction"]}], ["Attr(name='use_torch_optimizer_for_cpu_offload')", {"type": "leaf", "value": false, "paths": ["<root>.model.optim.config.use_torch_optimizer_for_cpu_offload"]}], ["Attr(name='overlap_cpu_optimizer_d2h_h2d')", {"type": "leaf", "value": false, "paths": ["<root>.model.optim.config.overlap_cpu_optimizer_d2h_h2d"]}], ["Attr(name='pin_cpu_grads')", {"type": "leaf", "value": true, "paths": ["<root>.model.optim.config.pin_cpu_grads"]}], ["Attr(name='pin_cpu_params')", {"type": "leaf", "value": true, "paths": ["<root>.model.optim.config.pin_cpu_params"]}], ["Attr(name='clip_grad')", {"type": "leaf", "value": 1.0, "paths": ["<root>.model.optim.config.clip_grad"]}], ["Attr(name='log_num_zeros_in_grad')", {"type": "leaf", "value": false, "paths": ["<root>.model.optim.config.log_num_zeros_in_grad"]}], ["Attr(name='barrier_with_L1_time')", {"type": "leaf", "value": false, "paths": ["<root>.model.optim.config.barrier_with_L1_time"]}], ["Attr(name='timers')", {"type": "leaf", "value": null, "paths": ["<root>.model.optim.config.timers"]}], ["Attr(name='config_logger_dir')", {"type": "leaf", "value": "", "paths": ["<root>.model.optim.config.config_logger_dir"]}]], "metadata": {"type": "ref", "key": "buildable_traverser_metadata_2"}, "paths": ["<root>.model.optim.config"]}, "tuple_3": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [["Index(index=0)", "config"]], "metadata": null}, "dict_5": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "dict_6": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "buildable_traverser_metadata_3": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}, "items": [["Attr(name='fn_or_cls')", {"type": "pyref", "module": "nemo.lightning.pytorch.optim.megatron", "name": "MegatronOptimizerModule"}], ["Attr(name='argument_names')", {"type": "ref", "key": "tuple_3"}], ["Attr(name='argument_tags')", {"type": "ref", "key": "dict_5"}], ["Attr(name='argument_history')", {"type": "ref", "key": "dict_6"}]], "metadata": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}}, "megatron_optimizer_module_1": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "Config"}, "items": [["Attr(name='config')", {"type": "ref", "key": "optimizer_config_1"}]], "metadata": {"type": "ref", "key": "buildable_traverser_metadata_3"}, "paths": ["<root>.model.optim"]}, "tuple_4": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [["Index(index=0)", "pretrained_model_name"], ["Index(index=1)", "trust_remote_code"]], "metadata": null}, "dict_7": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "dict_8": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "buildable_traverser_metadata_4": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}, "items": [["Attr(name='fn_or_cls')", {"type": "pyref", "module": "nemo.collections.common.tokenizers.huggingface.auto_tokenizer", "name": "AutoTokenizer"}], ["Attr(name='argument_names')", {"type": "ref", "key": "tuple_4"}], ["Attr(name='argument_tags')", {"type": "ref", "key": "dict_7"}], ["Attr(name='argument_history')", {"type": "ref", "key": "dict_8"}]], "metadata": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}}, "auto_tokenizer_1": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "Config"}, "items": [["Attr(name='pretrained_model_name')", {"type": "leaf", "value": "nemo_tokenizer", "paths": ["<root>.model.tokenizer.pretrained_model_name"]}], ["Attr(name='trust_remote_code')", {"type": "leaf", "value": true, "paths": ["<root>.model.tokenizer.trust_remote_code"]}]], "metadata": {"type": "ref", "key": "buildable_traverser_metadata_4"}, "paths": ["<root>.model.tokenizer"]}, "tuple_5": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [["Index(index=0)", "config"], ["Index(index=1)", "optim"], ["Index(index=2)", "tokenizer"]], "metadata": null}, "dict_9": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "dict_10": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "buildable_traverser_metadata_5": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}, "items": [["Attr(name='fn_or_cls')", {"type": "pyref", "module": "nemo.collections.llm.gpt.model.ssm", "name": "MambaModel"}], ["Attr(name='argument_names')", {"type": "ref", "key": "tuple_5"}], ["Attr(name='argument_tags')", {"type": "ref", "key": "dict_9"}], ["Attr(name='argument_history')", {"type": "ref", "key": "dict_10"}]], "metadata": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}}, "mamba_model_1": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "Config"}, "items": [["Attr(name='config')", {"type": "ref", "key": "nemotron_hconfig47_b_1"}], ["Attr(name='optim')", {"type": "ref", "key": "megatron_optimizer_module_1"}], ["Attr(name='tokenizer')", {"type": "ref", "key": "auto_tokenizer_1"}]], "metadata": {"type": "ref", "key": "buildable_traverser_metadata_5"}, "paths": ["<root>.model"]}, "tuple_6": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [["Index(index=0)", "always_save_context"]], "metadata": null}, "dict_11": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [["Key(key