Reboot Agents (#30387)

* Create CodeAgent and ReactAgent

* Fix formatting errors

* Update documentation for agents

* Add custom errors, improve logging

* Support variable usage in ReactAgent

* add messages

* Add message passing format

* Create React Code Agent

* Update

* Refactoring

* Fix errors

* Improve python interpreter

* Only non-tensor inputs should be sent to device

* Calculator tool slight refactor

* Improve docstrings

* Refactor

* Fix tests

* Fix more tests

* Fix even more tests

* Fix tests by replacing output and input types

* Fix operand type issue

* two small fixes

* EM TTS

* Fix agent running type errors

* Change text to speech tests to allow changed outputs

* Update doc with new agent types

* Improve code interpreter

* If max iterations reached, provide a real answer instead of an error

* Add edge case in interpreter

* Add safe imports to the interpreter

* Interpreter tweaks: tuples and listcomp

* Ma...

src/transformers/tools/speech_to_text.py → src/transformers/agents/speech_to_text.py

 #!/usr/bin/env python
 # coding=utf-8
 
-# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -14,28 +14,26 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
 from ..models.whisper import WhisperForConditionalGeneration, WhisperProcessor
-from .base import PipelineTool
+from .tools import PipelineTool
 
 
 class SpeechToTextTool(PipelineTool):
-    default_checkpoint = "openai/whisper-base"
-    description = (
-        "This is a tool that transcribes an audio into text. It takes an input named `audio` and returns the "
-        "transcribed text."
-    )
+    default_checkpoint = "distil-whisper/distil-large-v3"
+    description = "This is a tool that transcribes an audio into text. It returns the transcribed text."
     name = "transcriber"
     pre_processor_class = WhisperProcessor
     model_class = WhisperForConditionalGeneration
 
-    inputs = ["audio"]
-    outputs = ["text"]
+    inputs = {"audio": {"type": "audio", "description": "The audio to transcribe"}}
+    output_type = "text"
 
     def encode(self, audio):
-        return self.pre_processor(audio, return_tensors="pt").input_features
+        return self.pre_processor(audio, return_tensors="pt")
 
     def forward(self, inputs):
-        return self.model.generate(inputs=inputs)
+        return self.model.generate(inputs["input_features"])
 
     def decode(self, outputs):
         return self.pre_processor.batch_decode(outputs, skip_special_tokens=True)[0]

src/transformers/tools/text_to_speech.py → src/transformers/agents/text_to_speech.py

 #!/usr/bin/env python
 # coding=utf-8
 
-# Copyright 2023 The HuggingFace Inc. team. All rights reserved.
+# Copyright 2024 The HuggingFace Inc. team. All rights reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -14,11 +14,12 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+
 import torch
 
 from ..models.speecht5 import SpeechT5ForTextToSpeech, SpeechT5HifiGan, SpeechT5Processor
 from ..utils import is_datasets_available
-from .base import PipelineTool
+from .tools import PipelineTool
 
 
 if is_datasets_available():
@@ -28,16 +29,15 @@ if is_datasets_available():
 class TextToSpeechTool(PipelineTool):
     default_checkpoint = "microsoft/speecht5_tts"
     description = (
-        "This is a tool that reads an English text out loud. It takes an input named `text` which should contain the "
-        "text to read (in English) and returns a waveform object containing the sound."
+        "This is a tool that reads an English text out loud. It returns a waveform object containing the sound."
     )
-    name = "text_reader"
+    name = "text_to_speech"
     pre_processor_class = SpeechT5Processor
     model_class = SpeechT5ForTextToSpeech
     post_processor_class = SpeechT5HifiGan
 
-    inputs = ["text"]
-    outputs = ["audio"]
+    inputs = {"text": {"type": "text", "description": "The text to read out loud (in English)"}}
+    output_type = "audio"
 
     def setup(self):
         if self.post_processor is None:

src/transformers/tools/translation.py → src/transformers/agents/translation.py

@@ -15,7 +15,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 from ..models.auto import AutoModelForSeq2SeqLM, AutoTokenizer
-from .base import PipelineTool
+from .tools import PipelineTool
 
 
 LANGUAGE_CODES = {
@@ -231,27 +231,35 @@ class TranslationTool(PipelineTool):
     Example:
 
     ```py
-    from transformers.tools import TranslationTool
+    from transformers.agents import TranslationTool
 
     translator = TranslationTool()
     translator("This is a super nice API!", src_lang="English", tgt_lang="French")
     ```
     """
 
+    lang_to_code = LANGUAGE_CODES
     default_checkpoint = "facebook/nllb-200-distilled-600M"
     description = (
-        "This is a tool that translates text from a language to another. It takes three inputs: `text`, which should "
-        "be the text to translate, `src_lang`, which should be the language of the text to translate and `tgt_lang`, "
-        "which should be the language for the desired ouput language. Both `src_lang` and `tgt_lang` are written in "
-        "plain English, such as 'Romanian', or 'Albanian'. It returns the text translated in `tgt_lang`."
+        "This is a tool that translates text from a language to another."
+        f"Both `src_lang`and `tgt_lang` should belong to this list of languages: {list(lang_to_code.keys())}."
     )
     name = "translator"
     pre_processor_class = AutoTokenizer
     model_class = AutoModelForSeq2SeqLM
-    lang_to_code = LANGUAGE_CODES
 
-    inputs = ["text", "text", "text"]
-    outputs = ["text"]
+    inputs = {
+        "text": {"type": "text", "description": "The text to translate"},
+        "src_lang": {
+            "type": "text",
+            "description": "The language of the text to translate. Written in plain English, such as 'Romanian', or 'Albanian'",
+        },
+        "tgt_lang": {
+            "type": "text",
+            "description": "The language for the desired ouput language. Written in plain English, such as 'Romanian', or 'Albanian'",
+        },
+    }
+    output_type = "text"
 
     def encode(self, text, src_lang, tgt_lang):
         if src_lang not in self.lang_to_code: