Add PSR-3 compliant logging interface and default StreamLogger (#75)

* feat: implement configurable logging interface with PSR-3 support * feat: rename StdoutLogger to StreamLogger and add tests for the StreamLogger * fix: Possible null reference for logger in helpers * feat: Update docs on logger * feat: Update the StreamLogger to work with multiple streams
CodeWithKyrian · Jan 11, 2025 · 6547841 · 6547841
1 parent 920ae45
commit 6547841
Show file tree

Hide file tree

Showing 14 changed files with 460 additions and 36 deletions.
diff --git a/README.md b/README.md
@@ -152,7 +152,7 @@ Transformers::setup()
     ->setAuthToken('...') // Set the auth token for downloading models. Defaults to `null`
     ->setUserAgent('...') // Set the user agent for downloading models. Defaults to `transformers-php/{version}`
     ->setImageDriver('...') // Set the image driver for processing images. Defaults to `IMAGICK'
-    ->apply(); // Apply the configuration
+    ->setLogger('...'); // Set the logger for TransformersPHP. Defaults to `null`
 ```
 
 You can call the `set` methods in any order, or leave any out entirely, in which case, it uses the default values. For
@@ -399,4 +399,4 @@ This package is a WIP, but here's a list of tasks and architectures currently te
 1. **[YOLOS](https://huggingface.co/docs/transformers/model_doc/yolos)** (from Huazhong University of Science &
    Technology) released with the
    paper [You Only Look at One Sequence: Rethinking Transformer in Vision through Object Detection](https://arxiv.org/abs/2106.00666)
-   by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
+   by Yuxin Fang, Bencheng Liao, Xinggang Wang, Jiemin Fang, Jiyang Qi, Rui Wu, Jianwei Niu, Wenyu Liu.
diff --git a/composer.json b/composer.json
@@ -15,6 +15,7 @@
   "require": {
     "php": "^8.1",
     "ext-ffi": "*",
+    "psr/log": "^1.1.3|^2.0|^3.0",
     "codewithkyrian/jinja-php": "^1.0",
     "codewithkyrian/transformers-libsloader": "^2.0",
     "imagine/imagine": "^1.3",

diff --git a/docs/configuration.md b/docs/configuration.md
@@ -22,7 +22,8 @@ Transformers::setup()
         ->setRemotePathTemplate('custom/path/{model}/{file}')
         ->setAuthToken('your-token')
         ->setUserAgent('your-user-agent')
-        ->setImageDriver(ImageDriver::IMAGICK);
+        ->setImageDriver(ImageDriver::IMAGICK)
+        ->setLogger(new StreamLogger('transformers-php'));
 ```
 
 ::: tip
@@ -105,6 +106,14 @@ Transformers::setup()
     ->apply();
 ```
 
+### `setLogger(LoggerInterface $logger)`
+
+This setting allows you to set a custom logger for TransformersPHP. No logger is set by default, but you can set a
+logger to debug TransformersPHP's internal behavior. The logger should implement the `Psr\Log\LoggerInterface` interface. TransformersPHP
+comes with a `StreamLogger` class, similar to Monolog's `StreamHandler`, which can be used to log to a stream (STDOUT, STDERR,
+or a file) and can be customized to log at different levels (debug, info, warning, error, critical). You can also pass in a 
+logger that is already configured and ready to use e.g. a Laravel logger.