='always_save_context')", {"type": "leaf", "value": true, "paths": ["<root>.trainer.strategy.kwargs['always_save_context']"]}]], "metadata": {"type": "ref", "key": "tuple_6"}, "paths": ["<root>.trainer.strategy.kwargs"]}, "tuple_7": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [["Index(index=0)", "ckpt_save_optimizer"], ["Index(index=1)", "kwargs"]], "metadata": null}, "dict_12": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "dict_13": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "buildable_traverser_metadata_6": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}, "items": [["Attr(name='fn_or_cls')", {"type": "pyref", "module": "nemo.lightning.pytorch.strategies.megatron_strategy", "name": "MegatronStrategy"}], ["Attr(name='argument_names')", {"type": "ref", "key": "tuple_7"}], ["Attr(name='argument_tags')", {"type": "ref", "key": "dict_12"}], ["Attr(name='argument_history')", {"type": "ref", "key": "dict_13"}]], "metadata": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}}, "megatron_strategy_1": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "Config"}, "items": [["Attr(name='ckpt_save_optimizer')", {"type": "leaf", "value": false, "paths": ["<root>.trainer.strategy.ckpt_save_optimizer"]}], ["Attr(name='kwargs')", {"type": "ref", "key": "dict_11"}]], "metadata": {"type": "ref", "key": "buildable_traverser_metadata_6"}, "paths": ["<root>.trainer.strategy"]}, "tuple_8": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [["Index(index=0)", "accelerator"], ["Index(index=1)", "strategy"], ["Index(index=2)", "devices"]], "metadata": null}, "dict_14": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "dict_15": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "buildable_traverser_metadata_7": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}, "items": [["Attr(name='fn_or_cls')", {"type": "pyref", "module": "nemo.lightning.pytorch.trainer", "name": "Trainer"}], ["Attr(name='argument_names')", {"type": "ref", "key": "tuple_8"}], ["Attr(name='argument_tags')", {"type": "ref", "key": "dict_14"}], ["Attr(name='argument_history')", {"type": "ref", "key": "dict_15"}]], "metadata": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}}, "trainer_1": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "Config"}, "items": [["Attr(name='accelerator')", {"type": "leaf", "value": "cpu", "paths": ["<root>.trainer.accelerator"]}], ["Attr(name='strategy')", {"type": "ref", "key": "megatron_strategy_1"}], ["Attr(name='devices')", {"type": "leaf", "value": 1, "paths": ["<root>.trainer.devices"]}]], "metadata": {"type": "ref", "key": "buildable_traverser_metadata_7"}, "paths": ["<root>.trainer"]}, "dict_16": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}, "paths": ["<root>.extra"]}, "tuple_9": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [["Index(index=0)", "model"], ["Index(index=1)", "trainer"], ["Index(index=2)", "extra"]], "metadata": null}, "dict_17": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "dict_18": {"type": {"type": "pyref", "module": "builtins", "name": "dict"}, "items": [], "metadata": {"type": {"type": "pyref", "module": "builtins", "name": "tuple"}, "items": [], "metadata": null}}, "buildable_traverser_metadata_8": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}, "items": [["Attr(name='fn_or_cls')", {"type": "pyref", "module": "nemo.lightning.io.pl", "name": "TrainerContext"}], ["Attr(name='argument_names')", {"type": "ref", "key": "tuple_9"}], ["Attr(name='argument_tags')", {"type": "ref", "key": "dict_17"}], ["Attr(name='argument_history')", {"type": "ref", "key": "dict_18"}]], "metadata": {"type": "pyref", "module": "fiddle._src.config", "name": "BuildableTraverserMetadata"}}, "trainer_context_1": {"type": {"type": "pyref", "module": "fiddle._src.config", "name": "Config"}, "items": [["Attr(name='model')", {"type": "ref", "key": "mamba_model_1"}], ["Attr(name='trainer')", {"type": "ref", "key": "trainer_1"}], ["Attr(name='extra')", {"type": "ref", "key": "dict_16"}]], "metadata": {"type": "ref", "key": "buildable_traverser_metadata_8"}, "paths": ["<root>"]}}, "refcounts": {"tuple_1": 1, "dict_1": 1, "dict_2": 1, "buildable_traverser_metadata_1": 1, "nemotron_hconfig47_b_1": 1, "tuple_2": 1, "dict_3": 1, "dict_4": 1, "buildable_traverser_metadata_2": 1, "optimizer_config_1": 1, "tuple_3": 1, "dict_5": 1, "dict_6": 1, "buildable_traverser_metadata_3": 1, "megatron_optimizer_module_1": 1, "tuple_4": 1, "dict_7": 1, "dict_8": 1, "buildable_traverser_metadata_4": 1, "auto_tokenizer_1": 1, "tuple_5": 1, "dict_9": 1, "dict_10": 1, "buildable_traverser_metadata_5": 1, "mamba_model_1": 1, "tuple_6": 1, "dict_11": 1, "tuple_7": 1, "dict_12": 1, "dict_13": 1, "buildable_traverser_metadata_6": 1, "megatron_strategy_1": 1, "tuple_8": 1, "dict_14": 1, "dict_15": 1, "buildable_traverser_metadata_7": 1, "trainer_1": 1, "dict_16": 1, "tuple_9": 1, "dict_17": 1, "dict_18": 1, "buildable_traverser_metadata_8": 1, "trainer_context_1": 1}, "version": "0.0.1"}

nemo/context/model.yaml

@@ -0,0 +1,277 @@
+_target_: nemo.collections.llm.gpt.model.ssm.MambaModel
+config:
+  _cpu_offloading_context: null
+  _target_: nemo.collections.llm.gpt.model.ssm.NemotronHConfig47B
+  account_for_embedding_in_pipeline_split: false
+  account_for_loss_in_pipeline_split: false
+  activation_func:
+    _call_: false
+    _target_: nemo.collections.llm.gpt.model.ssm.NemotronHConfigBase.<lambda>
+  activation_func_fp8_input_store: false
+  add_bias_linear: false
+  add_qkv_bias: false
+  apply_query_key_layer_scaling: false
+  apply_residual_connection_post_layernorm: false
+  apply_rope_fusion: true
+  async_tensor_model_parallel_allreduce: false
+  attention_backend:
+    _call_: true
+    _target_: megatron.core.transformer.enums.AttnBackend
+  attention_dropout: 0.0
+  attention_softmax_in_fp32: false
+  autocast_dtype: null
+  barrier_with_L1_time: true
+  batch_p2p_comm: true
+  batch_p2p_sync: true
+  bf16: true
+  bias_activation_fusion: false
+  bias_dropout_fusion: true
+  calculate_per_token_loss: false
+  clone_scatter_output_in_embedding: true
+  config_logger_dir: ''
+  context_parallel_size: 1
+  cp_comm_type: null
+  cpu_offloading: false
+  cpu_offloading_activations: true
+  cpu_offloading_num_layers: 0
+  cpu_offloading_weights: true
+  cross_entropy_fusion_impl: native
+  cross_entropy_loss_fusion: true
+  cuda_graph_retain_backward_graph: false
+  cuda_graph_scope: full
+  cuda_graph_use_single_mempool: false
+  cuda_graph_warmup_steps: 3
+  data_step_fn:
+    _call_: false
+    _target_: nemo.collections.llm.gpt.model.base.gpt_data_step
+  deallocate_pipeline_outputs: true
+  defer_embedding_wgrad_compute: false
+  deterministic_mode: false
+  disable_parameter_transpose_cache: false
+  distribute_saved_activations: null
+  enable_autocast: false
+  enable_cuda_graph: false
+  expert_model_parallel_size: 1
+  expert_tensor_parallel_size: null
+  external_cuda_graph: false
+  ffn_hidden_size: 30720
+  finalize_model_grads_func: null
+  first_last_layers_bf16: true
+  flash_decode: false
+  forward_step_fn:
+    _call_: false
+    _target_: nemo.collections.llm.gpt.model.ssm.ssm_forward_step
+  fp16: false
+  fp16_lm_cross_entropy: false
+  fp32_residual_connection: false
+  fp8: null
+  fp8_amax_compute_algo: most_recent
+  fp8_amax_history_len: 1
+  fp8_dot_product_attention: false
+  fp8_interval: 1
+  fp8_margin: 0
+  fp8_multi_head_attention: false
+  fp8_recipe: delayed
+  fp8_wgrad: true
+  gated_linear_unit: false
+  get_attention_mask_from_fusion: false
+  grad_scale_func: null
+  grad_sync_func: null
+  gradient_accumulation_fusion: false
+  hidden_dropout: 0.0
+  hidden_size: 8192
+  hierarchical_context_parallel_sizes: null
+  hybrid_attention_ratio: 0.0
+  hybrid_mlp_ratio: 0.0
+  hybrid_override_pattern: M-M-M-M-M-M-M-M-M*-M-M-M-M-M-M-M-M-M-M*-M-M-M-M-M*-M-M-M-M-M*-M-M-M-M-M-M-M---MM---M-M*-M-M-M-M-M-
+  inference_rng_tracker: false
+  init_method: null
+  init_method_std: 0.02
+  init_model_with_meta_device: false
+  is_hybrid_model: true
+  kv_channels: null
+  layernorm_epsilon: 1.0e-05
+  layernorm_zero_centered_gamma: false
+  make_vocab_size_divisible_by: 128
