AutoRAG 0.2.0__tar.gz → 0.2.1__tar.gz

{autorag-0.2.0 → autorag-0.2.1}/AutoRAG.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: AutoRAG
-Version: 0.2.0
+Version: 0.2.1
 Summary: Automatically Evaluate RAG pipelines with your own data. Find optimal structure for new RAG product.
 Author-email: Marker-Inc <vkehfdl1@gmail.com>
 License:                                  Apache License
@@ -250,6 +250,7 @@ Requires-Dist: FlagEmbedding
 Requires-Dist: ragas
 Requires-Dist: ray
 Requires-Dist: kiwipiepy
+Requires-Dist: llmlingua
 Requires-Dist: llama-index>=0.10.1
 Requires-Dist: llama-index-core>=0.10.1
 Requires-Dist: llama-index-embeddings-openai

{autorag-0.2.0 → autorag-0.2.1}/AutoRAG.egg-info/SOURCES.txt

@@ -65,6 +65,7 @@ autorag/nodes/passageaugmenter/prev_next_augmenter.py
 autorag/nodes/passageaugmenter/run.py
 autorag/nodes/passagecompressor/__init__.py
 autorag/nodes/passagecompressor/base.py
+autorag/nodes/passagecompressor/longllmlingua.py
 autorag/nodes/passagecompressor/pass_compressor.py
 autorag/nodes/passagecompressor/refine.py
 autorag/nodes/passagecompressor/run.py
@@ -105,6 +106,7 @@ autorag/nodes/promptmaker/run.py
 autorag/nodes/queryexpansion/__init__.py
 autorag/nodes/queryexpansion/base.py
 autorag/nodes/queryexpansion/hyde.py
+autorag/nodes/queryexpansion/multi_query_expansion.py
 autorag/nodes/queryexpansion/pass_query_expansion.py
 autorag/nodes/queryexpansion/query_decompose.py
 autorag/nodes/queryexpansion/run.py
@@ -195,6 +197,7 @@ docs/source/nodes/generator/openai_llm.md
 docs/source/nodes/generator/vllm.md
 docs/source/nodes/passage_augmenter/passage_augmenter.md
 docs/source/nodes/passage_augmenter/prev_next_augmenter.md
+docs/source/nodes/passage_compressor/longllmlingua.md
 docs/source/nodes/passage_compressor/passage_compressor.md
 docs/source/nodes/passage_compressor/refine.md
 docs/source/nodes/passage_compressor/tree_summarize.md
@@ -221,6 +224,7 @@ docs/source/nodes/prompt_maker/fstring.md
 docs/source/nodes/prompt_maker/long_context_reorder.md
 docs/source/nodes/prompt_maker/prompt_maker.md
 docs/source/nodes/query_expansion/hyde.md
+docs/source/nodes/query_expansion/multi_query_expansion.md
 docs/source/nodes/query_expansion/query_decompose.md
 docs/source/nodes/query_expansion/query_expansion.md
 docs/source/nodes/retrieval/bm25.md
@@ -282,6 +286,7 @@ tests/autorag/nodes/passageaugmenter/test_pass_passage_augmenter.py
 tests/autorag/nodes/passageaugmenter/test_prev_next_augmenter.py
 tests/autorag/nodes/passageaugmenter/test_run_passage_augmenter.py
 tests/autorag/nodes/passagecompressor/test_base_passage_compressor.py
+tests/autorag/nodes/passagecompressor/test_longllmlingua.py
 tests/autorag/nodes/passagecompressor/test_pass_compressor.py
 tests/autorag/nodes/passagecompressor/test_refine.py
 tests/autorag/nodes/passagecompressor/test_run_passage_compressor_node.py
@@ -314,6 +319,7 @@ tests/autorag/nodes/promptmaker/test_long_context_reorder.py
 tests/autorag/nodes/promptmaker/test_prompt_maker_base.py
 tests/autorag/nodes/promptmaker/test_prompt_maker_run.py
 tests/autorag/nodes/queryexpansion/test_hyde.py
+tests/autorag/nodes/queryexpansion/test_multi_query_expansion.py
 tests/autorag/nodes/queryexpansion/test_pass_query_expansion.py
 tests/autorag/nodes/queryexpansion/test_query_decompose.py
 tests/autorag/nodes/queryexpansion/test_query_expansion_base.py

