[Full DTensor] Initial skeleton for full_dtensor mode #2049
Conversation
This file contains hidden or 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
fegin
added a commit
that referenced
this pull request
Nov 17, 2025
This PR provides a skelet This PR introduces an initial prototype and skeleton for fully DTensor-based training. The current codebase builds upon SimpleFSDP, but we anticipate developing our own Reparameterization to better serve our specific use case. There are several reasons why SimpleFSDP's Reparameterization is insufficient. For instance, the current parallelize_buffers() implementation in this PR will not function correctly when additional parallelization strategies are applied. Despite these limitations, this PR provides a starting point for experimenting with a full DTensor trainer. Accuracy verification: HSDP SimpleFSDP v.s. FSDP2 ``` python3 scripts/loss_compare.py . . \ --baseline-options='--activation_checkpoint.mode="none" --parallelism.data_parallel_replicate_degree=2' \ --test-options='--model.name full_dtensor.llama3 --activation_checkpoint.mode="none" --parallelism.data_parallel_replicate_degree=2' \ --test-train-file=torchtitan.experiments.full_dtensor.train \ --steps=10 --assert-equal --no-seed-checkpoint ``` ``` [LOSS_COMPARE] [LOSS_COMPARE] Asserting losses are equal... [LOSS_COMPARE] Baseline log: /tmp/baseline_training.log [LOSS_COMPARE] Test log: /tmp/test_training.log [LOSS_COMPARE] Extracted 100 steps from baseline log [LOSS_COMPARE] Extracted 100 steps from test log test_losses_equal (__main__.assert_losses_equal.<locals>.LossEqualityTest.test_losses_equal) ... ok ---------------------------------------------------------------------- Ran 1 test in 0.000s OK ``` Note that, `--no-seed-checkpoint` is used because when seed-checkpoint is used, we got accuracy mismatch. ghstack-source-id: 67cd703 Pull-Request: #2049
meta-cla
bot
added
the
CLA Signed
fegin
added a commit
that referenced
this pull request
Nov 17, 2025
This PR provides a skelet This PR introduces an initial prototype and skeleton for fully DTensor-based training. The current codebase builds upon SimpleFSDP, but we anticipate developing our own Reparameterization to better serve our specific use case. There are several reasons why SimpleFSDP's Reparameterization is insufficient. For instance, the current parallelize_buffers() implementation in this PR will not function correctly when additional parallelization strategies are applied. Despite these limitations, this PR provides a starting point for experimenting with a full DTensor trainer. Accuracy verification: HSDP SimpleFSDP v.s. FSDP2 ``` python3 scripts/loss_compare.py . . \ --baseline-options='--activation_checkpoint.mode="none" --parallelism.data_parallel_replicate_degree=2' \ --test-options='--model.name full_dtensor.llama3 --activation_checkpoint.mode="none" --parallelism.data_parallel_replicate_degree=2' \ --test-train-file=torchtitan.experiments.full_dtensor.train \ --steps=10 --assert-equal --no-seed-checkpoint ``` ``` [LOSS_COMPARE] [LOSS_COMPARE] Asserting losses are equal... [LOSS_COMPARE] Baseline log: /tmp/baseline_training.log [LOSS_COMPARE] Test log: /tmp/test_training.log [LOSS_COMPARE] Extracted 100 steps from baseline log [LOSS_COMPARE] Extracted 100 steps from test log test_losses_equal (__main__.assert_losses_equal.<locals>.LossEqualityTest.test_losses_equal) ... ok ---------------------------------------------------------------------- Ran 1 test in 0.000s OK ``` Note that, `--no-seed-checkpoint` is used because when seed-checkpoint is used, we got accuracy mismatch. ghstack-source-id: 6cf9b5e Pull-Request: #2049
fegin
added a commit
that referenced
this pull request
Nov 18, 2025
This PR provides a skelet This PR introduces an initial prototype and skeleton for fully DTensor-based training. The current codebase builds upon SimpleFSDP, but we anticipate developing our own Reparameterization to better serve our specific use case. There are several reasons why SimpleFSDP's Reparameterization is insufficient. For instance, the current parallelize_buffers() implementation in this PR will not function correctly when additional parallelization strategies are applied. Despite these limitations, this PR provides a starting point for experimenting with a full DTensor trainer. Accuracy verification: HSDP SimpleFSDP v.s. FSDP2 ``` python3 scripts/loss_compare.py . . \ --baseline-options='--activation_checkpoint.mode="none" --parallelism.data_parallel_replicate_degree=2' \ --test-options='--model.name full_dtensor.llama3 --activation_checkpoint.mode="none" --parallelism.data_parallel_replicate_degree=2' \ --test-train-file=torchtitan.experiments.full_dtensor.train \ --steps=10 --assert-equal --no-seed-checkpoint ``` ``` [LOSS_COMPARE] [LOSS_COMPARE] Asserting losses are equal... [LOSS_COMPARE] Baseline log: /tmp/baseline_training.log [LOSS_COMPARE] Test log: /tmp/test_training.log [LOSS_COMPARE] Extracted 100 steps from baseline log [LOSS_COMPARE] Extracted 100 steps from test log test_losses_equal (__main__.assert_losses_equal.<locals>.LossEqualityTest.test_losses_equal) ... ok ---------------------------------------------------------------------- Ran 1 test in 0.000s OK ``` Note that, `--no-seed-checkpoint` is used because when seed-checkpoint is used, we got accuracy mismatch. ghstack-source-id: 0d3e3f0 Pull-Request: #2049
fegin
added a commit
that referenced
this pull request
Nov 19, 2025
This PR provides a skelet This PR introduces an initial prototype and skeleton for fully DTensor-based training. The current codebase builds upon SimpleFSDP, but we anticipate developing our own Reparameterization to better serve our specific use case. There are several reasons why SimpleFSDP's Reparameterization is insufficient. For instance, the current parallelize_buffers() implementation in this PR will not function correctly when additional parallelization strategies are applied. Despite these limitations, this PR provides a starting point for experimenting with a full DTensor trainer. Accuracy verification: HSDP SimpleFSDP v.s. FSDP2 ``` python3 scripts/loss_compare.py . . \ --baseline-options='--activation_checkpoint.mode="none" --parallelism.data_parallel_replicate_degree=2' \ --test-options='--model.name full_dtensor.llama3 --activation_checkpoint.mode="none" --parallelism.data_parallel_replicate_degree=2' \ --test-train-file=torchtitan.experiments.full_dtensor.train \ --steps=10 --assert-equal --no-seed-checkpoint ``` ``` [LOSS_COMPARE] [LOSS_COMPARE] Asserting losses are equal... [LOSS_COMPARE] Baseline log: /tmp/baseline_training.log [LOSS_COMPARE] Test log: /tmp/test_training.log [LOSS_COMPARE] Extracted 100 steps from baseline log [LOSS_COMPARE] Extracted 100 steps from test log test_losses_equal (__main__.assert_losses_equal.<locals>.LossEqualityTest.test_losses_equal) ... ok ---------------------------------------------------------------------- Ran 1 test in 0.000s OK ``` Note that, `--no-seed-checkpoint` is used because when seed-checkpoint is used, we got accuracy mismatch. ghstack-source-id: c177628 Pull-Request: #2049
fegin
added a commit
that referenced
this pull request
Nov 19, 2025
This PR provides a skelet This PR introduces an initial prototype and skeleton for fully DTensor-based training. The current codebase builds upon SimpleFSDP, but we anticipate developing our own Reparameterization to better serve our specific use case. There are several reasons why SimpleFSDP's Reparameterization is insufficient. For instance, the current parallelize_buffers() implementation in this PR will not function correctly when additional parallelization strategies are applied. Despite these limitations, this PR provides a starting point for experimenting with a full DTensor trainer. Accuracy verification: HSDP SimpleFSDP v.s. FSDP2 ``` python3 scripts/loss_compare.py . . \ --baseline-options='--activation_checkpoint.mode="none" --parallelism.data_parallel_replicate_degree=2' \ --test-options='--model.name full_dtensor.llama3 --activation_checkpoint.mode="none" --parallelism.data_parallel_replicate_degree=2' \ --test-train-file=torchtitan.experiments.full_dtensor.train \ --steps=10 --assert-equal --no-seed-checkpoint ``` ``` [LOSS_COMPARE] [LOSS_COMPARE] Asserting losses are equal... [LOSS_COMPARE] Baseline log: /tmp/baseline_training.log [LOSS_COMPARE] Test log: /tmp/test_training.log [LOSS_COMPARE] Extracted 100 steps from baseline log [LOSS_COMPARE] Extracted 100 steps from test log test_losses_equal (__main__.assert_losses_equal.<locals>.LossEqualityTest.test_losses_equal) ... ok ---------------------------------------------------------------------- Ran 1 test in 0.000s OK ``` Note that, `--no-seed-checkpoint` is used because when seed-checkpoint is used, we got accuracy mismatch. ghstack-source-id: 955a260 Pull-Request: #2049
fegin
added a commit
that referenced
this pull request
Nov 19, 2025