+
 ## Standalone PHP Projects
 
 In a standalone PHP project, the best place to add global configuration is in your project's bootstrap or initialization

diff --git a/examples/bootstrap.php b/examples/bootstrap.php
@@ -4,9 +4,11 @@
 
 use Codewithkyrian\Transformers\Transformers;
 use Codewithkyrian\Transformers\Utils\ImageDriver;
+use Codewithkyrian\Transformers\Utils\StreamLogger;
 
 require_once './vendor/autoload.php';
 
 Transformers::setup()
     ->setCacheDir('/Users/Kyrian/.transformers')
-    ->setImageDriver(ImageDriver::VIPS);
+    ->setImageDriver(ImageDriver::VIPS)
+    ->setLogger(new StreamLogger(STDOUT));
diff --git a/examples/pipelines/asr.php b/examples/pipelines/asr.php
@@ -26,9 +26,9 @@
 //$audioUrl = __DIR__ . '/../sounds/sample-1.mp3';
 
 $streamer = WhisperTextStreamer::make()
-//->onTimestampStart(fn($timestamp) => dump($timestamp));
 ->onStream(fn($text) => print($text));
 
+
 $output = $transcriber($audioUrl,
     maxNewTokens: 256,
     chunkLengthSecs: 24,

diff --git a/src/Models/Auto/PretrainedMixin.php b/src/Models/Auto/PretrainedMixin.php
@@ -8,6 +8,7 @@
 use Codewithkyrian\Transformers\Exceptions\UnsupportedModelTypeException;
 use Codewithkyrian\Transformers\Models\ModelArchitecture;
 use Codewithkyrian\Transformers\Models\Pretrained\PretrainedModel;
+use Codewithkyrian\Transformers\Transformers;
 use Codewithkyrian\Transformers\Utils\AutoConfig;
 
 /**
@@ -18,6 +19,7 @@ abstract class PretrainedMixin
 {
     /**
      * Mapping from model type to model class.
+     *
      * @var array<string, array<string, string>> The model class mappings.
      */
     const MODEL_CLASS_MAPPINGS = [];
@@ -37,15 +39,16 @@ abstract class PretrainedMixin
      * @param string|null $cacheDir The cache directory to save the model in.
      * @param string $revision The revision of the model.
      * @param string|null $modelFilename The filename of the model.
+     *
      * @return PretrainedModel The instantiated pretrained model.
      */
     public static function fromPretrained(
-        string           $modelNameOrPath,
-        bool             $quantized = true,
-        ?array           $config = null,
-        ?string          $cacheDir = null,
-        string           $revision = 'main',
-        ?string          $modelFilename = null,
+        string    $modelNameOrPath,
+        bool      $quantized = true,
+        ?array    $config = null,
+        ?string   $cacheDir = null,
+        string    $revision = 'main',
+        ?string   $modelFilename = null,
         ?callable $onProgress = null
     ): PretrainedModel
     {
@@ -54,7 +57,6 @@ public static function fromPretrained(
         foreach (static::MODEL_CLASS_MAPPINGS as $modelClassMapping) {
             $modelClass = $modelClassMapping[$config->modelType] ?? null;
 
-
             if ($modelClass === null) continue;
 
             $modelArchitecture = self::getModelArchitecture($modelClass);
@@ -72,7 +74,7 @@ public static function fromPretrained(
         }
 
         if (static::BASE_IF_FAIL) {
-//            echo "Unknown model class for model type {$config->modelType}. Using base class PreTrainedModel.";
+            Transformers::getLogger()->warning("Unknown model class for model type {$config->modelType}. Using base class PreTrainedModel.");
 
             return PretrainedModel::fromPretrained(
                 modelNameOrPath: $modelNameOrPath,
@@ -109,4 +111,4 @@ protected static function getModelArchitecture($modelClass): ModelArchitecture
             default => ModelArchitecture::EncoderOnly,
         };
     }
-}
+}
diff --git a/src/Models/Pretrained/PretrainedModel.php b/src/Models/Pretrained/PretrainedModel.php
@@ -25,6 +25,7 @@
 use Codewithkyrian\Transformers\Models\ModelArchitecture;
 use Codewithkyrian\Transformers\Models\Output\ModelOutput;
 use Codewithkyrian\Transformers\Tensor\Tensor;
+use Codewithkyrian\Transformers\Transformers;
 use Codewithkyrian\Transformers\Utils\AutoConfig;
 use Codewithkyrian\Transformers\Utils\GenerationConfig;
 use Codewithkyrian\Transformers\Utils\Hub;
@@ -49,9 +50,7 @@ public function __construct(
         public InferenceSession  $session,
         public ModelArchitecture $modelArchitecture = ModelArchitecture::EncoderOnly,
                                  ...$args
-    )
-    {
-    }
+    ) {}
 
 
     /**
@@ -72,6 +71,7 @@ public function __construct(
      * @param string $revision The specific model version to use. It can be a branch name, a tag name,
      * @param string|null $modelFilename The name of the model file to load. If not provided, will default to the
      * @param ModelArchitecture $modelArchitecture
+     *
      * @return self The model instantiated from the configuration.