{autorag-0.2.0 → autorag-0.2.1}/AutoRAG.egg-info/requires.txt

@@ -29,6 +29,7 @@ FlagEmbedding
 ragas
 ray
 kiwipiepy
+llmlingua
 llama-index>=0.10.1
 llama-index-core>=0.10.1
 llama-index-embeddings-openai

{autorag-0.2.0 → autorag-0.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: AutoRAG
-Version: 0.2.0
+Version: 0.2.1
 Summary: Automatically Evaluate RAG pipelines with your own data. Find optimal structure for new RAG product.
 Author-email: Marker-Inc <vkehfdl1@gmail.com>
 License:                                  Apache License
@@ -250,6 +250,7 @@ Requires-Dist: FlagEmbedding
 Requires-Dist: ragas
 Requires-Dist: ray
 Requires-Dist: kiwipiepy
+Requires-Dist: llmlingua
 Requires-Dist: llama-index>=0.10.1
 Requires-Dist: llama-index-core>=0.10.1
 Requires-Dist: llama-index-embeddings-openai

autorag-0.2.1/autorag/VERSION

@@ -0,0 +1 @@
+0.2.1

{autorag-0.2.0 → autorag-0.2.1}/autorag/nodes/passagecompressor/__init__.py

@@ -1,3 +1,4 @@
+from .longllmlingua import longllmlingua
 from .pass_compressor import pass_compressor
 from .refine import refine
 from .tree_summarize import tree_summarize

{autorag-0.2.0 → autorag-0.2.1}/autorag/nodes/passagecompressor/base.py

@@ -46,6 +46,15 @@ def passage_compressor_node(func):
             )
             del llm
             result = list(map(lambda x: [x], result))
+        elif func.__name__ == 'longllmlingua':
+            result = func(
+                queries=queries,
+                contents=retrieved_contents,
+                scores=retrieve_scores,
+                ids=retrieved_ids,
+                **kwargs
+            )
+            result = list(map(lambda x: [x], result))
         elif func.__name__ == 'pass_compressor':
             result = func(contents=retrieved_contents)
         else:

autorag-0.2.1/autorag/nodes/passagecompressor/longllmlingua.py

@@ -0,0 +1,87 @@
+from typing import List, Optional
+
+import torch
+from llmlingua import PromptCompressor
+
+from autorag.nodes.passagecompressor.base import passage_compressor_node
+
+
+@passage_compressor_node
+def longllmlingua(queries: List[str],
+                  contents: List[List[str]],
+                  scores,
+                  ids,
+                  model_name: str = "NousResearch/Llama-2-7b-hf",
+                  instructions: Optional[str] = None,
+                  target_token: int = 300,
+                  **kwargs,
+                  ) -> List[str]:
+    """
+    Compresses the retrieved texts using LongLLMLingua.
+    For more information, visit https://github.com/microsoft/LLMLingua.
+
+    :param queries: The queries for retrieved passages.
+    :param contents: The contents of retrieved passages.
+    :param scores: The scores of retrieved passages.
+        Do not use in this function, so you can pass an empty list.
+    :param ids: The ids of retrieved passages.
+        Do not use in this function, so you can pass an empty list.
+    :param model_name: The model name to use for compression.
+        Default is "NousResearch/Llama-2-7b-hf".
+    :param instructions: The instructions for compression.
+        Default is None. When it is None, it will use default instructions.
+    :param target_token: The target token for compression.
+        Default is 300.
+    :param kwargs: Additional keyword arguments.
+    :return: The list of compressed texts.
+    """
+    if instructions is None:
+        instructions = "Given the context, please answer the final question"
+    llm_lingua = PromptCompressor(
+        model_name=model_name,
+    )
+    results = [llmlingua_pure(query, contents_, llm_lingua, instructions, target_token, **kwargs)
+               for query, contents_ in zip(queries, contents)]
+
+    del llm_lingua
+    if torch.cuda.is_available():
+        torch.cuda.empty_cache()
+
+    return results
+
+
+def llmlingua_pure(query: str,
+                   contents: List[str],
+                   llm_lingua: PromptCompressor,
+                   instructions: str,
+                   target_token: int = 300,
+                   **kwargs,
+                   ) -> str:
+    """
+    Return the compressed text.
+
+    :param query: The query for retrieved passages.
+    :param contents: The contents of retrieved passages.
+    :param llm_lingua: The llm instance that will be used to compress.
+    :param instructions: The instructions for compression.
+    :param target_token: The target token for compression.
+        Default is 300.
+    :param kwargs: Additional keyword arguments.
+    :return: The compressed text.
+    """
+    # split by "\n\n" (recommended by LongLLMLingua authors)
+    new_context_texts = [c for context in contents for c in context.split("\n\n")]
+    compressed_prompt = llm_lingua.compress_prompt(
+        new_context_texts,
+        question=query,
+        instruction=instructions,
+        rank_method="longllmlingua",
+        target_token=target_token,
+        **kwargs,
+    )
+    compressed_prompt_txt = compressed_prompt["compressed_prompt"]
+
+    # separate out the question and instruction
+    result = '\n\n'.join(compressed_prompt_txt.split("\n\n")[1:-1])
+
+    return result