+  mamba_head_dim: 64
+  mamba_nheads: 256
+  mamba_num_groups: 8
+  mamba_state_dim: 256
+  mapping_type: nvidia-hybrid-nemotronh
+  masked_softmax_fusion: true
+  memory_efficient_layer_norm: false
+  microbatch_group_size_per_vp_stage: 1
+  moe_aux_loss_coeff: 0
+  moe_enable_deepep: false
+  moe_expert_capacity_factor: null
+  moe_extended_tp: false
+  moe_ffn_hidden_size: null
+  moe_grouped_gemm: false
+  moe_input_jitter_eps: null
+  moe_layer_freq: 1
+  moe_layer_recompute: false
+  moe_pad_expert_input_to_capacity: false
+  moe_per_layer_logging: false
+  moe_permute_fusion: false
+  moe_router_bias_update_rate: 0.001
+  moe_router_dtype: null
+  moe_router_enable_expert_bias: false
+  moe_router_group_topk: null
+  moe_router_load_balancing_type: aux_loss
+  moe_router_num_groups: null
+  moe_router_pre_softmax: false
+  moe_router_score_function: softmax
+  moe_router_topk: 2
+  moe_router_topk_limited_devices: null
+  moe_router_topk_scaling_factor: null
+  moe_shared_expert_intermediate_size: null
+  moe_shared_expert_overlap: false
+  moe_token_dispatcher_type: allgather
+  moe_token_drop_policy: probs
+  moe_token_dropping: false
+  moe_use_legacy_grouped_gemm: false
+  moe_z_loss_coeff: null
+  mtp_loss_scaling_factor: null
+  mtp_num_layers: null
+  multi_latent_attention: false
+  no_sync_func: null
+  normalization: RMSNorm
+  num_attention_heads: 64
+  num_layers: 98
+  num_layers_at_end_in_bf16: 1
+  num_layers_at_start_in_bf16: 1
+  num_layers_in_first_pipeline_stage: null
+  num_layers_in_last_pipeline_stage: null
+  num_microbatches_with_partial_activation_checkpoints: null
+  num_moe_experts: null
+  num_query_groups: 8
+  output_layer_init_method: null
+  overlap_p2p_comm: false
+  overlap_p2p_comm_warmup_flush: false
+  parallel_output: true
+  param_sync_func: null
+  params_dtype:
+    _call_: false
+    _target_: torch.bfloat16
+  perform_initialization: true
+  persist_layer_norm: true
+  pipeline_dtype: null
+  pipeline_model_parallel_comm_backend: null
+  pipeline_model_parallel_size: 1
+  pipeline_model_parallel_split_rank: null
+  position_embedding_type: none
+  post_process: true
+  pre_process: true
+  qk_layernorm: false
+  recompute_granularity: null
+  recompute_method: null
+  recompute_num_layers: null
+  rotary_base: 10000
+  rotary_interleaved: false
+  rotary_percent: 1.0
+  seq_len_interpolation_factor: null
+  seq_length: 8192
+  sequence_parallel: false
+  share_embeddings_and_output_weights: false
+  softmax_scale: null
+  tensor_model_parallel_size: 1
+  test_mode: false
+  timers: null
+  tokenizer_library: tiktoken
+  tokenizer_model_path: null
+  tokenizer_name: TiktokenTokenizer
+  tp_comm_atomic_ag: false
+  tp_comm_atomic_rs: false
+  tp_comm_bootstrap_backend: nccl
+  tp_comm_bulk_dgrad: true
+  tp_comm_bulk_wgrad: true
+  tp_comm_overlap: false
+  tp_comm_overlap_ag: true
+  tp_comm_overlap_disable_fc1: false
+  tp_comm_overlap_disable_qkv: false
+  tp_comm_overlap_rs: true
+  tp_comm_overlap_rs_dgrad: false
+  tp_comm_split_ag: true
+  tp_comm_split_rs: true
+  tp_only_amax_red: false
+  use_cpu_initialization: false
+  use_custom_fsdp: false
+  use_ring_exchange_p2p: false
