flashrag-dev 0.1.4.dev20250720__tar.gz → 0.1.4.dev20250806__tar.gz

{flashrag_dev-0.1.4.dev20250720 → flashrag_dev-0.1.4.dev20250806}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: flashrag_dev
-Version: 0.1.4.dev20250720
+Version: 0.1.4.dev20250806
 Summary: A library for efficient Retrieval-Augmented Generation research
 Home-page: https://github.com/RUC-NLPIR/FlashRAG
 Author: Jiajie Jin, Yutao Zhu, Chenghao Zhang, Xinyu Yang, Zhicheng Dou
@@ -42,7 +42,7 @@ License-File: LICENSE
 </h4>
 
 
-FlashRAG is a Python toolkit for the reproduction and development of Retrieval Augmented Generation (RAG) research. Our toolkit includes 36 pre-processed benchmark RAG datasets and 17 state-of-the-art RAG algorithms.
+FlashRAG is a Python toolkit for the reproduction and development of Retrieval Augmented Generation (RAG) research. Our toolkit includes 36 pre-processed benchmark RAG datasets and **23 state-of-the-art RAG algorithms**, including **7 reasoning-based methods** that combine reasoning ability with retrieval.
 
 <p align="center">
 <img src="asset/framework.jpg">
@@ -76,7 +76,9 @@ https://github.com/user-attachments/assets/8ca00873-5df2-48a7-b853-89e7b18bc6e9
 
 - **Comprehensive Benchmark Datasets**: A collection of 36 pre-processed RAG benchmark datasets to test and validate RAG models' performances.
 
-- **Pre-implemented Advanced RAG Algorithms**: Features 17 advancing RAG algorithms with reported results, based on our framework. Easily reproducing results under different settings.
+- **Pre-implemented Advanced RAG Algorithms**: Features **23 advancing RAG algorithms** with reported results, based on our framework. Easily reproducing results under different settings.
+
+- **🚀 Reasoning-based Methods**: **NEW!** We now support **7 reasoning-based methods** that combine reasoning ability with retrieval, achieving superior performance on complex multi-hop tasks.
 
 - **Efficient Preprocessing Stage**: Simplifies the RAG workflow preparation by providing various scripts like corpus processing for retrieval, retrieval index building, and pre-retrieval of documents.
 
@@ -92,13 +94,15 @@ FlashRAG is still under development and there are many issues and room for impro
 - [x] Provdide instructions for each component
 - [x] Integrate sentence Transformers
 - [x] Support multimodal RAG
+- [x] Support reasoning-based methods
 - [ ] Inlcude more RAG approaches
-- [ ] Add more evaluation metrics (e.g., Unieval, name-entity F1) and benchmarks (e.g., RGB benchmark)
 - [ ] Enhance code adaptability and readability
 - [ ] Add support for api-based retriever (vllm server)
 
 ## :page_with_curl: Changelog
-[25/03/21] We have added support for **Reasoning Pipeline**, which is a new paradigm that combines reasoning ability and retrieval, representing work that includes [Search-o1](https://github.com/sunnynexus/Search-o1), [R1-Searcher](https://github.com/SsmallSong/R1-Searcher), [ReSearch](https://github.com/Agent-RL/ReSearch). We evaluate the performance of the pipeline on various RAG benchmarks, it can achieve F1 scores close to 60 on multi hop inference datasets such as Hotpotqe. See it in [**result table**](#robot-supporting-methods).
+[25/08/06] 🎯 **NEW!** We have added support for **Reasoning Pipeline**, which is a new paradigm that combines reasoning ability and retrieval, representing work that includes [R1-Searcher](https://github.com/SsmallSong/R1-Searcher), [Search-R1](https://github.com/PeterGriffinJin/Search-R1),.... We evaluate the performance of the pipeline on various RAG benchmarks, it can achieve F1 scores close to 60 on multi hop inference datasets such as HotpotQA. See it in [**result table**](#robot-supporting-methods).
+
+[25/03/21] 🚀 **Major Update!** We have expanded our toolkit to support **23 state-of-the-art RAG algorithms**, including **7 reasoning-based methods** that significantly improve performance on complex reasoning tasks. This represents a major milestone in our toolkit's evolution!
 