{autorag-0.2.0 → autorag-0.2.1}/autorag/nodes/queryexpansion/__init__.py

@@ -1,3 +1,4 @@
 from .hyde import hyde
+from .multi_query_expansion import multi_query_expansion
 from .pass_query_expansion import pass_query_expansion
 from .query_decompose import query_decompose

{autorag-0.2.0 → autorag-0.2.1}/autorag/nodes/queryexpansion/base.py

@@ -1,11 +1,11 @@
 import functools
 import logging
 from pathlib import Path
-from typing import List, Union
+from typing import List, Union, Dict, Optional
 
 import pandas as pd
 
-from autorag import generator_models
+from autorag.support import get_support_modules
 from autorag.utils import result_to_dataframe, validate_qa_dataset
 
 logger = logging.getLogger("AutoRAG")
@@ -28,31 +28,33 @@ def query_expansion_node(func):
         if func.__name__ == "pass_query_expansion":
             return func(queries=queries)
 
-        # set module parameters
-        llm_str = kwargs.pop("llm")
-
         # pop prompt from kwargs
         if "prompt" in kwargs.keys():
             prompt = kwargs.pop("prompt")
         else:
             prompt = ""
 
-        # pop batch from kwargs
-        if "batch" in kwargs.keys():
-            batch = kwargs.pop("batch")
-        else:
-            batch = 16
-
-        # set llm model for query expansion
-        if llm_str in generator_models:
-            llm = generator_models[llm_str](**kwargs)
-        else:
-            logger.error(f"llm_str {llm_str} does not exist.")
-            raise KeyError(f"llm_str {llm_str} does not exist.")
+        # set generator module for query expansion
+        generator_callable, generator_param = make_generator_callable_param(kwargs)
 
         # run query expansion function
-        expanded_queries = func(queries=queries, llm=llm, prompt=prompt, batch=batch)
-        del llm
+        expanded_queries = func(queries=queries,
+                                prompt=prompt,
+                                generator_func=generator_callable,
+                                generator_params=generator_param)
         return expanded_queries
 
     return wrapper
+
+
+def make_generator_callable_param(generator_dict: Optional[Dict]):
+    if 'generator_module_type' not in generator_dict.keys():
+        generator_dict = {
+            'generator_module_type': 'llama_index_llm',
+            'llm': 'openai',
+            'model': 'gpt-3.5-turbo',
+        }
+    module_str = generator_dict.pop('generator_module_type')
+    module_callable = get_support_modules(module_str)
+    module_param = generator_dict
+    return module_callable, module_param

