test_commit

.ipynb_checkpoints/app-checkpoint.py

@@ -1,43 +1,95 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+from huggingface_hub import HfApiModel
+import sys
+if './lib' not in sys.path :
+    sys.path.append('./lib')
+from ingestion_chroma import retrieve_info_from_db
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+############################################################################################
+################################### TOOLS ##################################################
+############################################################################################
+
+def find_key(data, target_key):
+    if isinstance(data, dict):
+        for key, value in data.items():
+            if key == target_key:
+                return value
+            else:
+                result = find_key(value, target_key)
+                if result is not None:
+                    return result
+    return "Indicator not found"
+
+############################################################################################
+
+class Chroma_retrieverTool(Tool):
+    name = "request"
+    description = "Using semantic similarity, retrieve the text from the knowledge base that has the embedding closest to the query."
+    inputs = {
+        "query": {
+            "type": "string",
+            "description": "The query to execute must be semantically close to the text to search. Use the affirmative form rather than a question.",
+        },
+    }
+    output_type = "string"
+    
+    def forward(self, query: str) -> str:
+        assert isinstance(query, str), "The request needs to be a string."
+        
+        query_results = retrieve_info_from_db(query)
+        str_result = "\nRetrieval texts : \n" + "".join([f"===== Text {str(i)} =====\n" + query_results['documents'][0][i] for i in range(len(query_results['documents'][0]))])
+
+        return str_result
+        
+############################################################################################
 
+class ESRS_info_tool(Tool):
+    name = "find_ESRS"
+    description = "Find ESRS description to help you to find what indicators the user want"
+    inputs = {
+        "indicator": {
+            "type": "string",
+            "description": "The indicator name. return the description of the indicator demanded.",
+        },
+    }
+    output_type = "string"
+    
+    def forward(self, indicator: str) -> str:
+        assert isinstance(indicator, str), "The request needs to be a string."
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+        with open('./data/dico_esrs.json') as json_data:
+            dico_esrs = json.load(json_data)
+        
+        result = find_key(dico_esrs, indicator)
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+        return result
+        
+############################################################################################
+############################################################################################
+############################################################################################
 
-    messages.append({"role": "user", "content": message})
+model = HfApiModel("Qwen/Qwen2.5-Coder-32B-Instruct")
 
-    response = ""
+retriever_tool = Chroma_retrieverTool()
+get_ESRS_info_tool = ESRS_info_tool()
+agent = CodeAgent(
+    tools=[
+            get_ESRS_info_tool,
+            retriever_tool,
+          ],
+    model=model,
+    max_steps=10,
+    max_print_outputs_length=16000,
+    additional_authorized_imports=['pandas', 'matplotlib', 'datetime']
+    )
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
 
-        response += token
-        yield response
+def respond(message):
+    system_prompt_added = """You are an expert in environmental and corporate social responsibility. You must respond to requests using the query function in the document database. 
+User's question : """
+    agent_output = agent.run(system_prompt_added+"""Find all informations about the ESRS E1–5: Energy consumption from fossil sources in Sartorius documents.""")
+    
+    yield agent_output
 
 
 """
@@ -45,18 +97,6 @@ For information on how to customize the ChatInterface, peruse the gradio docs: h
 """
 demo = gr.ChatInterface(
     respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
 )
 
 

