Merge branch 'main' into feat/split_with_threshold

haystack/components/evaluators/context_relevance.py

@@ -67,6 +67,7 @@ class ContextRelevanceEvaluator(LLMEvaluator):
     def __init__(
         self,
         examples: Optional[List[Dict[str, Any]]] = None,
+        progress_bar: bool = True,
         api: str = "openai",
         api_key: Secret = Secret.from_env_var("OPENAI_API_KEY"),
     ):
@@ -89,12 +90,13 @@ def __init__(
                     "statement_scores": [1],
                 },
             }]
+        :param progress_bar:
+            Whether to show a progress bar during the evaluation.
         :param api:
             The API to use for calling an LLM through a Generator.
             Supported APIs: "openai".
         :param api_key:
             The API key.
-
         """
         self.instructions = (
             "Your task is to judge how relevant the provided context is for answering a question. "
@@ -115,6 +117,7 @@ def __init__(
             examples=self.examples,
             api=self.api,
             api_key=self.api_key,
+            progress_bar=progress_bar,
         )
 
     @component.output_types(individual_scores=List[int], score=float, results=List[Dict[str, Any]])

haystack/components/evaluators/faithfulness.py

@@ -81,6 +81,7 @@ class FaithfulnessEvaluator(LLMEvaluator):
     def __init__(
         self,
         examples: Optional[List[Dict[str, Any]]] = None,
+        progress_bar: bool = True,
         api: str = "openai",
         api_key: Secret = Secret.from_env_var("OPENAI_API_KEY"),
     ):
@@ -104,6 +105,8 @@ def __init__(
                     "statement_scores": [1, 0],
                 },
             }]
+        :param progress_bar:
+            Whether to show a progress bar during the evaluation.
         :param api:
             The API to use for calling an LLM through a Generator.
             Supported APIs: "openai".
@@ -131,6 +134,7 @@ def __init__(
             examples=self.examples,
             api=self.api,
             api_key=self.api_key,
+            progress_bar=progress_bar,
         )
 
     @component.output_types(individual_scores=List[int], score=float, results=List[Dict[str, Any]])

haystack/components/evaluators/llm_evaluator.py

@@ -5,6 +5,8 @@
 import json
 from typing import Any, Dict, List, Tuple, Type
 
+from tqdm import tqdm
+
 from haystack import component, default_from_dict, default_to_dict
 from haystack.components.builders import PromptBuilder
 from haystack.components.generators import OpenAIGenerator
@@ -50,6 +52,7 @@ def __init__(
         inputs: List[Tuple[str, Type[List]]],
         outputs: List[str],
         examples: List[Dict[str, Any]],
+        progress_bar: bool = True,
         *,
         api: str = "openai",
         api_key: Secret = Secret.from_env_var("OPENAI_API_KEY"),
@@ -70,6 +73,8 @@ def __init__(
              `outputs` parameters.
             Each example is a dictionary with keys "inputs" and "outputs"
             They contain the input and output as dictionaries respectively.
+        :param progress_bar:
+            Whether to show a progress bar during the evaluation.
         :param api:
             The API to use for calling an LLM through a Generator.
             Supported APIs: "openai".
@@ -78,13 +83,13 @@ def __init__(
 
         """
         self.validate_init_parameters(inputs, outputs, examples)
-
         self.instructions = instructions
         self.inputs = inputs
         self.outputs = outputs
         self.examples = examples
         self.api = api
         self.api_key = api_key
+        self.progress_bar = progress_bar
 
         if api == "openai":
             self.generator = OpenAIGenerator(
@@ -173,7 +178,7 @@ def run(self, **inputs) -> Dict[str, Any]:
         list_of_input_names_to_values = [dict(zip(input_names, v)) for v in values]
 
         results = []
-        for input_names_to_values in list_of_input_names_to_values:
+        for input_names_to_values in tqdm(list_of_input_names_to_values, disable=not self.progress_bar):
             prompt = self.builder.run(**input_names_to_values)
             result = self.generator.run(prompt=prompt["prompt"])
 
@@ -243,6 +248,7 @@ def to_dict(self) -> Dict[str, Any]:
             examples=self.examples,
             api=self.api,
             api_key=self.api_key.to_dict(),
+            progress_bar=self.progress_bar,
         )
 
     @classmethod

haystack/components/readers/extractive.py

@@ -210,6 +210,7 @@ def _preprocess(
         """
         texts = []
         document_ids = []
+        document_contents = []
         for i, doc in enumerate(documents):
             if doc.content is None:
                 warnings.warn(
@@ -219,9 +220,11 @@ def _preprocess(
                 continue
             texts.append(doc.content)
             document_ids.append(i)
+            document_contents.append(doc.content)
+
         encodings_pt = self.tokenizer(  # type: ignore
             queries,
-            [document.content for document in documents],
+            document_contents,
             padding=True,
             truncation=True,
             max_length=max_seq_length,
@@ -571,6 +574,9 @@ def run(
         :raises ComponentError:
             If the component was not warmed up by calling 'warm_up()' before.
         """
+        if not documents:
+            return {"answers": []}
+
         queries = [query]  # Temporary solution until we have decided what batching should look like in v2
         nested_documents = [documents]
         if self.model is None:

haystack/components/websearch/serper_dev.py

@@ -124,7 +124,7 @@ def run(self, query: str) -> Dict[str, Union[List[Document], List[str]]]:
 
         # we get the snippet from the json result and put it in the content field of the document
         organic = [
-            Document(meta={k: v for k, v in d.items() if k != "snippet"}, content=d["snippet"])
+            Document(meta={k: v for k, v in d.items() if k != "snippet"}, content=d.get("snippet"))
             for d in json_result["organic"]
         ]
 

releasenotes/notes/reader-crash-no-docs-53085ce48baaae81.yaml

@@ -0,0 +1,4 @@
+---
+fixes:
+  - |
+    Return an empty list of answers when `ExtractiveReader` receives an empty list of documents instead of raising an exception.

releasenotes/notes/serperdev-more-robust-229ba25c8fc9306d.yaml

@@ -0,0 +1,4 @@
+---
+enhancements:
+  - |
+    Make the `SerperDevWebSearch` more robust when `snippet` is not present in the request response.

test/components/evaluators/test_llm_evaluator.py

@@ -206,6 +206,7 @@ def test_to_dict_default(self, monkeypatch):
                 "instructions": "test-instruction",
                 "inputs": [("predicted_answers", List[str])],
                 "outputs": ["score"],
+                "progress_bar": True,
                 "examples": [
                     {"inputs": {"predicted_answers": "Football is the most popular sport."}, "outputs": {"score": 0}}
                 ],
@@ -266,6 +267,7 @@ def test_to_dict_with_parameters(self, monkeypatch):
                 "instructions": "test-instruction",
                 "inputs": [("predicted_answers", List[str])],
                 "outputs": ["custom_score"],
+                "progress_bar": True,
                 "examples": [
                     {
                         "inputs": {"predicted_answers": "Damn, this is straight outta hell!!!"},

test/components/readers/test_extractive.py

@@ -266,13 +266,17 @@ def test_from_dict_no_token():
     assert component.token is None
 
 
+def test_run_no_docs(mock_reader: ExtractiveReader):
+    mock_reader.warm_up()
+    assert mock_reader.run(query="hello", documents=[]) == {"answers": []}
+
+
 def test_output(mock_reader: ExtractiveReader):
-    answers = mock_reader.run(example_queries[0], example_documents[0], top_k=3)[
-        "answers"
-    ]  # [0] Uncomment and remove first two indices when batching support is reintroduced
+    answers = mock_reader.run(example_queries[0], example_documents[0], top_k=3)["answers"]
     doc_ids = set()
     no_answer_prob = 1
     for doc, answer in zip(example_documents[0], answers[:3]):
+        assert answer.document_offset is not None
         assert answer.document_offset.start == 11
         assert answer.document_offset.end == 16
         assert doc.content is not None

test/components/websearch/test_serperdev.py

@@ -1,6 +1,8 @@
 # SPDX-FileCopyrightText: 2022-present deepset GmbH <info@deepset.ai>
 #
 # SPDX-License-Identifier: Apache-2.0
+
+import json
 import os
 from unittest.mock import Mock, patch
 from haystack.utils.auth import Secret
@@ -111,6 +113,15 @@ def mock_serper_dev_search_result():
         yield mock_run
 
 
+@pytest.fixture
+def mock_serper_dev_search_result_no_snippet():
+    resp = {**EXAMPLE_SERPERDEV_RESPONSE}
+    del resp["organic"][0]["snippet"]
+    with patch("haystack.components.websearch.serper_dev.requests") as mock_run:
+        mock_run.post.return_value = Mock(status_code=200, json=lambda: resp)
+        yield mock_run
+
+
 class TestSerperDevSearchAPI:
     def test_init_fail_wo_api_key(self, monkeypatch):
         monkeypatch.delenv("SERPERDEV_API_KEY", raising=False)
@@ -142,6 +153,10 @@ def test_web_search_top_k(self, mock_serper_dev_search_result, top_k: int):
         assert all(isinstance(link, str) for link in links)
         assert all(link.startswith("http") for link in links)
 
+    def test_no_snippet(self, mock_serper_dev_search_result_no_snippet):
+        ws = SerperDevWebSearch(api_key=Secret.from_token("test-api-key"), top_k=1)
+        ws.run(query="Who is the boyfriend of Olivia Wilde?")
+
     @patch("requests.post")
     def test_timeout_error(self, mock_post):
         mock_post.side_effect = Timeout