autorag-0.2.1/autorag/nodes/queryexpansion/hyde.py

@@ -0,0 +1,31 @@
+from typing import List, Dict, Callable
+
+import pandas as pd
+
+from autorag.nodes.queryexpansion.base import query_expansion_node
+
+hyde_prompt = "Please write a passage to answer the question"
+
+
+@query_expansion_node
+def hyde(queries: List[str],
+         generator_func: Callable,
+         generator_params: Dict,
+         prompt: str = hyde_prompt) -> List[List[str]]:
+    """
+    HyDE, which inspired by "Precise Zero-shot Dense Retrieval without Relevance Labels" (https://arxiv.org/pdf/2212.10496.pdf)
+    LLM model creates a hypothetical passage.
+    And then, retrieve passages using hypothetical passage as a query.
+    :param queries: List[str], queries to retrieve.
+    :param generator_func: Callable, generator functions.
+    :param generator_params: Dict, generator parameters.
+    :param prompt: prompt to use when generating hypothetical passage
+    :return: List[List[str]], List of hyde results.
+    """
+    full_prompts = list(
+        map(lambda x: (prompt if not bool(prompt) else hyde_prompt) + f"\nQuestion: {x}\nPassage:", queries))
+    input_df = pd.DataFrame({"prompts": full_prompts})
+    result_df = generator_func(project_dir=None, previous_result=input_df, **generator_params)
+    answers = result_df['generated_texts'].tolist()
+    results = list(map(lambda x: [x], answers))
+    return results

autorag-0.2.1/autorag/nodes/queryexpansion/multi_query_expansion.py

@@ -0,0 +1,52 @@
+from typing import List, Dict, Callable
+
+import pandas as pd
+
+from autorag.nodes.queryexpansion.base import query_expansion_node
+
+multi_query_expansion_prompt = """You are an AI language model assistant. 
+    Your task is to generate 3 different versions of the given user 
+    question to retrieve relevant documents from a vector  database. 
+    By generating multiple perspectives on the user question, 
+    your goal is to help the user overcome some of the limitations 
+    of distance-based similarity search. Provide these alternative 
+    questions separated by newlines. Original question: {question}"""
+
+
+@query_expansion_node
+def multi_query_expansion(queries: List[str],
+                          generator_func: Callable,
+                          generator_params: Dict,
+                          prompt: str = multi_query_expansion_prompt) -> List[List[str]]:
+    """
+    Expand a list of queries using a multi-query expansion approach.
+    LLM model generate 3 different versions queries for each input query.
+
+    :param queries: List[str], queries to decompose.
+    :param generator_func: Callable, generator functions.
+    :param generator_params: Dict, generator parameters.
+    :param prompt: str, prompt to use for multi-query expansion.
+        default prompt comes from langchain MultiQueryRetriever default query prompt.
+    :return: List[List[str]], list of expansion query.
+    """
+    full_prompts = []
+    for query in queries:
+        if bool(prompt):
+            full_prompt = f"prompt: {prompt}\n\n question: {query}"
+        else:
+            full_prompt = multi_query_expansion_prompt.format(question=query)
+        full_prompts.append(full_prompt)
+    input_df = pd.DataFrame({"prompts": full_prompts})
+    result_df = generator_func(project_dir=None, previous_result=input_df, **generator_params)
+    answers = result_df['generated_texts'].tolist()
+    results = list(map(lambda x: get_multi_query_expansion(x[0], x[1]), zip(queries, answers)))
+    return results
+
+
+def get_multi_query_expansion(query: str, answer: str) -> List[str]:
+    try:
+        queries = answer.split("\n")
+        queries.insert(0, query)
+        return queries
+    except:
+        return [query]

{autorag-0.2.0 → autorag-0.2.1}/autorag/nodes/queryexpansion/query_decompose.py

@@ -1,10 +1,8 @@
-import asyncio
-from typing import List
+from typing import List, Callable, Dict
 
-from llama_index.core.service_context_elements.llm_predictor import LLMPredictorType
+import pandas as pd
 
 from autorag.nodes.queryexpansion.base import query_expansion_node
-from autorag.utils.util import process_batch
 
 decompose_prompt = """Decompose a question in self-contained sub-questions. Use \"The question needs no decomposition\" when no decomposition is needed.
 