      * @throws HubException
      */
@@ -217,7 +217,7 @@ public static function fromPretrained(
             default:
             {
                 if ($modelArchitecture != ModelArchitecture::EncoderOnly) {
-                    echo "WARNING: {$modelArchitecture->value} is not a valid model group. Defaulting to EncoderOnly.";
+                    Transformers::getLogger()?->warning("{$modelArchitecture->value} is not a valid model group. Defaulting to EncoderOnly.");
                 }
 
 
@@ -251,6 +251,7 @@ public static function fromPretrained(
      * @param bool $fatal Whether to raise an error if the file could not be loaded.
      * @param callable|null $onProgress
      * @param mixed ...$sessionOptions
+     *
      * @return InferenceSession|null
      * @throws HubException
      */
@@ -283,7 +284,9 @@ public function __invoke(array $modelInputs): array|ModelOutput
     /**
      * Forward method for a pretrained model. If not overridden by a subclass, the correct forward method
      *  will be chosen based on the model type.
+     *
      * @param array $modelInputs The input data to the model in the format specified in the ONNX model.
+     *
      * @return array{logits: Tensor, hidden_states: Tensor, attentions: Tensor} The output data from the model in the format specified in the ONNX model.
      */
     public function forward(array $modelInputs): array
@@ -315,6 +318,7 @@ public function runSession(InferenceSession $session, array $inputs): array
     /**
      * @param InferenceSession $session
      * @param Tensor[] $inputs
+     *
      * @return Tensor[]
      * @throws MissingModelInputException
      */
@@ -345,20 +349,27 @@ public function validateInputs(array $inputNames, array $inputs): array
 
 
         if ($numInputsProvided > $numInputsNeeded) {
-            // No missing inputs, but too many inputs were provided.
-            // Warn the user and ignore the extra inputs.
+            // No missing inputs, but too many inputs were provided so we warn the user and ignore the extra inputs.
             $ignored = array_diff(array_keys($inputs), $inputNames);
-            echo 'WARNING: Too many inputs were provided (' . $numInputsProvided . ' > ' . $numInputsNeeded . '). 
-            The following inputs will be ignored: "' . implode(', ', $ignored) . '".';
+
+            $warning = sprintf(
+                'Too many inputs were provided (%d > %d). The following inputs will be ignored: "%s".',
+                $numInputsProvided,
+                $numInputsNeeded,
+                implode(', ', $ignored)
+            );
+
+            Transformers::getLogger()->warning($warning);
         }
 
-//        return array_map(fn($i) => $i->toArray(), $inputs);
         return $inputs;
     }
 
     /**
      * Prepares an attention mask for a sequence of tokens based on configuration options.
+     *
      * @param Tensor $tokens The input tokens.
+     *
      * @return Tensor The attention mask tensor.
      * @private
      */
@@ -379,7 +390,7 @@ public function prepareAttentionMask(Tensor $tokens): Tensor
         if ($isPadTokenInInputs && $isPadTokenNotEqualToEosTokenId) {
             $mo = Tensor::mo();
 
-            $data = $mo->f(fn($x) => $x != $padTokenId, $tokens);
+            $data = $mo->f(fn ($x) => $x != $padTokenId, $tokens);
 
             return new Tensor($data, $tokens->dtype(), $tokens->shape());
         } else {
@@ -389,9 +400,11 @@ public function prepareAttentionMask(Tensor $tokens): Tensor
 
     /**
      * Add position IDs to the feeds object.
+     *
      * @param array $inputNames The names of the inputs to the model.
      * @param array $feeds The input to the model.
      * @param bool $useCacheBranch Whether to use the cache branch of the model.
+     *
      * @return void
      */
     public function preparePositionIds(array $inputNames, array &$feeds, bool $useCacheBranch): void
@@ -430,6 +443,7 @@ public function preparePositionIds(array $inputNames, array &$feeds, bool $useCa
      *
      * @param array $decoderResults The decoder results object.
      * @param ?array $pastKeyValues The previous past key values.
+     *
      * @return array An object containing past key values.
      */
     public function getPastKeyValues(array $decoderResults, ?array $pastKeyValues): array
@@ -458,6 +472,7 @@ public function getPastKeyValues(array $decoderResults, ?array $pastKeyValues):
      * Returns an object containing attentions from the given decoder results object.
      *
      * @param array $decoderResults The decoder results object.
+     *
      * @return array An object containing attentions.
      */
     public function getAttentions(array $decoderResults): array
@@ -540,11 +555,13 @@ public function addPastKeyValues(array &$decoderFeeds, ?array $pastKeyValues): v
     }
 
     /** Generates text based on the given inputs and generation configuration using the model.
+     *
      * @param Tensor $inputs The input token ids.
      * @param GenerationConfig|null $generationConfig The generation configuration to use. If null, default configuration will be used.
      * @param LogitsProcessorList|null $logitsProcessor An optional logits processor to use. If null, a new LogitsProcessorList instance will be created.
      * @param Tensor|null $inputsAttentionMask An optional attention mask for the inputs.
      * @param Streamer|null $streamer
+     *
      * @return array An array of generated output sequences, where each sequence is an array of token IDs.
      * @throws Exception
      */
@@ -615,7 +632,7 @@ public function generate(
 
         $beams = $this->getStartBeams($inputs, $generationConfig, $numOutputTokens, $inputsAttentionMask);
 
-        while (array_some($beams, fn($beam) => !$beam['done']) && $numOutputTokens < $maxOutputTokens) {
+        while (array_some($beams, fn ($beam) => !$beam['done']) && $numOutputTokens < $maxOutputTokens) {
             $newestBeams = [];
             foreach ($beams as $beam) {
                 if ($beam['done']) {
@@ -676,7 +693,7 @@ public function generate(
             // Group and select best beams
             $newestBeams = array_merge(...array_map(
                 function ($group) use ($generationConfig) {
-                    usort($group, fn($a, $b) => $b['score'] <=> $a['score']);
+                    usort($group, fn ($a, $b) => $b['score'] <=> $a['score']);
                     return array_slice(
                         $group,
                         0,
@@ -702,7 +719,7 @@ function ($group) use ($generationConfig) {
                 function ($batch) use ($key, $generationConfig) {
                     if ($generationConfig->num_return_sequences > 1) {
                         return array_slice(
-                            array_map(fn($beam) => $beam[$key], $batch),
+                            array_map(fn ($beam) => $beam[$key], $batch),
                             0,
                             $generationConfig->num_return_sequences
                         );
@@ -752,7 +769,9 @@ function ($batch) use ($key, $generationConfig) {
     /**
      * This function merges multiple generation configs together to form a final generation config to be used by the model for text generation.
      * It first creates an empty `GenerationConfig` object, then it applies the model's own `generation_config` property to it. Finally, if a `generation_config` object was passed in the arguments, it overwrites the corresponding properties in the final config with those of the passed config object.
+     *
      * @param ?GenerationConfig $generationConfig A `GenerationConfig` object containing generation parameters.
+     *
      * @return GenerationConfig The final generation config object to be used by the model for text generation.