.ipynb_checkpoints/archive-checkpoint.py

@@ -0,0 +1,64 @@
+import gradio as gr
+from huggingface_hub import InferenceClient
+
+"""
+For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
+"""
+client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+
+
+def respond(
+    message,
+    history: list[tuple[str, str]],
+    system_message,
+    max_tokens,
+    temperature,
+    top_p,
+):
+    messages = [{"role": "system", "content": system_message}]
+
+    for val in history:
+        if val[0]:
+            messages.append({"role": "user", "content": val[0]})
+        if val[1]:
+            messages.append({"role": "assistant", "content": val[1]})
+
+    messages.append({"role": "user", "content": message})
+
+    response = ""
+
+    for message in client.chat_completion(
+        messages,
+        max_tokens=max_tokens,
+        stream=True,
+        temperature=temperature,
+        top_p=top_p,
+    ):
+        token = message.choices[0].delta.content
+
+        response += token
+        yield response
+
+
+"""
+For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
+"""
+demo = gr.ChatInterface(
+    respond,
+    additional_inputs=[
+        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
+        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
+        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
+        gr.Slider(
+            minimum=0.1,
+            maximum=1.0,
+            value=0.95,
+            step=0.05,
+            label="Top-p (nucleus sampling)",
+        ),
+    ],
+)
+
+
+if __name__ == "__main__":
+    demo.launch()

.ipynb_checkpoints/requirements-checkpoint.txt

@@ -0,0 +1,2 @@
+huggingface_hub==0.25.2
+chromadb==0.6.3

.ipynb_checkpoints/test-checkpoint.txt

@@ -1 +0,0 @@
-blablabla

app.py

@@ -1,43 +1,95 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+from huggingface_hub import HfApiModel
+import sys
+if './lib' not in sys.path :
+    sys.path.append('./lib')
+from ingestion_chroma import retrieve_info_from_db
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+############################################################################################
+################################### TOOLS ##################################################
+############################################################################################
+
+def find_key(data, target_key):
+    if isinstance(data, dict):
+        for key, value in data.items():
+            if key == target_key:
+                return value
+            else:
+                result = find_key(value, target_key)
+                if result is not None:
+                    return result
+    return "Indicator not found"
+
+############################################################################################
+
+class Chroma_retrieverTool(Tool):
+    name = "request"
+    description = "Using semantic similarity, retrieve the text from the knowledge base that has the embedding closest to the query."
+    inputs = {
+        "query": {
+            "type": "string",
+            "description": "The query to execute must be semantically close to the text to search. Use the affirmative form rather than a question.",
+        },
+    }
+    output_type = "string"
+    
+    def forward(self, query: str) -> str:
+        assert isinstance(query, str), "The request needs to be a string."
+        
+        query_results = retrieve_info_from_db(query)
+        str_result = "\nRetrieval texts : \n" + "".join([f"===== Text {str(i)} =====\n" + query_results['documents'][0][i] for i in range(len(query_results['documents'][0]))])
+
+        return str_result
+        
+############################################################################################
 
+class ESRS_info_tool(Tool):
+    name = "find_ESRS"
+    description = "Find ESRS description to help you to find what indicators the user want"
+    inputs = {
+        "indicator": {
+            "type": "string",
+            "description": "The indicator name. return the description of the indicator demanded.",
+        },
+    }
+    output_type = "string"
+    
+    def forward(self, indicator: str) -> str:
+        assert isinstance(indicator, str), "The request needs to be a string."
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
-    messages = [{"role": "system", "content": system_message}]
+        with open('./data/dico_esrs.json') as json_data:
+            dico_esrs = json.load(json_data)
+        
+        result = find_key(dico_esrs, indicator)
 
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+        return result
+        
+############################################################################################
+############################################################################################
+############################################################################################
 
-    messages.append({"role": "user", "content": message})
+model = HfApiModel("Qwen/Qwen2.5-Coder-32B-Instruct")
 
-    response = ""
+retriever_tool = Chroma_retrieverTool()
+get_ESRS_info_tool = ESRS_info_tool()
+agent = CodeAgent(
+    tools=[
+            get_ESRS_info_tool,
+            retriever_tool,
+          ],
+    model=model,
+    max_steps=10,
+    max_print_outputs_length=16000,
+    additional_authorized_imports=['pandas', 'matplotlib', 'datetime']
+    )
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
 