@@ -54,44 +52,44 @@ decompose_prompt = """Decompose a question in self-contained sub-questions. Use
 
 
 @query_expansion_node
-def query_decompose(queries: List[str], llm: LLMPredictorType,
-                    prompt: str = decompose_prompt,
-                    batch: int = 16) -> List[List[str]]:
+def query_decompose(queries: List[str],
+                    generator_func: Callable,
+                    generator_params: Dict,
+                    prompt: str = decompose_prompt) -> List[List[str]]:
     """
     decompose query to little piece of questions.
     :param queries: List[str], queries to decompose.
-    :param llm: LLMPredictorType, language model to use.
+    :param generator_func: Callable, generator functions.
+    :param generator_params: Dict, generator parameters.
     :param prompt: str, prompt to use for query decomposition.
         default prompt comes from Visconde's StrategyQA few-shot prompt.
-    :param batch: int, batch size for llm.
-        Default is 16.
     :return: List[List[str]], list of decomposed query. Return input query if query is not decomposable.
     """
-    # Run async query_decompose_pure function
-    tasks = [query_decompose_pure(query, llm, prompt) for query in queries]
-    loop = asyncio.get_event_loop()
-    results = loop.run_until_complete(process_batch(tasks, batch_size=batch))
+    full_prompts = []
+    for query in queries:
+        if bool(prompt):
+            full_prompt = f"prompt: {prompt}\n\n question: {query}"
+        else:
+            full_prompt = decompose_prompt.format(question=query)
+        full_prompts.append(full_prompt)
+    input_df = pd.DataFrame({"prompts": full_prompts})
+    result_df = generator_func(project_dir=None, previous_result=input_df, **generator_params)
+    answers = result_df['generated_texts'].tolist()
+    results = list(map(lambda x: get_query_decompose(x[0], x[1]), zip(queries, answers)))
     return results
 
 
-async def query_decompose_pure(query: str, llm: LLMPredictorType,
-                               prompt: str = decompose_prompt) -> List[str]:
+def get_query_decompose(query: str, answer: str) -> List[str]:
     """
     decompose query to little piece of questions.
     :param query: str, query to decompose.
-    :param llm: LLMPredictorType, language model to use.
-    :param prompt: str, prompt to use for query decomposition.
-        default prompt comes from Visconde's StrategyQA few-shot prompt.
+    :param answer: str, answer from query_decompose function.
     :return: List[str], list of a decomposed query. Return input query if query is not decomposable.
     """
-    if prompt == "":
-        prompt = decompose_prompt
-    full_prompt = "prompt: " + prompt + "\n\n" "question: " + query
-    answer = await llm.acomplete(full_prompt)
-    if answer.text == "the question needs no decomposition.":
+    if answer.lower() == "the question needs no decomposition":
         return [query]
     try:
-        lines = [line.strip() for line in answer.text.splitlines() if line.strip()]
+        lines = [line.strip() for line in answer.splitlines() if line.strip()]
         if lines[0].startswith("Decompositions:"):
             lines.pop(0)
         questions = [line.split(':', 1)[1].strip() for line in lines if ':' in line]

{autorag-0.2.0 → autorag-0.2.1}/autorag/strategy.py

@@ -93,6 +93,7 @@ def select_best(results: List[pd.DataFrame],
     strategy_func_dict = {
         'mean': select_best_average,
         'rank': select_best_rr,
+        'normalize_mean': select_normalize_mean,
     }
     if strategy_name not in strategy_func_dict:
         raise ValueError(f'Input strategy name {strategy_name} is not in {strategy_func_dict.keys()}')
@@ -129,3 +130,13 @@ def select_best_rr(results: List[pd.DataFrame], columns: Iterable[str],
     rr_df = rank_df.applymap(lambda x: 1 / x)
     best_index = np.array(rr_df.sum(axis=1)).argmax()
     return results[best_index], metadatas[best_index]
+
+
+def select_normalize_mean(results: List[pd.DataFrame], columns: Iterable[str],
+                          metadatas: Optional[List[Any]] = None) -> Tuple[pd.DataFrame, Any]:
+    results, columns, metadatas = validate_strategy_inputs(results, columns, metadatas)
+    each_mean_df = pd.DataFrame([df[columns].mean(axis=0).to_dict() for df in results])
+    normalized_means = (each_mean_df - each_mean_df.min()) / (each_mean_df.max() - each_mean_df.min())
+    normalized_mean_sums = normalized_means.sum(axis=1)
+    best_index = normalized_mean_sums.argmax()
+    return results[best_index], metadatas[best_index]

