[BE] Normalized to use model_args: ModelArgs
#58
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
facebook-github-bot
added
the
CLA Signed
awgu
changed the title
model_args: ModelArgsmodel_args: ModelArgs
lessw2020
approved these changes
Feb 13, 2024
lessw2020
pushed a commit
that referenced
this pull request
Apr 18, 2024
Some modules used `args: ModelArgs`, others `params: ModelArgs`, and
others `model_args: ModelArgs`. This PR normalizes everything to use
`model_args: ModelArgs` for consistency. (`params` might be confused
with `nn.Parameter`s, and `model_args` was more explicit than `args`.)
**Test Plan**
```
./run_llama_train.sh
```
<details>
<summary> Output </summary>
```
+ TRAINER_DIR=/home/andgu/local/torchtrain
+ MODEL=debugmodel
+ NGPU=8
+ PP=1
+ SP=1
+ DP=-1
+ LOG_RANK=0
+ CHECKPOINT_FOLDER=
+ CHECKPOINT_INTERVAL=5
+ torchrun --nproc_per_node=8 --local-ranks-filter 0 --role rank --tee 3 train.py --steps 10 --compile --pp_degree 1 --sp_degree 1 --dp_degree -1
[2024-02-13 09:53:31,345] torch.distributed.run: [WARNING]
[2024-02-13 09:53:31,345] torch.distributed.run: [WARNING] *****************************************
[2024-02-13 09:53:31,345] torch.distributed.run: [WARNING] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed.
[2024-02-13 09:53:31,345] torch.distributed.run: [WARNING] *****************************************
[rank0]:2024-02-13 09:53:33,644 - torchtrain.parallelisms - INFO - Building 1-D device mesh with ('dp',), [8]
[rank0]:2024-02-13 09:53:36,955 - root - INFO - Reloaded SentencePiece model from ./torchtrain/datasets/tokenizer/tokenizer.model
[rank0]:2024-02-13 09:53:36,955 - root - INFO - #words: 32000 - BOS ID: 1 - EOS ID: 2
[rank0]:/home/andgu/.conda/envs/pytorch-3.10/lib/python3.10/site-packages/scipy/__init__.py:146: UserWarning: A NumPy version >=1.17.3 and <1.25.0 is required for this version of SciPy (detected version 1.26.0
[rank0]: warnings.warn(f"A NumPy version >={np_minversion} and <{np_maxversion}"
[rank0]:2024-02-13 09:53:41,571 - root - INFO - Applied FSDP to the model...
[rank0]:2024-02-13 09:53:41,572 - root - INFO - Gradient scaling not enabled.
[rank0]:2024-02-13 09:53:41,572 - root - INFO - Compiling model llama with torch.compile...
[rank0]:2024-02-13 09:53:43,892 - root - INFO - Profiling active. Traces will be saved at ./torchtrain/outputs/profiling/traces
[rank0]:NCCL version 2.19.3+cuda12.0
[rank0]:[rank0]:[2024-02-13 09:53:43,995] [0/0] torch._dynamo.variables.torch: [WARNING] Profiler function <class 'torch.autograd.profiler.record_function'> will be ignored
[rank0]:/data/users/andgu/pytorch/torch/_inductor/lowering.py:1697: UserWarning: Torchinductor does not support code generation for complex operators. Performance may be worse than eager.
[rank0]: warnings.warn(
[rank0]:2024-02-13 09:54:06,085 - root - INFO - step: 1, current loss: 10.54707145690918, lr: [0.0002666666666666667]
[rank0]:2024-02-13 09:54:06,153 - root - INFO - step: 2, current loss: 10.481386184692383, lr: [0.0005333333333333334]
[rank0]:2024-02-13 09:54:06,222 - root - INFO - step: 3, current loss: 10.334623336791992, lr: [0.0008]
[rank0]:2024-02-13 09:54:06,288 - root - INFO - step: 4, current loss: 10.121940612792969, lr: [0.0007]
[rank0]:2024-02-13 09:54:06,355 - root - INFO - step: 5, current loss: 9.922933578491211, lr: [0.0006000000000000001]
[rank0]:2024-02-13 09:54:06,422 - root - INFO - step: 6, current loss: 9.710294723510742, lr: [0.0005]
[rank0]:2024-02-13 09:54:06,487 - root - INFO - step: 7, current loss: 9.587849617004395, lr: [0.0004]
[rank0]:2024-02-13 09:54:06,773 - root - INFO - step: 8, current loss: 9.474313735961914, lr: [0.00030000000000000003]
[rank0]:STAGE:2024-02-13 09:54:06 3243810:3243810 ActivityProfilerController.cpp:314] Completed Stage: Warm Up
[rank0]:2024-02-13 09:54:06,845 - root - INFO - step: 9, current loss: 9.282522201538086, lr: [0.0002]
[rank0]:[rank0]:[W CPUAllocator.cpp:249] Memory block of unknown size was allocated before the profiling started, profiler results will not include the deallocation event
[rank0]:STAGE:2024-02-13 09:54:06 3243810:3243810 ActivityProfilerController.cpp:320] Completed Stage: Collection
[rank0]:STAGE:2024-02-13 09:54:06 3243810:3243810 ActivityProfilerController.cpp:324] Completed Stage: Post Processing
[rank0]:2024-02-13 09:54:06,999 - root - INFO - exporting profile traces to ./torchtrain/outputs/profiling/traces/iteration_10
[rank0]:2024-02-13 09:54:07,002 - root - INFO - step: 10, current loss: 9.34823989868164, lr: [0.0001]
```
</details>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Some modules used
args: ModelArgs, othersparams: ModelArgs, and othersmodel_args: ModelArgs. This PR normalizes everything to usemodel_args: ModelArgsfor consistency. (paramsmight be confused withnn.Parameters, andmodel_argswas more explicit thanargs.)Test Plan
Output