-        response += token
-        yield response
+def respond(message):
+    system_prompt_added = """You are an expert in environmental and corporate social responsibility. You must respond to requests using the query function in the document database. 
+User's question : """
+    agent_output = agent.run(system_prompt_added+"""Find all informations about the ESRS E1–5: Energy consumption from fossil sources in Sartorius documents.""")
+    
+    yield agent_output
 
 
 """
@@ -45,18 +97,6 @@ For information on how to customize the ChatInterface, peruse the gradio docs: h
 """
 demo = gr.ChatInterface(
     respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
 )
 
 

archive.py

@@ -0,0 +1,64 @@
+import gradio as gr
+from huggingface_hub import InferenceClient
+
+"""
+For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
+"""
+client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+
+
+def respond(
+    message,
+    history: list[tuple[str, str]],
+    system_message,
+    max_tokens,
+    temperature,
+    top_p,
+):
+    messages = [{"role": "system", "content": system_message}]
+
+    for val in history:
+        if val[0]:
+            messages.append({"role": "user", "content": val[0]})
+        if val[1]:
+            messages.append({"role": "assistant", "content": val[1]})
+
+    messages.append({"role": "user", "content": message})
+
+    response = ""
+
+    for message in client.chat_completion(
+        messages,
+        max_tokens=max_tokens,
+        stream=True,
+        temperature=temperature,
+        top_p=top_p,
+    ):
+        token = message.choices[0].delta.content
+
+        response += token
+        yield response
+
+
+"""
+For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
+"""
+demo = gr.ChatInterface(
+    respond,
+    additional_inputs=[
+        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
+        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
+        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
+        gr.Slider(
+            minimum=0.1,
+            maximum=1.0,
+            value=0.95,
+            step=0.05,
+            label="Top-p (nucleus sampling)",
+        ),
+    ],
+)
+
+
+if __name__ == "__main__":
+    demo.launch()

