OOMptimizer: bucketing batch size profiles to make GPUs go 🔥 (#9763)

* Initial working draft of the OOMptimizer. Signed-off-by: Piotr Żelasko <[email protected]> * Support model config. Add bucket merging. Signed-off-by: Piotr Żelasko <[email protected]> * fix Signed-off-by: Piotr Żelasko <[email protected]> * code review Signed-off-by: Piotr Żelasko <[email protected]> * Support bucket_batch_size option for lhotse dataloading Signed-off-by: Piotr Żelasko <[email protected]> * Ability to force a memory fraction to be unused in OOMptimizer Signed-off-by: Piotr Żelasko <[email protected]> * Ability to force a memory fraction to be unused in OOMptimizer Signed-off-by: Piotr Żelasko <[email protected]> * Fix for autocast and configurable dtype Signed-off-by: Piotr Żelasko <[email protected]> * Allow token-per-second filtering Signed-off-by: Piotr Żelasko <[email protected]> * Fix an issue with canary tokenizer Signed-off-by: Piotr Żelasko <[email protected]> * Lift the requirement to use CanaryTokenizer with canary prompt format * Fixes Signed-off-by: Piotr Żelasko <[email protected]> * Initial 2D bucketing draft Signed-off-by: Piotr Żelasko <[email protected]> * Separate script for 2D bucket estimation Signed-off-by: Piotr Żelasko <[email protected]> * Full 2D bucketing support: estimate_uduration_bins_2d, oomptimizer, training Signed-off-by: Piotr Żelasko <[email protected]> * fix Signed-off-by: Piotr Żelasko <[email protected]> * fix Signed-off-by: Piotr Żelasko <[email protected]> * fix Signed-off-by: Piotr Żelasko <[email protected]> * fix Signed-off-by: Piotr Żelasko <[email protected]> * fix Signed-off-by: Piotr Żelasko <[email protected]> * fix Signed-off-by: Piotr Żelasko <[email protected]> * Unit tests for bucket_batch_size and 2D bucketing for audio Signed-off-by: Piotr Żelasko <[email protected]> * Docs for 2D estimate duration bins Signed-off-by: Piotr Żelasko <[email protected]> * Fixes Signed-off-by: Piotr Żelasko <[email protected]> * Preliminary support for prompt format in estimate_duration_bins_2d Signed-off-by: Piotr Żelasko <[email protected]> * fixes Signed-off-by: Piotr Żelasko <[email protected]> * fix for bucket selection edge case Signed-off-by: Piotr Żelasko <[email protected]> * Add more info about the distribution to estimate_duration_bins_2d.py Signed-off-by: Piotr Żelasko <[email protected]> * Include CUDA RAM usage tracking in OOMptimizer Signed-off-by: Piotr Żelasko <[email protected]> * Track batch_size, num frames/tokens, and their padding ratio for AED multi task models Signed-off-by: Piotr Żelasko <[email protected]> * OOMptimizer documentation Signed-off-by: Piotr Żelasko <[email protected]> * Resolve TODOs and support any combination of (audio|text)->(audio|text) modalities Signed-off-by: Piotr Żelasko <[email protected]> * Add missing property