This PR provides a skelet This PR introduces an initial prototype and skeleton for fully DTensor-based training. The current codebase builds upon SimpleFSDP, but we anticipate developing our own Reparameterization to better serve our specific use case. There are several reasons why SimpleFSDP's Reparameterization is insufficient. For instance, the current parallelize_buffers() implementation in this PR will not function correctly when additional parallelization strategies are applied. Despite these limitations, this PR provides a starting point for experimenting with a full DTensor trainer. Accuracy verification: HSDP SimpleFSDP v.s. FSDP2 ``` python3 scripts/loss_compare.py . . \ --baseline-options='--activation_checkpoint.mode="none" --parallelism.data_parallel_replicate_degree=2' \ --test-options='--model.name full_dtensor.llama3 --activation_checkpoint.mode="none" --parallelism.data_parallel_replicate_degree=2' \ --test-train-file=torchtitan.experiments.full_dtensor.train \ --steps=10 --assert-equal --no-seed-checkpoint ``` ``` [LOSS_COMPARE] [LOSS_COMPARE] Asserting losses are equal... [LOSS_COMPARE] Baseline log: /tmp/baseline_training.log [LOSS_COMPARE] Test log: /tmp/test_training.log [LOSS_COMPARE] Extracted 100 steps from baseline log [LOSS_COMPARE] Extracted 100 steps from test log test_losses_equal (__main__.assert_losses_equal.<locals>.LossEqualityTest.test_losses_equal) ... ok ---------------------------------------------------------------------- Ran 1 test in 0.000s OK ``` Note that, `--no-seed-checkpoint` is used because when seed-checkpoint is used, we got accuracy mismatch. ghstack-source-id: 92d8e21 Pull-Request: #2049
fegin
added a commit
that referenced
this pull request
Nov 19, 2025
Stack from [ghstack](https://github.com/ezyang/ghstack/tree/0.12.0) (oldest at bottom): * #2049 * __->__ #2029 ## Summary This PR adds `scripts/loss_compare.py` for comparing training losses between different git commits and/or training configurations. ## Key Features - Commit Comparison: Compare losses between two different git commits with deterministic training - Configuration Comparison: Compare different training configurations on the same commit - Reproducibility: Automatically enables deterministic mode and seed checkpointing for reproducible comparisons - Real-time Output: Streams training output to both console and log files during execution - Statistical Analysis: Generates step-by-step loss comparisons and summary statistics - CI Testing: Includes --assert-equal flag for automated testing to verify identical losses ## Usage Examples #### Compare two commits ``` python3 ./scripts/loss_compare.py main my_branch ``` #### Compare two commits with custom configuration ``` python3 ./scripts/loss_compare.py main my_branch \ --baseline-config="./custom.toml" --baseline-options="--parallelism.tensor_parallel_degree=2" \ ``` #### Compare different parallelization strategies on same commit ``` python3 ./scripts/loss_compare.py . . \ --baseline-config="./llama3_8b.toml" --baseline-options="--parallelism.tensor_parallel_degree=2" \ --test-options="--parallelism.tensor_parallel_degree=1" \ ``` #### Assert equality for CI testing ``` python3 ./scripts/loss_compare.py main my_branch --assert-equal ``` ## Real Use Cases Compare full dtensor simple fsdp with fsdp2: ``` python3 scripts/loss_compare.py . . \ --baseline-options='--activation_checkpoint.mode="none"' \ --test-train-file='torchtitan.experiments.full_dtensor.train' \ --test-options='--model.name full_dtensor.llama3 --activation_checkpoint.mode="none"' \ --assert-equal --no-seed-checkpoint [LOSS_COMPARE] [LOSS_COMPARE] Asserting losses are equal... [LOSS_COMPARE] Baseline log: /tmp/baseline_training.log [LOSS_COMPARE] Test log: /tmp/test_training.log [LOSS_COMPARE] Extracted 100 steps from baseline log [LOSS_COMPARE] Extracted 100 steps from test log test_losses_equal (__main__.assert_losses_equal.<locals>.LossEqualityTest.test_losses_equal) ... ok ```
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.
Stack from ghstack (oldest at bottom):
This PR provides a skelet
This PR introduces an initial prototype and skeleton for fully DTensor-based training. The current codebase builds upon SimpleFSDP, but we anticipate developing our own parameterization to better serve our specific use case. There are several reasons why SimpleFSDP's parameterization is insufficient. For instance, the current parallelize_buffers() implementation in this PR will not function correctly when additional parallelization strategies are applied. Despite these limitations, this PR provides a starting point for experimenting with a full DTensor trainer.
Accuracy verification:
HSDP
SimpleFSDP v.s. FSDP2
Note that,
--no-seed-checkpointis used because when seed-checkpoint is used, we got accuracy mismatch.