lib/.ipynb_checkpoints/ingestion_chroma-checkpoint.py

@@ -0,0 +1,121 @@
+import chromadb
+from chromadb.utils import embedding_functions
+from tqdm import tqdm
+import time
+
+
+####################################################################################################################################
+############################################# GLOBAL INGESTION #####################################################################
+####################################################################################################################################
+def prepare_chunks_for_ingestion(df):
+    """
+    Specialisé pour les fichiers RSE
+    """
+    chunks = list(df.full_chunk)
+    metadatas = [
+        {
+         "source": str(source),
+         "chunk_size": str(chunk_size),
+        }
+        for source, chunk_size in zip(list(df.source), list(df.chunk_size))
+    ]
+    return chunks, metadatas
+
+
+###################################################################################################################################
+def ingest_chunks(df=None, batch_size=100, create_collection=False, chroma_data_path="./chroma_data/", embedding_model="intfloat/multilingual-e5-large", collection_name=None):
+    """
+    Adds to a RAG database from a dataframe with metadata and text already read. And returns the question answering pipeline.
+    Documents already chunked ! 
+    Custom file slicing from self-care data.
+    Parameters:
+    - df the dataframe of chunked docs with their metadata and text
+    - batch_size (optional)
+    Returns:
+    - collection: the resulting chroma collection
+    - duration: the list of duration of batch ingestion
+    """
+    
+    print("Modèle d'embedding choisi: ", embedding_model)
+    print("Collection où ingérer: ", collection_name)
+    # La collection du vector store est censée déjà exister.
+    client = chromadb.PersistentClient(path=chroma_data_path)
+    embedding_func = embedding_functions.SentenceTransformerEmbeddingFunction(
+                 model_name=embedding_model)
+    
+    if create_collection:
+        collection = client.create_collection(
+            name=collection_name,
+            embedding_function=embedding_func,
+            metadata={"hnsw:space": "cosine"},
+        )
+        next_id = 0
+    else:
+        collection = client.get_collection(name=collection_name, embedding_function=embedding_func)
+        print("Computing next chroma id. Please wait a few minutes...")
+        next_id = compute_next_id_chroma(chroma_data_path, collection_name)
+    print("Préparation des métadatas des chunks :")
+    documents, metadatas = prepare_chunks_for_ingestion(df)
+    # batch adding to do it faster
+    durations = []
+    total_batches = len(documents)/batch_size
+    initialisation=True
+    for i in tqdm(range(0, len(documents), batch_size)):
+        # print(f"Processing batch number {i/batch_size} of {total_batches}...")
+        if initialisation:
+            print(f"Processing first batch of {total_batches}.")
+            print("This can take 10-15 mins if this is the first time the model is loaded. Please wait...")
+            initialisation=False
+        with open("ingesting.log", "a") as file:
+            file.write(f"Processing batch number {i/batch_size} of {total_batches}..." +"\n")
+        batch_documents = documents[i:i+batch_size]
+        batch_ids = [f"id{j}" for j in range(next_id+i, next_id+i+len(batch_documents))]
+        batch_metadatas = metadatas[i:i+batch_size]
+        start_time = time.time() # start measuring execution time
+        collection.add(
+            documents=batch_documents,
+            ids=batch_ids, # [f"id{i}" for i in range(len(documents))],
+            metadatas=batch_metadatas
+        )
+        end_time = time.time() # end measuring execution time
+        with open("ingesting.log", "a") as file:
+            file.write(f"Done. Collection adding time: {end_time-start_time}"+"\n")
+        durations.append(end_time-start_time) # store execution times per batch
+    return collection, durations
+
+
+###################################################################################################################################
+def clean_rag_collection(collname,chroma_data_path):
+    """ Removes the old ollection for the RAG to ingest data new.
+    """
+    client = chromadb.PersistentClient(path=chroma_data_path)
+    res = client.delete_collection(name=collname)
+    return res
+
+
+###################################################################################################################################
+def retrieve_info_from_db(prompt: str, entreprise=None):
+    EMBED_MODEL = 'intfloat/multilingual-e5-large'
+    collection_name = "RSE_CSRD_REPORTS_TEST"
+    # création du client
+    client = chromadb.PersistentClient(path="./data/chroma_data/")
+    # chargement du modèle d'embedding permettant le calcul de proximité sémantique
+    embedding_func = embedding_functions.SentenceTransformerEmbeddingFunction(
+        model_name=EMBED_MODEL
+    )
+    collection = client.get_collection(name=collection_name, embedding_function=embedding_func)
+    if entreprise is not None:
+        # requête
+        query_results = collection.query(
+                query_texts=[prompt],
+                n_results=3,
+                where={'source': entreprise}
+            )
+    else:
+        # requête
+        query_results = collection.query(
+                query_texts=[prompt],
+                n_results=3
+            )
+        
+    return query_results