decorator Signed-off-by: Piotr Żelasko <[email protected]> * fixes Signed-off-by: Piotr Żelasko <[email protected]> * Add docs about 2D bucketing with tokenizer and prompts Signed-off-by: Piotr Żelasko <[email protected]> * Fix bucket allocation logic for 2D bucketing Signed-off-by: Piotr Żelasko <[email protected]> * Bump lhotse version Signed-off-by: Piotr Żelasko <[email protected]> * fix... Signed-off-by: Piotr Żelasko <[email protected]> * Reverse bucket iteration order; move oomptimizer_schema to AsrModel Signed-off-by: Piotr Żelasko <[email protected]> * Make OOMptimizer compatible with dataclass mini-batches Signed-off-by: Piotr Żelasko <[email protected]> * Refine the schema Signed-off-by: Piotr Żelasko <[email protected]> * fixes after merging main Signed-off-by: Piotr Żelasko <[email protected]> * fix oomptimizer with pretrained models; verified canary, parakeet tdt and ctc Signed-off-by: Piotr Żelasko <[email protected]> * Disable concurrent bucketing to prevent spawning extra threads in tests Signed-off-by: Piotr Żelasko <[email protected]> * fix tests and make life more colorful Signed-off-by: Piotr Żelasko <[email protected]> * formatting Signed-off-by: Piotr Żelasko <[email protected]> * more reasonable starting batch size settings Signed-off-by: Piotr Żelasko <[email protected]> * Disable clearing of cuda memory cache Signed-off-by: Piotr Żelasko <[email protected]> * Even more conservative profile by incorporating DDP overhead simulation Signed-off-by: Piotr Żelasko <[email protected]> * Bucket selection fix and an extended unit test * Refactor registered_prompt_format_fn to enable prompt formatting before Sampler Signed-off-by: Piotr Żelasko <[email protected]> * porting fix Signed-off-by: Piotr Żelasko <[email protected]> * Fixes, move fast-path to prompted dataset Signed-off-by: Piotr Żelasko <[email protected]> * Changes from Daniel's review Signed-off-by: Piotr Żelasko <[email protected]> * OOMptimizer tests + fixes for 1D bucketing case Signed-off-by: Piotr Żelasko <[email protected]> * estimate duration bins tests Signed-off-by: Piotr Żelasko <[email protected]> * address Daniel's review Signed-off-by: Piotr Żelasko <[email protected]> * fix CPU unit test Signed-off-by: Piotr Żelasko <[email protected]> * try to fix CI test Signed-off-by: Piotr Żelasko <[email protected]> * Apply suggestions from code review Co-authored-by: oliver könig <[email protected]> Signed-off-by: Piotr Żelasko <[email protected]> * Disable 2D bucketing test with prompt due to quoting issue Signed-off-by: Piotr Żelasko <[email protected]> --------- Signed-off-by: Piotr Żelasko <[email protected]> Signed-off-by: Piotr Żelasko <[email protected]> Co-authored-by: oliver könig <[email protected]>
NVIDIA · Aug 16, 2024 · a2c1627 · a2c1627
1 parent 5f332bf
commit a2c1627
Show file tree