+  use_te_rng_tracker: false
+  variable_seq_lengths: false
+  virtual_pipeline_model_parallel_size: null
+  vocab_file: null
+  vocab_size: 131072
+  wgrad_deferral_limit: 0
+  window_size: null
+model_transform: null
+optim:
+  _target_: nemo.lightning.pytorch.optim.megatron.MegatronOptimizerModule
+  config:
+    _target_: megatron.core.optimizer.optimizer_config.OptimizerConfig
+    adam_beta1: 0.9
+    adam_beta2: 0.999
+    adam_eps: 1.0e-08
+    barrier_with_L1_time: false
+    bf16: false
+    clip_grad: 1.0
+    config_logger_dir: ''
+    decoupled_lr: null
+    decoupled_min_lr: null
+    exp_avg_dtype:
+      _call_: false
+      _target_: torch.float32
+    exp_avg_sq_dtype:
+      _call_: false
+      _target_: torch.float32
+    fp16: false
+    hysteresis: 2
+    initial_loss_scale: 4294967296
+    log_num_zeros_in_grad: false
+    loss_scale: null
+    loss_scale_window: 1000
+    lr: 0.0001
+    main_grads_dtype:
+      _call_: false
+      _target_: torch.float32
+    main_params_dtype:
+      _call_: false
+      _target_: torch.float32
+    min_loss_scale: 1.0
+    min_lr: null
+    optimizer: adam
+    optimizer_cpu_offload: false
+    optimizer_offload_fraction: 0.0
+    overlap_cpu_optimizer_d2h_h2d: false
+    overlap_param_gather_with_optimizer_step: false
+    params_dtype:
+      _call_: false
+      _target_: torch.float32
+    pin_cpu_grads: true
+    pin_cpu_params: true
+    sgd_momentum: 0.9
+    timers: null
+    use_distributed_optimizer: true
+    use_precision_aware_optimizer: false
+    use_torch_optimizer_for_cpu_offload: false
+    weight_decay: 0.01
+  lr_mult: 1.0
+  lr_scheduler: null
+  no_weight_decay_cond: null
+  scale_lr_cond: null
+tokenizer:
+  _target_: nemo.collections.common.tokenizers.huggingface.auto_tokenizer.AutoTokenizer
+  additional_special_tokens: []
+  bos_token: null
+  cls_token: null
+  eos_token: null
+  include_special_tokens: false
+  mask_token: null
+  merges_file: null
+  pad_token: null
+  pretrained_model_name: nemo_tokenizer
+  sep_token: null
+  trust_remote_code: true
+  unk_token: null
+  use_fast: false
+  vocab_file: null

nemo/context/nemo_tokenizer/special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

nemo/context/nemo_tokenizer/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3277c00fe5fb3963b3cb7c07b7f183722d2af4d775a4aea7cfb3684d7cccbc2f
+size 17078330

nemo/weights/.metadata

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c351cb6da43f06c2cbea80f6b64247dacaee17e4a7b0baee8c664709307b0e7
+size 425519

nemo/weights/__0_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87418abcf047f4dd5aa2924d8de88ccff8225e555249043a8916206ceebb17da
+size 46792303844

nemo/weights/__0_1.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9270a49378b0de2284768e37136f913abffd74caa9cf0684b0455e22d89704f7
+size 46792361482

nemo/weights/common.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf2d72b4f4fef17ea0a99868f378f27ac9989452f3ffdd0691821ae1a3b9f285
+size 3099

nemo/weights/metadata.json

@@ -0,0 +1 @@
+{"sharded_backend": "torch_dist", "sharded_backend_version": 1, "common_backend": "torch", "common_backend_version": 1}