{autorag-0.2.0 → autorag-0.2.1}/autorag/support.py

@@ -18,6 +18,7 @@ def get_support_modules(module_name: str) -> Callable:
         'query_decompose': ('autorag.nodes.queryexpansion', 'query_decompose'),
         'hyde': ('autorag.nodes.queryexpansion', 'hyde'),
         'pass_query_expansion': ('autorag.nodes.queryexpansion', 'pass_query_expansion'),
+        'multi_query_expansion': ('autorag.nodes.queryexpansion', 'multi_query_expansion'),
         # retrieval
         'bm25': ('autorag.nodes.retrieval', 'bm25'),
         'vectordb': ('autorag.nodes.retrieval', 'vectordb'),
@@ -53,6 +54,7 @@ def get_support_modules(module_name: str) -> Callable:
         'tree_summarize': ('autorag.nodes.passagecompressor', 'tree_summarize'),
         'pass_compressor': ('autorag.nodes.passagecompressor', 'pass_compressor'),
         'refine': ('autorag.nodes.passagecompressor', 'refine'),
+        'longllmlingua': ('autorag.nodes.passagecompressor', 'longllmlingua'),
         # prompt_maker
         'fstring': ('autorag.nodes.promptmaker', 'fstring'),
         'long_context_reorder': ('autorag.nodes.promptmaker', 'long_context_reorder'),

{autorag-0.2.0 → autorag-0.2.1}/docs/source/api_spec/autorag.nodes.passagecompressor.rst

@@ -12,6 +12,14 @@ autorag.nodes.passagecompressor.base module
    :undoc-members:
    :show-inheritance:
 
+autorag.nodes.passagecompressor.longllmlingua module
+----------------------------------------------------
+
+.. automodule:: autorag.nodes.passagecompressor.longllmlingua
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
 autorag.nodes.passagecompressor.pass\_compressor module
 -------------------------------------------------------
 

{autorag-0.2.0 → autorag-0.2.1}/docs/source/api_spec/autorag.nodes.queryexpansion.rst

@@ -20,6 +20,14 @@ autorag.nodes.queryexpansion.hyde module
    :undoc-members:
    :show-inheritance:
 
+autorag.nodes.queryexpansion.multi\_query\_expansion module
+-----------------------------------------------------------
+
+.. automodule:: autorag.nodes.queryexpansion.multi_query_expansion
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
 autorag.nodes.queryexpansion.pass\_query\_expansion module
 ----------------------------------------------------------
 

autorag-0.2.1/docs/source/nodes/passage_compressor/longllmlingua.md

@@ -0,0 +1,25 @@
+# Long LLM Lingua
+
+The `Long LLM Lingua` module is compressor based on [llmlingua](https://github.com/microsoft/LLMLingua).
+
+Compresses the retrieved texts using LongLLMLingua.
+
+## **Module Parameters**
+
+**model_name**: The name of the LLM to be used for compression, defaulting to "NousResearch/Llama-2-7b-hf".
+
+**instructions**: Optional instructions for the LLM, defaulting to "Given the context, please answer the final
+question".
+
+**target_token**: The target token count for the output, default to 300.
+
+- **Additional Parameters**:
+  You can put any additional parameters at llm_lingua.
+  Find additional parameters [here](https://github.com/microsoft/LLMLingua)
+
+## **Example config.yaml**
+
+```yaml
+modules:
+  - module_type: longllmlingua
+```

{autorag-0.2.0 → autorag-0.2.1}/docs/source/nodes/passage_compressor/passage_compressor.md

