Fix LoRA contiguous tensor (NVIDIA#10611)

* contiguous

Signed-off-by: Chen Cui <chcui@nvidia.com>

* fix load

Signed-off-by: Chen Cui <chcui@nvidia.com>

* add test script

Signed-off-by: Chen Cui <chcui@nvidia.com>

* Apply isort and black reformatting

Signed-off-by: cuichenx <cuichenx@users.noreply.github.com>

* Apply isort and black reformatting

Signed-off-by: artbataev <artbataev@users.noreply.github.com>

---------

Signed-off-by: Chen Cui <chcui@nvidia.com>
Signed-off-by: cuichenx <cuichenx@users.noreply.github.com>
Signed-off-by: artbataev <artbataev@users.noreply.github.com>
Co-authored-by: cuichenx <cuichenx@users.noreply.github.com>
Co-authored-by: artbataev <artbataev@users.noreply.github.com>

nemo/collections/llm/peft/lora.py

@@ -50,7 +50,7 @@ def forward(self, x):
             linear_output, bias, layernorm_output = linear_output
             x = layernorm_output
 
-        adapter_output = self.adapter(x)
+        adapter_output = self.adapter(x.contiguous())
         return linear_output + adapter_output, bias
 
 

nemo/lightning/pytorch/callbacks/peft.py

@@ -107,6 +107,9 @@ def setup(self, trainer: pl.Trainer, pl_module: pl.LightningModule, stage: str)
 
     def apply_transform(self, trainer):
         super().apply_transform(trainer)
+        self.trainable_params = set(
+            name for name, param in trainer.lightning_module.named_parameters() if param.requires_grad
+        )
 
         adapter_sharded_state_dict = {}
         if self.wrapped_io.adapter_ckpt_path is not None:
@@ -137,10 +140,6 @@ def apply_transform(self, trainer):
             if trainer.state.fn == TrainerFn.FITTING:
                 trainer.strategy.load_optimizer_state_dict(adapter_state, selective_restore=True)
 
-        self.trainable_params = set(
-            name for name, param in trainer.lightning_module.named_parameters() if param.requires_grad
-        )
-
     def adapter_key_filter(self, key: str) -> bool:
         return key in self.trainable_params or ".adapter." in key or key.endswith(".adapters")
 

tests/collections/llm/gpt_finetuning.py

@@ -0,0 +1,115 @@
+# Copyright (c) 2024, NVIDIA CORPORATION.  All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import argparse
+import os
+from dataclasses import dataclass
+
+from megatron.core.optimizer import OptimizerConfig
+
+from nemo import lightning as nl
+from nemo.collections import llm
+from nemo.collections.nlp.modules.common.tokenizer_utils import get_nmt_tokenizer
+
+## NOTE: This script is present for github-actions testing only.
+
+
+@dataclass
+class Llama3Config96M(llm.Llama3Config8B):
+    seq_length: int = 2048
+    num_layers: int = 2
+    hidden_size: int = 768
+    ffn_hidden_size: int = 3072
+    num_attention_heads: int = 8
+
+
+def get_args():
+    parser = argparse.ArgumentParser(description='Finetune a small GPT model using NeMo 2.0')
+    parser.add_argument('--restore_path', type=str, help="Path to model to be finetuned")
+    parser.add_argument('--experiment_dir', type=str, help="directory to write results and checkpoints to")
+    parser.add_argument('--devices', type=int, default=1, help="number of devices")
+    parser.add_argument('--mbs', type=int, default=1, help="micro batch size")
+    parser.add_argument('--tp_size', type=int, default=1, help="tensor parallel size")
+    parser.add_argument('--pp_size', type=int, default=1, help="pipeline parallel size")
+
+    return parser.parse_args()
+
+
+if __name__ == '__main__':
+    args = get_args()
+
+    strategy = nl.MegatronStrategy(
+        tensor_model_parallel_size=args.tp_size,
+        pipeline_parallel_size=args.pp_size,
+    )
+
+    trainer = nl.Trainer(
+        devices=args.devices,
+        max_steps=2,
+        accelerator="gpu",
+        strategy=strategy,
+        plugins=nl.MegatronMixedPrecision(precision="bf16-mixed"),
+        log_every_n_steps=1,
+        limit_val_batches=2,
+        val_check_interval=2,
+        num_sanity_val_steps=0,
+    )
+
+    ckpt = nl.ModelCheckpoint(
+        save_last=True,
+        monitor="reduced_train_loss",
+        save_top_k=1,
+        save_on_train_epoch_end=True,
+        save_optim_on_train_end=True,
+    )
+
+    logger = nl.NeMoLogger(
+        log_dir=args.experiment_dir,
+        use_datetime_version=False,  # must be false if using auto resume
+        ckpt=ckpt,
+    )
+
+    adam = nl.MegatronOptimizerModule(
+        config=OptimizerConfig(
+            optimizer="adam",
+            lr=0.0001,
+            adam_beta2=0.98,
+            use_distributed_optimizer=True,
+            clip_grad=1.0,
+            bf16=True,
+        ),
+    )
+
+    lora = llm.peft.LoRA()
+
+    squad = llm.SquadDataModule(seq_length=2048, micro_batch_size=args.mbs, global_batch_size=8, num_workers=0)
+
+    tokenizer = get_nmt_tokenizer(
+        tokenizer_model="/lustre/fsw/coreai_dlalgo_llm/nemo_home/models/llama_96M/dummy_tokenizer.model"
+    )
+    llama3_8b = llm.LlamaModel(Llama3Config96M(), tokenizer=tokenizer)
+
+    resume = nl.AutoResume(
+        restore_config=nl.RestoreConfig(path=args.restore_path),
+        resume_if_exists=True,
+    )
+
+    llm.finetune(
+        model=llama3_8b,
+        data=squad,
+        trainer=trainer,
+        peft=lora,
+        log=logger,
+        optim=adam,
+        resume=resume,
+    )