Hide file tree

Showing 24 changed files with 1,756 additions and 184 deletions.
diff --git a/.github/workflows/cicd-main.yml b/.github/workflows/cicd-main.yml
@@ -841,6 +841,67 @@ jobs:
       AFTER_SCRIPT: |
         rm -rf examples/asr/speech_to_text_adapters_mha_results
 
+  # L2: OOMptimizer
+  L2_Speech_Estimate_Duration_Bins:
+    needs: [cicd-test-container-setup]
+    uses: ./.github/workflows/_test_template.yml
+    with:
+      RUNNER: self-hosted-azure
+      SCRIPT: |
+        set -x
+        # 1D buckets [SSL, CTC]
+        python scripts/speech_recognition/estimate_duration_bins.py \
+          /home/TestData/an4_dataset/an4_train.json \
+          --buckets 5 
+        # 2D buckets [CTC, RNNT, TDT] / with tokenizer
+        python scripts/speech_recognition/estimate_duration_bins_2d.py \
+          /home/TestData/an4_dataset/an4_train_lang.json \
+          --tokenizer /home/TestData/asr_tokenizers/canary/en/tokenizer_spe_bpe_v1024_max_4/tokenizer.model \
+          --buckets 5 \
+          --sub-buckets 2
+        # TODO(pzelasko): Figure out how to quote the value in the test properly for CI to accept it...
+        # 2D buckets with prompt [AED/Canary, SpeechLM] / with aggregate tokenizer + prompt format 
+        # python scripts/speech_recognition/estimate_duration_bins_2d.py \
+        #   /home/TestData/an4_dataset/an4_train_lang.json \
+        #   --tokenizer /home/TestData/asr_tokenizers/canary/canary_spl_tokenizer_v32/tokenizer.model \
+        #      /home/TestData/asr_tokenizers/canary/en/tokenizer_spe_bpe_v1024_max_4/tokenizer.model \
+        #      /home/TestData/asr_tokenizers/canary/es/tokenizer_spe_bpe_v1024_max_4/tokenizer.model \
+        #   --langs spl_tokens en es \
+        #   --prompt-format canary \
+        #   --prompt '[{"role":"user","slots":{"source_lang":"en","target_lang":"en","task":"asr","pnc":"yes"}}]' \
+        #   --buckets 5 \
+        #   --sub-buckets 2
+
+  # L2: OOMptimizer
+  L2_Speech_Batch_Size_OOMptimizer:
+    needs: [cicd-test-container-setup]
+    uses: ./.github/workflows/_test_template.yml
+    with:
+      RUNNER: self-hosted-azure
+      SCRIPT: |
+        # 1D bucketing
+        python scripts/speech_recognition/oomptimizer.py \
+          -c /home/TestData/oomptimizer/fast-conformer_ctc_bpe.yaml \
+          -m nemo.collections.asr.models.EncDecCTCModelBPE \
+          -b "[5.0,10.0]"
+        # 2D bucketing
+        python scripts/speech_recognition/oomptimizer.py \
+          -c /home/TestData/oomptimizer/fast-conformer_ctc_bpe.yaml \
+          -m nemo.collections.asr.models.EncDecCTCModelBPE \
+          -b "[[5.0,30],[5.0,45],[10.0,57],[10.0,71]]"
+
+  # L2: OOMptimizer Canary (has a different batch schema)
+  L2_Speech_Batch_Size_OOMptimizer_Canary:
+    needs: [cicd-test-container-setup]
+    uses: ./.github/workflows/_test_template.yml
+    with:
+      RUNNER: self-hosted-azure
+      SCRIPT: |
+        python scripts/speech_recognition/oomptimizer.py \
+          -c /home/TestData/oomptimizer/fast-conformer_aed.yaml \
+          -m nemo.collections.asr.models.EncDecMultiTaskModel \
+          -b "[[5.0,30],[5.0,45],[10.0,57],[10.0,71]]"
+
   # L2: Speech Transcription
   L2_Speech_Transcription_Speech_to_Text_Transcribe:
     needs: [cicd-test-container-setup]

diff --git a/docs/source/asr/datasets.rst b/docs/source/asr/datasets.rst
@@ -803,21 +803,153 @@ The following script may be used:
 .. code-block:: bash
 
     $ python scripts/speech_recognition/estimate_duration_bins.py -b 30 manifest.json
+
+    # The script's output:
     Use the following options in your config:
             num_buckets=30
             bucket_duration_bins=[1.78,2.34,2.69,...
     <other diagnostic information about the dataset>
 
-For multi-dataset setups, one may provide multiple manifests and even their weights:
+For multi-dataset setups, one may provide a dataset config directly:
+
+.. code-block:: bash
+
+    $ python scripts/speech_recognition/estimate_duration_bins.py -b 30 input_cfg.yaml
+
+    # The script's output:
+    Use the following options in your config:
+            num_buckets=30
+            bucket_duration_bins=[1.91,3.02,3.56,...
+    <other diagnostic information about the dataset>
+
+It's also possible to manually specify the list of data manifests (optionally together with weights):
 
 .. code-block:: bash
 
     $ python scripts/speech_recognition/estimate_duration_bins.py -b 30 [[manifest.json,0.7],[other.json,0.3]]
+
+    # The script's output:
     Use the following options in your config:
             num_buckets=30
             bucket_duration_bins=[1.91,3.02,3.56,...
     <other diagnostic information about the dataset>
 