@@ -53,4 +53,5 @@ maxdepth: 1
 ---
 tree_summarize.md
 refine.md
+longllmlingua.md
 ```

autorag-0.2.1/docs/source/nodes/query_expansion/hyde.md

@@ -0,0 +1,27 @@
+# HyDE
+
+The HyDE is inspired by the paper "[Precise Zero-shot Dense Retrieval without Relevance Labels](https://arxiv.org/abs/2212.10496)". It uses a LLM model to generate a hypothetical passage for a given query.
+
+## **Module Parameters**
+
+**llm**: The query expansion node requires setting parameters related to our generator modules.
+
+- **generator_module_type**: The type of the generator module to use.
+- **llm**: The type of llm.
+- Other LLM-related parameters such as `model`, `temperature`, and `max_token` can be set. These are passed as keyword
+  arguments (`kwargs`) to the LLM object, allowing for further customization of the LLM's behavior.
+
+**Additional Parameters**:
+
+- **prompt**: You can use your own custom prompt for the LLM model.
+  Default prompt is come from the
+  paper "[Precise Zero-shot Dense Retrieval without Relevance Labels](https://arxiv.org/abs/2212.10496)".
+
+## **Example config.yaml**
+```yaml
+modules:
+  - module_type: hyde
+    generator_module_type: llama_index_llm
+    llm: openai
+    max_token: 64
+```

autorag-0.2.1/docs/source/nodes/query_expansion/multi_query_expansion.md

@@ -0,0 +1,30 @@
+# Multi Query Expansion
+
+The `multi_query_expansion` automates the process of prompt tuning by using an LLM to generate multiple queries from
+different perspectives for a given user input query. The module uses a default multi-query prompt from
+the [langchain MultiQueryRetriever](https://python.langchain.com/v0.1/docs/modules/data_connection/retrievers/MultiQueryRetriever/)'
+s default query prompt.
+
+## **Module Parameters**
+
+**llm**: The query expansion node requires setting parameters related to our generator modules.
+
+- **generator_module_type**: The type of the generator module to use.
+- **llm**: The type of llm.
+- Other LLM-related parameters such as `model`, `temperature`, and `max_token` can be set. These are passed as keyword
+  arguments (`kwargs`) to the LLM object, allowing for further customization of the LLM's behavior.
+
+**Additional Parameters**:
+
+- **prompt**: You can use your own custom prompt for the LLM model.
+  Default prompt comes from langchain MultiQueryRetriever default query prompt.
+
+## **Example config.yaml**
+
+```yaml
+modules:
+- module_type: multi_query_expansion
+  generator_module_type: llama_index_llm
+  llm: openai
+  model: [ gpt-3.5-turbo-16k, gpt-3.5-turbo-1106 ]
+```

autorag-0.2.1/docs/source/nodes/query_expansion/query_decompose.md

@@ -0,0 +1,26 @@
+# Query Decompose
+
+The `query_decompose` is used to decompose a ‘multi-hop question’ into ‘multiple single-hop questions’ using a LLM model. The module uses a default decomposition prompt from the [Visconde paper](https://arxiv.org/pdf/2212.09656.pdf)'s StrategyQA few-shot prompt.
+
+## **Module Parameters**
+
+**llm**: The query expansion node requires setting parameters related to our generator modules.
+
+- **generator_module_type**: The type of the generator module to use.
+- **llm**: The type of llm.
+- Other LLM-related parameters such as `model`, `temperature`, and `max_token` can be set. These are passed as keyword
+  arguments (`kwargs`) to the LLM object, allowing for further customization of the LLM's behavior.
+
+**Additional Parameters**:
+
+- **prompt**: You can use your own custom prompt for the LLM model.
+  default prompt comes from Visconde's StrategyQA few-shot prompt.
+
+## **Example config.yaml**
+```yaml
+modules:
+- module_type: query_decompose
+  generator_module_type: llama_index_llm
+  llm: openai
+  model: [ gpt-3.5-turbo-16k, gpt-3.5-turbo-1106 ]
+```

{autorag-0.2.0 → autorag-0.2.1}/docs/source/nodes/query_expansion/query_expansion.md

@@ -71,4 +71,5 @@ maxdepth: 1
 ---
 query_decompose.md
 hyde.md
+multi_query_expansion.md
 ```