lib/ingestion_chroma.py

@@ -0,0 +1,121 @@
+import chromadb
+from chromadb.utils import embedding_functions
+from tqdm import tqdm
+import time
+
+
+####################################################################################################################################
+############################################# GLOBAL INGESTION #####################################################################
+####################################################################################################################################
+def prepare_chunks_for_ingestion(df):
+    """
+    Specialisé pour les fichiers RSE
+    """
+    chunks = list(df.full_chunk)
+    metadatas = [
+        {
+         "source": str(source),
+         "chunk_size": str(chunk_size),
+        }
+        for source, chunk_size in zip(list(df.source), list(df.chunk_size))
+    ]
+    return chunks, metadatas
+
+
+###################################################################################################################################
+def ingest_chunks(df=None, batch_size=100, create_collection=False, chroma_data_path="./chroma_data/", embedding_model="intfloat/multilingual-e5-large", collection_name=None):
+    """
+    Adds to a RAG database from a dataframe with metadata and text already read. And returns the question answering pipeline.
+    Documents already chunked ! 
+    Custom file slicing from self-care data.
+    Parameters:
+    - df the dataframe of chunked docs with their metadata and text
+    - batch_size (optional)
+    Returns:
+    - collection: the resulting chroma collection
+    - duration: the list of duration of batch ingestion
+    """
+    
+    print("Modèle d'embedding choisi: ", embedding_model)
+    print("Collection où ingérer: ", collection_name)
+    # La collection du vector store est censée déjà exister.
+    client = chromadb.PersistentClient(path=chroma_data_path)
+    embedding_func = embedding_functions.SentenceTransformerEmbeddingFunction(
+                 model_name=embedding_model)
+    
+    if create_collection:
+        collection = client.create_collection(
+            name=collection_name,
+            embedding_function=embedding_func,
+            metadata={"hnsw:space": "cosine"},
+        )
+        next_id = 0
+    else:
+        collection = client.get_collection(name=collection_name, embedding_function=embedding_func)
+        print("Computing next chroma id. Please wait a few minutes...")
+        next_id = compute_next_id_chroma(chroma_data_path, collection_name)
+    print("Préparation des métadatas des chunks :")
+    documents, metadatas = prepare_chunks_for_ingestion(df)
+    # batch adding to do it faster
+    durations = []
+    total_batches = len(documents)/batch_size
+    initialisation=True
+    for i in tqdm(range(0, len(documents), batch_size)):
+        # print(f"Processing batch number {i/batch_size} of {total_batches}...")
+        if initialisation:
+            print(f"Processing first batch of {total_batches}.")
+            print("This can take 10-15 mins if this is the first time the model is loaded. Please wait...")
+            initialisation=False
+        with open("ingesting.log", "a") as file:
+            file.write(f"Processing batch number {i/batch_size} of {total_batches}..." +"\n")
+        batch_documents = documents[i:i+batch_size]
+        batch_ids = [f"id{j}" for j in range(next_id+i, next_id+i+len(batch_documents))]
+        batch_metadatas = metadatas[i:i+batch_size]
+        start_time = time.time() # start measuring execution time
+        collection.add(
+            documents=batch_documents,
+            ids=batch_ids, # [f"id{i}" for i in range(len(documents))],
+            metadatas=batch_metadatas
+        )
+        end_time = time.time() # end measuring execution time
+        with open("ingesting.log", "a") as file:
+            file.write(f"Done. Collection adding time: {end_time-start_time}"+"\n")
+        durations.append(end_time-start_time) # store execution times per batch
+    return collection, durations
+
+
+###################################################################################################################################
+def clean_rag_collection(collname,chroma_data_path):
+    """ Removes the old ollection for the RAG to ingest data new.
+    """
+    client = chromadb.PersistentClient(path=chroma_data_path)
+    res = client.delete_collection(name=collname)
+    return res
+
+
+###################################################################################################################################
+def retrieve_info_from_db(prompt: str, entreprise=None):
+    EMBED_MODEL = 'intfloat/multilingual-e5-large'
+    collection_name = "RSE_CSRD_REPORTS_TEST"
+    # création du client
+    client = chromadb.PersistentClient(path="./data/chroma_data/")
+    # chargement du modèle d'embedding permettant le calcul de proximité sémantique
+    embedding_func = embedding_functions.SentenceTransformerEmbeddingFunction(
+        model_name=EMBED_MODEL
+    )
+    collection = client.get_collection(name=collection_name, embedding_function=embedding_func)
+    if entreprise is not None:
+        # requête
+        query_results = collection.query(
+                query_texts=[prompt],
+                n_results=3,
+                where={'source': entreprise}
+            )
+    else:
+        # requête
+        query_results = collection.query(
+                query_texts=[prompt],
+                n_results=3
+            )
+        
+    return query_results

requirements.txt

@@ -1 +1,2 @@
-huggingface_hub==0.25.2
+huggingface_hub==0.25.2
+chromadb==0.6.3

test.txt

@@ -1 +0,0 @@
-blablabla