 [25/02/24] 🔥🔥🔥 We have added support for **multimodal RAG**, including [**MLLMs like Llava, Qwen, InternVL**](https://ruc-nlpir.github.io/FlashRAG/#/zh-cn/component/generator?id=%e5%a4%9a%e6%a8%a1%e6%80%81%e7%94%9f%e6%88%90%e5%99%a8), and various [**multimodal retrievers with Clip architecture**](https://ruc-nlpir.github.io/FlashRAG/#/zh-cn/component/retriever?id=%e5%a4%9a%e6%a8%a1%e6%80%81%e6%a3%80%e7%b4%a2%e5%99%a8). More information can be found in our new version of arxiv article and our documentation. Try it!
 
@@ -587,7 +591,7 @@ python interface.py
 
 ## :robot: Supporting Methods
 
-We have implemented 15 works with a consistent setting of:
+We have implemented **23 works** with a consistent setting of:
 
 - **Generator:** LLAMA3-8B-instruct with input length of 2048
 - **Retriever:** e5-base-v2 as embedding model, retrieve 5 docs per query
@@ -597,7 +601,7 @@ For open-source methods, we implemented their processes using our framework. For
 
 For necessary settings and hyperparameters specific to some methods, we have documented them in the **specific settings** column. For more details, please consult our [<u>reproduce guidance</u>](./docs/original_docs/reproduce_experiment.md) and [<u>method details</u>](./docs/original_docs/baseline_details.md).
 
-It’s important to note that, to ensure consistency, we have utilized a uniform setting. However, this setting may differ from the original setting of the method, leading to variations in results compared to the original outcomes.
+It's important to note that, to ensure consistency, we have utilized a uniform setting. However, this setting may differ from the original setting of the method, leading to variations in results compared to the original outcomes.
 
 | Method                                                                                    | Type        | NQ (EM) | TriviaQA (EM) | Hotpotqa (F1) | 2Wiki (F1) | PopQA (F1) | WebQA(EM) | Specific setting                                |
 | ----------------------------------------------------------------------------------------- | ----------- | ------- | ------------- | ------------- | ---------- | ---------- | --------- | ----------------------------------------------- |
@@ -618,10 +622,21 @@ It’s important to note that, to ensure consistency, we have utilized a uniform
 | [FLARE](https://arxiv.org/abs/2305.06983)                                                 | Loop        | 22.5    | 55.8          | 28.0          | 33.9       | 20.7       | 20.2      |                                                 |
 | [Iter-Retgen](https://arxiv.org/abs/2305.15294), [ITRG](https://arxiv.org/abs/2310.05149) | Loop        | 36.8    | 60.1          | 38.3          | 21.6       | 37.9       | 18.2      |                                                 |
 | [IRCoT](https://aclanthology.org/2023.acl-long.557.pdf)                                   | Loop        | 33.3    | 56.9          | 41.5          | 32.4       | 45.6       | 20.7      |                                                 |
-| [RQRAG](https://arxiv.org/abs/2404.00610)                                   | Loop        | 32.6    | 52.5          | 33.5          | 35.8       | 46.4       | 26.2      |  Use trained rqrag-llama2-7B                                               |
-| [R1-Searcher](https://arxiv.org/pdf/2503.05592) | Reasoning | 37.3 | 58.7 | 59.5 | 55.5 | 43.9 | 27.5 | Using trained Qwen2.5-7B model |
- 
+| [RQRAG](https://arxiv.org/abs/2404.00610)                                   | Loop        | 32.6    | 52.5          | 33.5          | 35.8       | 46.4       | 26.2      |  Use trained rqrag-llama2-7B                                               | 
+
+#### 🚀 Reasoning-based Methods (NEW!)
+
+We now support **7 reasoning-based methods** that combine reasoning ability with retrieval, achieving superior performance on complex multi-hop tasks:
 
+| Method                                                                                    | Type        | NQ (EM) | TriviaQA (EM) | PopQA (EM) | Hotpotqa (F1) | 2Wiki (F1) |  Musique (F1) | Bamboogle (F1) | Specific setting                             |
+| ----------------------------------------------------------------------------------------- | ----------- | ------- | ------- | ------------- | ------------- | ---------- | ---------- | --------- | ----------------------------------------------- |
+| [Search-R1](https://arxiv.org/abs/2503.09516) | Reasoning | 45.2 | 62.2 | 49.2 | 54.5 | 42.6 | 29.2 |  59.9 | SearchR1-nq_hotpotqa_train-qwen2.5-7b-em-ppo |
+| [R1-Searcher](https://arxiv.org/pdf/2503.05592) | Reasoning | 36.9 | 61.6 | 42.0 | 49.0 | 49.1 | 24.7 | 57.7 | Qwen-2.5-7B-base-RAG-RL |
+| [O2-Searcher](https://arxiv.org/pdf/2505.16582) | Reasoning | 41.4 | 51.4 | 46.8 | 43.4 | 48.6 | 19.0 | 47.6 | O2-Searcher-Qwen2.5-3B-GRPO |
+| [AutoRefine](https://www.arxiv.org/pdf/2505.11277) | Reasoning | 43.8 | 59.8 | 32.4 | 54.0 | 50.3 | 23.6 | 46.6 | AutoRefine-Qwen2.5-3B-Base |
+| [ReaRAG](https://arxiv.org/abs/2503.21729) | Reasoning | 26.3 | 51.8 | 24.6 | 42.9 | 41.6 | 21.2 | 41.9 | ReaRAG-9B |
+| [CoRAG](https://arxiv.org/abs/2503.21729) | Reasoning | 40.9 | 63.1 | 36.0 | 56.6 | 60.7 | 31.9 | 54.1 | CoRAG-Llama3.1-8B-MultihopQA |
+| [SimpleDeepSearcher](https://arxiv.org/pdf/2505.16834) | Reasoning | 36.1 | 61.6 | 42.0 | 49.0 | 49.1 | 24.7 | 57.7 | Qwen-7B-SimpleDeepSearcher |
 
 ## :notebook: Supporting Datasets & Document Corpus
 

{flashrag_dev-0.1.4.dev20250720 → flashrag_dev-0.1.4.dev20250806}/README.md

@@ -25,7 +25,7 @@
 </h4>
 
 
-FlashRAG is a Python toolkit for the reproduction and development of Retrieval Augmented Generation (RAG) research. Our toolkit includes 36 pre-processed benchmark RAG datasets and 17 state-of-the-art RAG algorithms.
+FlashRAG is a Python toolkit for the reproduction and development of Retrieval Augmented Generation (RAG) research. Our toolkit includes 36 pre-processed benchmark RAG datasets and **23 state-of-the-art RAG algorithms**, including **7 reasoning-based methods** that combine reasoning ability with retrieval.
 
 <p align="center">
 <img src="asset/framework.jpg">
@@ -59,7 +59,9 @@ https://github.com/user-attachments/assets/8ca00873-5df2-48a7-b853-89e7b18bc6e9
 
 - **Comprehensive Benchmark Datasets**: A collection of 36 pre-processed RAG benchmark datasets to test and validate RAG models' performances.
 
-- **Pre-implemented Advanced RAG Algorithms**: Features 17 advancing RAG algorithms with reported results, based on our framework. Easily reproducing results under different settings.
+- **Pre-implemented Advanced RAG Algorithms**: Features **23 advancing RAG algorithms** with reported results, based on our framework. Easily reproducing results under different settings.
+
+- **🚀 Reasoning-based Methods**: **NEW!** We now support **7 reasoning-based methods** that combine reasoning ability with retrieval, achieving superior performance on complex multi-hop tasks.
 
 - **Efficient Preprocessing Stage**: Simplifies the RAG workflow preparation by providing various scripts like corpus processing for retrieval, retrieval index building, and pre-retrieval of documents.
 
@@ -75,13 +77,15 @@ FlashRAG is still under development and there are many issues and room for impro
 - [x] Provdide instructions for each component
 - [x] Integrate sentence Transformers
 - [x] Support multimodal RAG
+- [x] Support reasoning-based methods
 - [ ] Inlcude more RAG approaches
-- [ ] Add more evaluation metrics (e.g., Unieval, name-entity F1) and benchmarks (e.g., RGB benchmark)
 - [ ] Enhance code adaptability and readability
 - [ ] Add support for api-based retriever (vllm server)
 
 ## :page_with_curl: Changelog
-[25/03/21] We have added support for **Reasoning Pipeline**, which is a new paradigm that combines reasoning ability and retrieval, representing work that includes [Search-o1](https://github.com/sunnynexus/Search-o1), [R1-Searcher](https://github.com/SsmallSong/R1-Searcher), [ReSearch](https://github.com/Agent-RL/ReSearch). We evaluate the performance of the pipeline on various RAG benchmarks, it can achieve F1 scores close to 60 on multi hop inference datasets such as Hotpotqe. See it in [**result table**](#robot-supporting-methods).
+[25/08/06] 🎯 **NEW!** We have added support for **Reasoning Pipeline**, which is a new paradigm that combines reasoning ability and retrieval, representing work that includes [R1-Searcher](https://github.com/SsmallSong/R1-Searcher), [Search-R1](https://github.com/PeterGriffinJin/Search-R1),.... We evaluate the performance of the pipeline on various RAG benchmarks, it can achieve F1 scores close to 60 on multi hop inference datasets such as HotpotQA. See it in [**result table**](#robot-supporting-methods).
+
+[25/03/21] 🚀 **Major Update!** We have expanded our toolkit to support **23 state-of-the-art RAG algorithms**, including **7 reasoning-based methods** that significantly improve performance on complex reasoning tasks. This represents a major milestone in our toolkit's evolution!
 
 [25/02/24] 🔥🔥🔥 We have added support for **multimodal RAG**, including [**MLLMs like Llava, Qwen, InternVL**](https://ruc-nlpir.github.io/FlashRAG/#/zh-cn/component/generator?id=%e5%a4%9a%e6%a8%a1%e6%80%81%e7%94%9f%e6%88%90%e5%99%a8), and various [**multimodal retrievers with Clip architecture**](https://ruc-nlpir.github.io/FlashRAG/#/zh-cn/component/retriever?id=%e5%a4%9a%e6%a8%a1%e6%80%81%e6%a3%80%e7%b4%a2%e5%99%a8). More information can be found in our new version of arxiv article and our documentation. Try it!
 
@@ -570,7 +574,7 @@ python interface.py
 
 ## :robot: Supporting Methods
 
-We have implemented 15 works with a consistent setting of:
+We have implemented **23 works** with a consistent setting of:
 
 - **Generator:** LLAMA3-8B-instruct with input length of 2048
 - **Retriever:** e5-base-v2 as embedding model, retrieve 5 docs per query
@@ -580,7 +584,7 @@ For open-source methods, we implemented their processes using our framework. For
 
 For necessary settings and hyperparameters specific to some methods, we have documented them in the **specific settings** column. For more details, please consult our [<u>reproduce guidance</u>](./docs/original_docs/reproduce_experiment.md) and [<u>method details</u>](./docs/original_docs/baseline_details.md).
 
-It’s important to note that, to ensure consistency, we have utilized a uniform setting. However, this setting may differ from the original setting of the method, leading to variations in results compared to the original outcomes.
+It's important to note that, to ensure consistency, we have utilized a uniform setting. However, this setting may differ from the original setting of the method, leading to variations in results compared to the original outcomes.
 
 | Method                                                                                    | Type        | NQ (EM) | TriviaQA (EM) | Hotpotqa (F1) | 2Wiki (F1) | PopQA (F1) | WebQA(EM) | Specific setting                                |
 | ----------------------------------------------------------------------------------------- | ----------- | ------- | ------------- | ------------- | ---------- | ---------- | --------- | ----------------------------------------------- |
@@ -601,10 +605,21 @@ It’s important to note that, to ensure consistency, we have utilized a uniform
 | [FLARE](https://arxiv.org/abs/2305.06983)                                                 | Loop        | 22.5    | 55.8          | 28.0          | 33.9       | 20.7       | 20.2      |                                                 |
 | [Iter-Retgen](https://arxiv.org/abs/2305.15294), [ITRG](https://arxiv.org/abs/2310.05149) | Loop        | 36.8    | 60.1          | 38.3          | 21.6       | 37.9       | 18.2      |                                                 |
 | [IRCoT](https://aclanthology.org/2023.acl-long.557.pdf)                                   | Loop        | 33.3    | 56.9          | 41.5          | 32.4       | 45.6       | 20.7      |                                                 |
-| [RQRAG](https://arxiv.org/abs/2404.00610)                                   | Loop        | 32.6    | 52.5          | 33.5          | 35.8       | 46.4       | 26.2      |  Use trained rqrag-llama2-7B                                               |
-| [R1-Searcher](https://arxiv.org/pdf/2503.05592) | Reasoning | 37.3 | 58.7 | 59.5 | 55.5 | 43.9 | 27.5 | Using trained Qwen2.5-7B model |
- 
+| [RQRAG](https://arxiv.org/abs/2404.00610)                                   | Loop        | 32.6    | 52.5          | 33.5          | 35.8       | 46.4       | 26.2      |  Use trained rqrag-llama2-7B                                               | 
+
+#### 🚀 Reasoning-based Methods (NEW!)
+
+We now support **7 reasoning-based methods** that combine reasoning ability with retrieval, achieving superior performance on complex multi-hop tasks:
 
+| Method                                                                                    | Type        | NQ (EM) | TriviaQA (EM) | PopQA (EM) | Hotpotqa (F1) | 2Wiki (F1) |  Musique (F1) | Bamboogle (F1) | Specific setting                             |
+| ----------------------------------------------------------------------------------------- | ----------- | ------- | ------- | ------------- | ------------- | ---------- | ---------- | --------- | ----------------------------------------------- |
+| [Search-R1](https://arxiv.org/abs/2503.09516) | Reasoning | 45.2 | 62.2 | 49.2 | 54.5 | 42.6 | 29.2 |  59.9 | SearchR1-nq_hotpotqa_train-qwen2.5-7b-em-ppo |
+| [R1-Searcher](https://arxiv.org/pdf/2503.05592) | Reasoning | 36.9 | 61.6 | 42.0 | 49.0 | 49.1 | 24.7 | 57.7 | Qwen-2.5-7B-base-RAG-RL |
+| [O2-Searcher](https://arxiv.org/pdf/2505.16582) | Reasoning | 41.4 | 51.4 | 46.8 | 43.4 | 48.6 | 19.0 | 47.6 | O2-Searcher-Qwen2.5-3B-GRPO |
+| [AutoRefine](https://www.arxiv.org/pdf/2505.11277) | Reasoning | 43.8 | 59.8 | 32.4 | 54.0 | 50.3 | 23.6 | 46.6 | AutoRefine-Qwen2.5-3B-Base |
+| [ReaRAG](https://arxiv.org/abs/2503.21729) | Reasoning | 26.3 | 51.8 | 24.6 | 42.9 | 41.6 | 21.2 | 41.9 | ReaRAG-9B |
+| [CoRAG](https://arxiv.org/abs/2503.21729) | Reasoning | 40.9 | 63.1 | 36.0 | 56.6 | 60.7 | 31.9 | 54.1 | CoRAG-Llama3.1-8B-MultihopQA |
+| [SimpleDeepSearcher](https://arxiv.org/pdf/2505.16834) | Reasoning | 36.1 | 61.6 | 42.0 | 49.0 | 49.1 | 24.7 | 57.7 | Qwen-7B-SimpleDeepSearcher |
 
 ## :notebook: Supporting Datasets & Document Corpus
 

{flashrag_dev-0.1.4.dev20250720 → flashrag_dev-0.1.4.dev20250806}/flashrag/generator/generator.py

@@ -12,6 +12,7 @@ from transformers import (
     AutoConfig,
 )
 from flashrag.generator.utils import resolve_max_tokens
+from flashrag.utils import get_device
 
 
 class BaseGenerator:
@@ -553,13 +554,13 @@ class FastChatGenerator(HFCausalLMGenerator):
             max_gpu_memory = None
             import torch
             self.gpu_num = torch.cuda.device_count()
-            if self.gpu_num != 1:
+            if self.gpu_num > 1:
                 available_gpu_memory = get_gpu_memory()
                 max_gpu_memory = str(int(min(available_gpu_memory) * gpu_memory_utilization)) + "GiB"
 
             model, tokenizer = load_model(
                 self.model_path,
-                device="cuda",
+                device=get_device(),
                 num_gpus=self.gpu_num,
                 max_gpu_memory=max_gpu_memory,
                 load_8bit=False,

flashrag_dev-0.1.4.dev20250806/flashrag/pipeline/ReaRAG_utils.py

@@ -0,0 +1,150 @@
+import re
+from termcolor import colored
+
+class AgentUtilsBase():
+    def __init__(self):
+        pass
+
+    def truncate(self, tokenizer, tokenized_prompt, model_max_length):
+        half = int(model_max_length/2)
+        prompt = tokenizer.decode(tokenized_prompt[:half], skip_special_tokens=False) + \
+                tokenizer.decode(tokenized_prompt[-half:], skip_special_tokens=False)
+        
+        return prompt
+
+    def preprocess_query(self, query):
+        if "'" in query and '"' in query:
+            query = query.replace("'", "\\'").replace('"', '\\"')
+        return query
+
+    def extract_code(self, text: str) -> str:
+        triple_match = re.search(r'```[^\n]*\n(.+?)```', text, re.DOTALL)
+        single_match = re.search(r'`([^`]*)`', text, re.DOTALL)
+        if triple_match:
+            return triple_match.group(1)
+        elif single_match:
+            return single_match.group(1)
+        return text
+    
+    def postprocess_agent_response(self, response):
+        """
+        Implement the postprocess_agent_response function
+        """
+        raise NotImplementedError("postprocess_agent_response function must be implemented in a subclass")
+
+class AgentUtils(AgentUtilsBase):
+    def __init__(self):
+        super().__init__()
+
+    def parse_reasoning_steps(self, text: str):
+        """
+        Parse a string containing Thought/Action/Observation steps (including multi-line)
+        and return a list of dictionaries of the form:
+        
+        [
+            {
+                "1": {
+                    "Thought": "...",
+                    "Action": "...",  # Only content inside backticks (if present)
+                    "Observation": "..."
+                }
+            },
+            {
+                "2": {
+                    "Thought": "...",
+                    "Action": "...",
+                    "Observation": "..."
+                }
+            },
+            ...
+        ]
+        """
+        # Regex pattern to match lines that start with "Thought X:", "Action X:", or "Observation X:".
+        pattern = re.compile(r'^(Thought|Action|Observation)\s+(\d+):', re.MULTILINE)
+
+        # This dictionary will accumulate:
+        # data_dict[step_number] = {"Thought": ..., "Action": ..., "Observation": ...}
+        data_dict = {}
+
+        # We'll track the current label (Thought/Action/Observation) and step number
+        current_label = None
+        current_step = None
+        last_pos = 0
+
+        # Find all pattern occurrences in the text
+        matches = list(pattern.finditer(text))
+
+        for match in matches:
+            # If we already have a label in progress, we can record its content
+            if current_label is not None:
+                # Slice the text from the last match's end to the start of this new match
+                content = text[last_pos:match.start()].strip()
+                # Store that content in data_dict
+                data_dict[current_step][current_label] = content
+
+            # Extract the new label and step
+            label = match.group(1)       # "Thought", "Action", or "Observation"
+            step = match.group(2)        # e.g. "1", "2", "3"
+
+            # Ensure a dict for this step
+            if step not in data_dict:
+                data_dict[step] = {"Thought": None, "Action": None, "Observation": None}
+
+            # Update current label/step
+            current_label = label
+            current_step = step
+            # We'll slice from here next time
+            last_pos = match.end()
+
+        # Handle the final block after the last match
+        if current_label is not None:
+            content = text[last_pos:].strip()
+            data_dict[current_step][current_label] = content
+
+        # Post-process:
+        #  - For each step, extract only the text inside triple backticks for "Action".
+        for step_number in data_dict:
+            action_text = data_dict[step_number]["Action"]
+            if action_text:
+                # Extract content inside triple backticks
+                data_dict[step_number]["Action"] = self.extract_code(action_text)
+
+        # Convert our dictionary to the desired list-of-dicts structure
+        structured_data = []
+        for step_number in sorted(data_dict.keys(), key=lambda x: int(x)):
+            structured_data.append({step_number: data_dict[step_number]})
+
+        return structured_data
+    
+    def postprocess_agent_response(self, response):
+        """
+        Extract Thought and Action, then extract the dict from Action.
+
+        Return:
+        - thought: str
+        - action: dict
+        - is_valid: bool
+        """
+        parsed_codes = self.parse_reasoning_steps(response)
+
+        thoughts = []
+        actions = []
+        for steps in parsed_codes:
+            for step_idx, step in steps.items():
+                thought = f"Thought {step_idx}: {step['Thought']}"
+                action = eval(self.extract_code(step['Action']))
+
+                assert "function" in action, f"Action does not contain 'function' key: {action}"
+                assert "parameters" in action, f"Action does not contain 'parameters' key: {action}"
+
+                thoughts.append(thought)
+                actions.append(action)
+
+        return thoughts, actions
+    
+def print_code(codes):
+    for idx, step in enumerate(codes):
+        print(f"{colored(step['thought'], 'blue')}")
+        print(colored(f"Action {idx+1}:\n```\n{step['action']}\n```", 'red'))
+        print(colored(f"Observation {idx+1}: {step['observation']}", 'yellow'))
+        # print(f"{'-'*60}")

{flashrag_dev-0.1.4.dev20250720 → flashrag_dev-0.1.4.dev20250806}/flashrag/pipeline/__init__.py

@@ -2,4 +2,5 @@ from flashrag.pipeline.mm_pipeline import *
 from flashrag.pipeline.pipeline import *
 from flashrag.pipeline.branching_pipeline import REPLUGPipeline, SuRePipeline
 from flashrag.pipeline.active_pipeline import IterativePipeline, SelfRAGPipeline, FLAREPipeline, SelfAskPipeline, IRCOTPipeline, RQRAGPipeline
-from flashrag.pipeline.reasoning_pipeline import ReasoningPipeline
+from flashrag.pipeline.reasoning_pipeline import *
+from flashrag.pipeline.ReaRAG_utils import *