      */
     protected function getGenerationConfig(?GenerationConfig $generationConfig): GenerationConfig
@@ -854,6 +873,7 @@ protected function getLogitsProcessor(
      * @param GenerationConfig $generationConfig The generation config.
      * @param int $numOutputTokens The number of tokens to generate.
      * @param Tensor|null $inputsAttentionMask The attention mask for the input token ids.
+     *
      * @return array{ inputs: Tensor, output_token_ids: Tensor, score: float, done: bool, id: int } The initial beam for text generation.
      *
      */
@@ -877,6 +897,7 @@ public function getStartBeams(
      *  Runs the beam for text generation task
      *
      * @param array $beam The current beam being generated.
+     *
      * @return array The updated beam after a single generation step.
      *
      */
@@ -890,14 +911,15 @@ public function runBeam(array &$beam): array
      *
      * @param array $beam
      * @param array $output
+     *
      * @throws Exception
      */
     public function addAttentionsToBeam(array &$beam, array $output): void
     {
         if ($this->config->isEncoderDecoder) {
             if (empty($output['cross_attentions'])) {
                 throw new Exception(
-                    "`output_attentions` is true, but the model did not produce cross-attentions. " .
+                    "`output_attentions` is true, but the model did not produce cross-attentions. ".
                     "This is most likely because the model was not exported with `output_attentions=True`."
                 );
             }
@@ -909,7 +931,7 @@ public function addAttentionsToBeam(array &$beam, array $output): void
 
         if (empty($output['decoder_attentions'])) {
             throw new Exception(
-                "`output_attentions` is true, but the model did not produce decoder-attentions. " .
+                "`output_attentions` is true, but the model did not produce decoder-attentions. ".
                 "This is most likely because the model was not exported with `output_attentions=True`."
             );
         }
@@ -935,6 +957,7 @@ public function updateBeam(array &$beam, int $newTokenId): void
      * Groups an array of beam objects by their ids.
      *
      * @param array $beams The array of beam objects to group.
+     *
      * @return array An array of arrays, where each inner array contains beam objects with the same id.
      */
     public function groupBeams(array $beams): array

diff --git a/src/Models/Pretrained/VisionEncoderDecoderModel.php b/src/Models/Pretrained/VisionEncoderDecoderModel.php
@@ -9,6 +9,7 @@
 use Codewithkyrian\Transformers\Models\Auto\AutoModel;
 use Codewithkyrian\Transformers\Models\Auto\AutoModelForCausalLM;
 use Codewithkyrian\Transformers\Models\ModelArchitecture;
+use Codewithkyrian\Transformers\Transformers;
 use Codewithkyrian\Transformers\Utils\AutoConfig;
 use Codewithkyrian\Transformers\Utils\GenerationConfig;
 use Codewithkyrian\Transformers\Utils\InferenceSession;
@@ -32,6 +33,7 @@ class VisionEncoderDecoderModel extends PretrainedModel
 
     /**
      * Creates a new instance of the `VisionEncoderDecoderModel` class.
+     *
      * @param AutoConfig $config The configuration array specifying the hyperparameters and other model settings.
      * @param mixed $session The ONNX session containing the encoder model.
      * @param InferenceSession $decoderMergedSession The ONNX session containing the merged decoder model.
@@ -60,7 +62,7 @@ public function __construct(
             ?? AutoModel::ENCODER_DECODER_MODEL_MAPPING[$encoderModelType];
 
         if (!$encoderModel) {
-            echo "Model type for encoder '{$encoderModelType}' not found, assuming encoder-only architecture. Please report this at https://github.com/CodeWithKyrian/transformers-php/issues/new/choose.";
+            Transformers::getLogger()?->warning("Model type for encoder '{$encoderModelType}' not found, assuming encoder-only architecture. Please report this at https://github.com/CodeWithKyrian/transformers-php/issues/new/choose.");
         }
 
         // Validate decoder