[DAP/Whisper] Remove redundant code.

examples/BuddyWhisper/README.md

@@ -1,7 +1,7 @@
 # Buddy Compiler WHISPER Example
 
 ## Introduction
-This example shows how to use Buddy Compiler to compile a WHISPER model to MLIR code then run it.  The [model](openai/whisper-base) is a pre-trained model for automatic speech recognition (ASR) and speech translation.
+This example shows how to use Buddy Compiler to compile a WHISPER model to MLIR code then run it.  The [model](openai/whisper-base) is a pre-trained model for automatic speech recognition (ASR) and speech translation (ST).
 
 
 ## How to run
@@ -13,15 +13,13 @@ This example shows how to use Buddy Compiler to compile a WHISPER model to MLIR
 $ export PYTHONPATH=/path-to-buddy-mlir/llvm/build/tools/mlir/python_packages/mlir_core:/path-to-buddy-mlir/build/python_packages:${PYTHONPATH}
 ```
 
-3. Set model and dataset environment variable.
+3. Set model environment variable.
 
 ```bash
 $ export WHISPER_MODEL_PATH=/path-to-whisper-model/
-$ export AUDIO_DATASET_PATH=/path-to-audio-dataset/
 
 // For example:
 $ export WHISPER_MODEL_PATH=/home/xxx/whisper-base
-$ export AUDIO_DATASET_PATH=/home/xxx/librispeech_asr_dummy
 ```
 
 4. Build and run the WHISPER example
@@ -33,4 +31,4 @@ $ cd bin
 $ ./buddy-whisper-run
 ```
 
-4. Enjoy it!
+5. Enjoy it!

examples/BuddyWhisper/import-whisper.py

@@ -14,16 +14,15 @@
 #
 # ===---------------------------------------------------------------------------
 #
-# This is the example of whisper model.
+# This is an example for whisper model.
 #
 # ===---------------------------------------------------------------------------
 
 import os
 import torch
 import torch._dynamo as dynamo
 from torch._inductor.decomposition import decompositions as inductor_decomp
-from transformers import WhisperProcessor, WhisperForConditionalGeneration
-from datasets import load_dataset
+from transformers import WhisperForConditionalGeneration
 import numpy
 
 from buddy.compiler.frontend import DynamoCompiler
@@ -36,22 +35,18 @@
 if model_path is None:
     model_path = "openai/whisper-base"
 
-# Initialize the tokenizer and model from the specified model path.
-processor = WhisperProcessor.from_pretrained(model_path)
+# Initialize the model from the specified model path.
 model = WhisperForConditionalGeneration.from_pretrained(model_path)
 model.config.use_cache = False
 
-ds = load_dataset("hf-internal-testing/librispeech_asr_dummy", "clean", split="validation")
-sample = ds[1]["audio"]
-input_features = processor(
-    sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt"
-).input_features
-
-decoder_input_ids = torch.tensor([[50258] * 448], dtype=torch.long)
+# Generate placeholder for inputs.
+input_features = torch.zeros(size=(1, 80, 3000), dtype=torch.float32)
+decoder_input_ids = torch.zeros(size=(1, 448), dtype=torch.long)
 inputs = {
     "input_features": input_features,
     "decoder_input_ids": decoder_input_ids,
 }
+
 # Initialize Dynamo Compiler with specific configurations as an importer.
 dynamo_compiler = DynamoCompiler(
     primary_registry=tosa.ops_registry,

examples/BuddyWhisper/whisper-main.cpp

@@ -13,6 +13,10 @@
 // limitations under the License.
 //
 //===----------------------------------------------------------------------===//
+//
+// This file implements an example for Whisper Model Inference. 
+//
+// ------------------------------------------------------------------------===//
 
 #include "whisper-main.h"
 
@@ -29,8 +33,6 @@ int main() {
   /// Define directories of vacabulary and parameter file.
   const std::string vocabDir = "../../examples/BuddyWhisper/vocab.txt";
   const std::string paramsDir = "../../examples/BuddyWhisper/arg0.data";
-  const std::string input_featuresDir =
-      "../../examples/BuddyWhisper/input_features.data";
 
   /// Initialize data containers
   //  - Result container
@@ -47,7 +49,8 @@ int main() {
 
   /// Fill data into containers
   //  - Output: register vocabulary.
-  //  - Parameters: generate audioInput from rawAudioData.
+  //  - Parameters: load parameters from the `arg0` file into the container.
+  //  - Input: generate audioInput from rawAudioData.
   outputContainer.loadVocab(vocabDir);
   loadParameters(paramsDir, paramsContainer);
   rawAudioData = std::move(MemRef<double, 1>(rawSpeech, inputShape));