Merge branch 'main' into jingyux/megatron-lora

Author: jingyu-ml

.gitlab/tests.yml

@@ -54,20 +54,12 @@ example-torch:
   timeout: 30m
   parallel:
     matrix:
-      - EXAMPLE: [llm_distill, llm_sparsity, speculative_decoding]
+      - EXAMPLE: [llm_distill, llm_qat, llm_sparsity, speculative_decoding]
   script:
     - pip install ".[hf,dev-test]"
     - find examples/$EXAMPLE -name "requirements.txt" | while read req_file; do pip install -r "$req_file" || exit 1; done
     - pytest -s tests/examples/$EXAMPLE
 
-# TODO: Fix llm_qat test hang in GitLab CI
-example-failing:
-  extends: example-torch
-  allow_failure: true
-  parallel:
-    matrix:
-      - EXAMPLE: [llm_qat]
-
 example-trtllm:
   extends: example-torch
   timeout: 60m

docs/source/guides/7_nas.rst

@@ -635,3 +635,12 @@ The difference between NAS and pruning is summarized below.
         increased training time.
       - May provide similar performance to NAS in particular applications, however, usually exhibits
         worse performance due to the limited search space and training time.
+
+
+[Advanced] Adding a new NAS/Prune Algorithm
+===========================================
+
+* Please refer to this `template <https://github.com/NVIDIA/TensorRT-Model-Optimizer/compare/template/new-nas-mode>`_ 
+  for adding a new NAS algorithm.
+* Please refer to `mcore_minitron.py <https://github.com/NVIDIA/TensorRT-Model-Optimizer/blob/main/modelopt/torch/prune/plugins/mcore_minitron.py>`_
+  for an actual example of adding Minitron Pruning algorithm.

examples/llm_distill/main.py

@@ -13,10 +13,11 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import logging
 import os
 from dataclasses import dataclass
 
+os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True"
+
 import datasets
 import torch
 import torch.distributed
@@ -29,10 +30,7 @@
 import modelopt.torch.opt as mto
 from modelopt.torch.distill.plugins.huggingface import KDTrainer, LMLogitsLoss
 
-os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "max_split_size_mb:512"
-
-logger = get_logger(__name__)
-logging.basicConfig(level=logging.INFO)
+logger = get_logger(__name__, log_level="INFO")
 
 
 @dataclass
@@ -69,6 +67,29 @@ class KDSFTTrainer(SFTTrainer, KDTrainer):
     pass
 
 
+def _save_model_fsdp_compat(
+    self,
+    output_dir: str | None = None,
+    _internal_call: bool = False,
+    *args,
+    **kwargs,
+):
+    output_dir = output_dir or self.args.output_dir
+    model = self.accelerator.unwrap_model(self.model)
+    if not _internal_call and self.is_fsdp_enabled:
+        state_dict = self.accelerator.get_state_dict(self.model)
+        if self.accelerator.is_main_process:
+            model.save_pretrained(
+                output_dir,
+                is_main_process=self.accelerator.is_main_process,
+                save_function=self.accelerator.save,
+                state_dict=state_dict,
+            )
+            self.processing_class.save_pretrained(output_dir)
+    else:
+        super(SFTTrainer, self).save_model(output_dir, _internal_call, *args, **kwargs)
+
+
 def train():
     parser = transformers.HfArgumentParser((ModelArguments, TrainingArguments))
     model_args, training_args = parser.parse_args_into_dataclasses()
@@ -77,6 +98,9 @@ def train():
     # modelopt state will be saved automatically to "modelopt_state.pth"
     mto.enable_huggingface_checkpointing()
 
+    # HACK: Fix FSDP2-incompatible save_model() function for SFTTrainer
+    SFTTrainer.save_model = _save_model_fsdp_compat
+
     # Set total batch size across all ranks to equal 64
     total_batch_size = 64
     num_accum_steps = total_batch_size / (
@@ -91,19 +115,22 @@ def train():
         f"Using {int(num_accum_steps)} grad accumulation steps for effective batchsize of {total_batch_size}."
     )
 
+    # Dataset
     logger.info("Loading dataset...")
     dset = datasets.load_dataset("Open-Orca/OpenOrca", split="train")
     dset_splits = dset.train_test_split(train_size=25600, test_size=1700, seed=420)
     dset_train, dset_eval = dset_splits["train"], dset_splits["test"]
     logger.info("Dataset loaded.")
 
+    # Tokenizer
     logger.info("Loading tokenizer...")
     model_path = model_args.teacher_name_or_path or model_args.student_name_or_path
     tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=True)
     tokenizer.pad_token = tokenizer.eos_token
     tokenizer.padding_side = "right"
     logger.info("Tokenizer loaded.")
 
+    # Model
     if model_args.single_model:
         logger.info("Loading single model only...")
         model = transformers.AutoModelForCausalLM.from_pretrained(

examples/llm_distill/requirements.txt

@@ -1,2 +1,3 @@
 pyarrow
+transformers<5.0
 trl>=0.23.0

modelopt/torch/__init__.py

@@ -34,7 +34,7 @@
 
     if not (_Version("4.48") <= _Version(_transformers_version) < _Version("5.0")):
         _warnings.warn(
-            f"transformers version {_transformers_version} is incompatible with nvidia-modelopt and may cause issues. "
+            f"transformers version {_transformers_version} is not tested with nvidia-modelopt and may cause issues. "
             "Please install recommended version with `pip install nvidia-modelopt[hf]` if working with HF models.",
         )
 except ImportError:

modelopt/torch/opt/plugins/__init__.py

@@ -19,9 +19,6 @@
 
 from .huggingface import *
 
-with import_plugin("megatron core model config"):
-    from .megatron_model_config import *
-
 with import_plugin("megatron core dist checkpointing"):
     from .mcore_dist_checkpointing import *
 

tests/gpu/torch/nas/plugins/test_megatron_mamba_dynamic_modules.py

@@ -173,7 +173,7 @@ def _test_mamba_parameter_sorting(rank, size):
     prompt_tokens = torch.randint(0, vocab_size, (batch_size, max_sequence_length)).cuda()
     y1 = run_mcore_inference(model, prompt_tokens)
 
-    dynamic_space.sort_parameters()
+    mtn.utils.sort_parameters(model)
 
     # check if all mamba_num_heads, mamba_head_dim, hidden_size have been sorted
     sortable_per_pp = [