+2D bucketing
+~~~~~~~~~~~~
+
+To achieve maximum training efficiency for some classes of models it is necessary to stratify the sampling
+both on the input sequence lengths and the output sequence lengths.
+One such example are attention encoder-decoder models, where the overall GPU memory usage can be factorized
+into two main components: input-sequence-length bound (encoder activations) and output-sequence-length bound
+(decoder activations).
+Classical bucketing techniques only stratify on the input sequence length (e.g. duration in speech),
+which leverages encoder effectively but leads to excessive padding on on decoder's side.
+
+To amend this we support a 2D bucketing technique which estimates the buckets in two stages.
+The first stage is identical to 1D bucketing, i.e. we determine the input-sequence bucket bins so that
+every bin holds roughly an equal duration of audio.
+In the second stage, we use a tokenizer and optionally a prompt formatter (for prompted models) to
+estimate the total number of tokens in each duration bin, and sub-divide it into several sub-buckets,
+where each sub-bucket again holds roughly an equal number of tokens.
+
+To run 2D bucketing with 30 buckets sub-divided into 5 sub-buckets each (150 buckets total), use the following script:
+
+.. code-block:: bash
+
+    $ python scripts/speech_recognition/estimate_duration_bins_2d.py \
+        --tokenizer path/to/tokenizer.model \
+        --buckets 30 \
+        --sub-buckets 5 \
+        input_cfg.yaml
+
+    # The script's output:
+    Use the following options in your config:
+            num_buckets=30
+            bucket_duration_bins=[[1.91,10],[1.91,17],[1.91,25],...
+            max_duration=...
+            max_tps=...
+    <other diagnostic information about the dataset>
+
+Note that the output in ``bucket_duration_bins`` is a nested list, where every bin specifies
+the maximum duration and the maximum number of tokens that go into the bucket.
+Passing this option to Lhotse dataloader will automatically enable 2D bucketing.
+Note the presence of ``max_duration`` and ``max_tps`` (token-per-second) options:
+these need to be included in dataloader's configuration to ensure we can use the buckets correctly at runtime
+in case of outliers.
+In general, if you change your data in training, it is highly advisable to re-estimate the duration bins.
+
+Note that reasonable values for tokens-per-second rarely exceed 12tps with reasonably good tokenizers.
+If you find your dataset's TPS is much higher than that, you may have some bad data outliers.
+In that case you may specify ``--max_tps`` option to discard those both in bin estimation and dataloading.
+
+We also support aggregate tokenizers for 2D bucketing estimation:
+
+.. code-block:: bash
+
+    $ python scripts/speech_recognition/estimate_duration_bins_2d.py \
+        --tokenizer path/to/en/tokenizer.model path/to/pl/tokenizer1.model \
+        --langs en pl \
+        --buckets 30 \
+        --sub-buckets 5 \
+        input_cfg.yaml
+
+To estimate 2D buckets for a prompted model such as Canary-1B, provide prompt format name and an example prompt.
+For Canary-1B, we'll also provide the special tokens tokenizer. Example:
+
+.. code-block:: bash
+
+    $ python scripts/speech_recognition/estimate_duration_bins_2d.py \
+        --prompt-format canary \
+        --prompt "[{'role':'user','slots':{'source_lang':'en','target_lang':'de','task':'ast','pnc':'yes'}}]" \
+        --tokenizer path/to/spl_tokens/tokenizer.model path/to/en/tokenizer.model path/to/de/tokenizer1.model \
+        --langs spl_tokens en de \
+        --buckets 30 \
+        --sub-buckets 5 \
+        input_cfg.yaml
+
+Pushing GPU utilization to the limits with bucketing and OOMptimizer
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The default approach of specifying a ``batch_duration``, ``bucket_duration_bins`` and ``quadratic_duration``
+is quite flexible, but is not maximally efficient. We observed that in practice it often leads to under-utilization
+of GPU memory and compute for most buckets (especially those with shorter durations).
+While it is impossible to estimate GPU memory usage up-front, we can determine it empirically with a bit of search.
+
+OOMptimizer is an approach that given a NeMo model, optimizer, and a list of buckets (1D or 2D)
+estimates the maximum possible batch size to use for each bucket.
+It performs a binary search over batch sizes that succeed or lead to CUDA OOM until convergence.
+We find that the resulting bucketing batch size profiles enable full GPU utilization in training,
+while it only takes a couple of minutes to complete the search.
+
+In order to run OOMptimizer, you only need the bucketing bins (from previous sections) and a model configuration:
+
+.. code-block:: bash
+
+    $ python scripts/speech_recognition/oomptimizer.py \
+        --config-path fast-conformer_aed.yaml \
+        --module-name nemo.collections.asr.models.EncDecMultiTaskModel \
+        --buckets '[[3.975,30],[3.975,48],[4.97,37],[4.97,60],[5.851,42],[5.851,71],[6.563,46],[6.563,79],[7.32,49],[7.32,88],[8.19,54],[8.19,99],[8.88,61],[8.88,107],[9.75,66],[9.75,117],[10.55,72],[10.55,127],[11.21,76],[11.21,135],[11.87,79],[11.87,143],[12.54,82],[12.54,151],[13.08,87],[13.08,157],[13.62,91],[13.62,164],[14.16,93],[14.16,170],[14.7,96],[14.7,177],[15.19,99],[15.19,183],[15.67,101],[15.67,189],[16.13,103],[16.13,194],[16.66,105],[16.66,200],[17.2,108],[17.2,207],[17.73,111],[17.73,213],[18.2,114],[18.2,219],[18.69,117],[18.69,225],[19.15,120],[19.15,230],[19.62,123],[19.62,236],[20.264,122],[20.264,244],[32.547,173],[32.547,391],[36.587,227],[36.587,440],[40.0,253],[40.0,480]]'
+
+    # The script's output:
+    <output logs from the search>
+    The final profile is:
+            bucket_duration_bins=[[3.975,30],[3.975,48],[4.97,37],[4.97,60],[5.851,42],[5.851,71],[6.563,46],[6.563,79],[7.32,49],[7.32,88],[8.19,54],[8.19,99],[8.88,61],[8.88,107],[9.75,66],[9.75,117],[10.55,72],[10.55,127],[11.21,76],[11.21,135],[11.87,79],[11.87,143],[12.54,82],[12.54,151],[13.08,87],[13.08,157],[13.62,91],[13.62,164],[14.16,93],[14.16,170],[14.7,96],[14.7,177],[15.19,99],[15.19,183],[15.67,101],[15.67,189],[16.13,103],[16.13,194],[16.66,105],[16.66,200],[17.2,108],[17.2,207],[17.73,111],[17.73,213],[18.2,114],[18.2,219],[18.69,117],[18.69,225],[19.15,120],[19.15,230],[19.62,123],[19.62,236],[20.264,122],[20.264,244],[32.547,173],[32.547,391],[36.587,227],[36.587,440],[40.0,253],[40.0,480]]
+            bucket_batch_size=[352,308,280,245,245,206,206,180,186,163,168,142,151,132,136,119,126,106,116,98,110,92,104,88,99,83,94,79,90,76,86,72,86,72,81,68,80,65,78,63,74,60,72,58,70,58,68,54,66,52,65,52,62,50,37,28,31,24,28,21]
+            max_tps=12.0
+            max_duration=40.0
+
+Use the resulting options in your training configuration (typically under namespace ``model.train_ds``) to apply the profile.
+
+It's also possible to run OOMptimizer using a pretrained model's name and bucket bins corresponding
+to your fine-tuning data:
+
+    $ python scripts/speech_recognition/oomptimizer.py \
+        --pretrained-name nvidia/canary-1b \
+        --buckets '[2.0,3.1,5.6,6.6,...]'
+
+Note that your training script can perform some additional actions using GPU RAM that cannot be anticipated by the OOMptimizer.
+By default, we let the script use up to 90% of GPU's RAM for this estimation to account for that.
+In the unlikely case you run into an OutOfMemoryError during training, you can try re-estimating the profile with the option ``--memory-fraction 0.75`` (or another value) that will further cap OOMptimizer's available GPU RAM.
 
 Seeds and randomness
 ~~~~~~~~~~~~~~~~~~~~

diff --git a/nemo/collections/asr/data/audio_to_text_lhotse.py b/nemo/collections/asr/data/audio_to_text_lhotse.py
@@ -50,7 +50,19 @@ def __init__(self, tokenizer):
 
     def __getitem__(self, cuts) -> Tuple[torch.Tensor, ...]:
         audio, audio_lens, cuts = self.load_audio(cuts)
-        tokens = [torch.as_tensor(self.tokenizer(c.supervisions[0].text, c.supervisions[0].language)) for c in cuts]
+        tokens = [
+            torch.as_tensor(
+                sum(
+                    (
+                        # Supervisions may come pre-tokenized from the dataloader.
+                        s.tokens if hasattr(s, "tokens") else self.tokenizer(s.text, s.language)
+                        for s in c.supervisions
+                    ),
+                    start=[],
+                )
+            )
+            for c in cuts
+        ]
         token_lens = torch.tensor([t.size(0) for t in tokens], dtype=torch.long)
         tokens = collate_vectors(tokens, padding_value=0)
         return audio, audio_lens, tokens, token_lens