Final touches

Diegi97 · Diegi97 · commit fc20d9c9de91 · 2024-08-30T13:42:36.000+02:00
diff --git a/Dogefile b/Dogefile
@@ -117,7 +117,7 @@ workers:
       tag: "chatf-1315"
     envs: [ django, ray_worker ]
     instance_size: XL
-    command: /.venv/bin/ray start --address=back:6375 --resources='{"tasks":100,"rags":100}' --block
+    command: /.venv/bin/ray start --address=back:6375 --resources='{"tasks":100,"ai_components":100}' --block
 
 jobs:
   - name: migrate
diff --git a/back/back/apps/broker/admin.py b/back/back/apps/broker/admin.py
@@ -2,17 +2,20 @@
 
 from .models.message import Message, UserFeedback, AdminReview
 from .models import ConsumerRoundRobinQueue, RemoteSDKParsers
+import time
 
 
 class MessageAdmin(admin.ModelAdmin):
     list_display = ["conversation_id", "id", "sender_type", "payload_text", "created_date"]
 
     def payload_text(self, obj):
         payload = obj.stack[0]['payload']
-        if isinstance(payload, str):
-            return payload
-        else:
+        if 'content' in payload and isinstance(payload, dict):
             return payload['content']
+        elif 'model_response' in payload and isinstance(payload, dict):
+            return payload['model_response']
+        else:
+            return payload
 
     def sender_type(self, obj):
         return obj.sender["type"]
diff --git a/back/back/apps/broker/models/message.py b/back/back/apps/broker/models/message.py
@@ -125,7 +125,7 @@ def get_formatted_conversation(self, chain):
                     messages.append({"role": "assistant", "content": bot_content})
                     bot_content = ""
 
-                messages.append({"role": "user", "content": m.stack[0]["payload"]})
+                messages.append({"role": "user", "content": m.stack[0]["payload"]["content"]})
                 human_messages_ids.append(m.id)
             elif m.sender["type"] == "bot":
                 bot_content += m.stack[0]["payload"]["content"]
diff --git a/back/back/apps/language_model/ray_deployments/colbert_deployment.py b/back/back/apps/language_model/ray_deployments/colbert_deployment.py
@@ -14,7 +14,7 @@
     ray_actor_options={
         "num_cpus": 1,
         "resources": {
-            "rags": 1,
+            "ai_components": 1,
         },
     },
 )
diff --git a/back/back/apps/language_model/ray_deployments/e5_deployment.py b/back/back/apps/language_model/ray_deployments/e5_deployment.py
@@ -13,7 +13,7 @@
     ray_actor_options={
             "num_cpus": 1,
             "resources": {
-                "rags": 1,
+                "ai_components": 1,
             }
         }
 )
diff --git a/back/back/apps/language_model/ray_deployments/llm_deployment.py b/back/back/apps/language_model/ray_deployments/llm_deployment.py
@@ -10,7 +10,7 @@
     ray_actor_options={
         "num_cpus": 0.01,
         "resources": {
-            "rags": 1,
+            "ai_components": 1,
         },
     },
 )
diff --git a/back/back/apps/language_model/tasks/intent_tasks.py b/back/back/apps/language_model/tasks/intent_tasks.py
@@ -393,8 +393,8 @@ def generate_suggested_intents_task(
         Message.objects.filter(
             stack__contains=[
                 {
-                    "payload": {"references": {"knowledge_base_id": "1"}},
-                    "type": "rag_generated_text",
+                    "payload": {"references": {"knowledge_base_id": knowledge_base_pk}},
+                    # "type": "rag_generated_text",
                 }
             ]
         )
@@ -410,7 +410,7 @@ def generate_suggested_intents_task(
     messages = Message.objects.filter(id__in=Subquery(subquery), sender__type="human")
 
     messages_text = [
-        Message.objects.get(id=item["message_id"]).stack[0]["payload"]
+        Message.objects.get(id=item["message_id"]).stack[0]["payload"]["content"]
         for item in messages
     ]
 
@@ -428,7 +428,7 @@ def generate_suggested_intents_task(
         < new_intents_thresholds["max"]
     ]
 
-    messages_text = [message.stack[0]["payload"] for message in messages]
+    messages_text = [message.stack[0]["payload"]["content"] for message in messages]
 
     logger.info(f"Number of messages after filtering: {messages.count()}")
 
diff --git a/back/run_ray_worker.sh b/back/run_ray_worker.sh
@@ -1,3 +1,3 @@
 export $(xargs <.env)
 
-ray start --address=back:6375 --resources='{"tasks": 100, "rags": 100}' --block
+ray start --address=back:6375 --resources='{"tasks": 100, "ai_components": 100}' --block
diff --git a/doc/source/installation.md b/doc/source/installation.md
@@ -17,7 +17,7 @@ The system comprises seven main components, here are their relationships and the
 - **Widget** <!-- (<a href="/en/latest/modules/installations/index.html#widget-installation">local install</a>) --> is a JS browser client application from which the user interacts with the bot.
 
 
-- **Admin** is a JS browser client application to manage the chatbots, datasets, retriever, models, RAG configs, etc...
+- **Admin** is a JS browser client application to manage the chatbots, knowledge bases, retrievers, models, configs, etc...
 
 
 - **Ray workers** are used to run distributed inference on the models.
@@ -26,7 +26,7 @@ The system comprises seven main components, here are their relationships and the
 - **Channel layer** (Redis) is used to communicate through WebSockets between the back-end and the SDK, admin and widget.
 
 
-- **Relational Database** (PostgreSQL) is used to store all the data related to the chatbots, datasets, retriever, models, RAG configs, etc...
+- **Relational Database** (PostgreSQL) is used to store all the data related to the chatbots, knowledge bases, retrievers, models, configs, etc...
 
 ### Docker
 
@@ -101,8 +101,6 @@ To manage the chatbot and view the model configuration we provided, go to:
 
 ## Deeper into ChatFAQ
 
-If you want to use your own dataset, you can check the [Dataset Configuration](./modules/configuration/index.html#knowledge-base) documentation.
-
-If you want to learn how to configure your own RAG (LLM model, retriever model, prompt configuration, etc...) you can check the [RAG Configuration](./modules/configuration/index.html#rag-config) documentation.
+If you want to use your own knowledge base, you can check the [Knowledge Base Configuration](./modules/configuration/index.html#knowledge-base) documentation.
 
 If you want to learn how to use the SDK, so you can create your own chatbot behavior, you can check the [SDK](./modules/sdk/index.md) documentation.
diff --git a/doc/source/modules/configuration/index.md b/doc/source/modules/configuration/index.md
@@ -32,7 +32,7 @@ Here you can set up multiple AI components that can be used to compose AI pipeli
 - You can create a simple chatbot using only a LLM.
 - You can create any kind of agent using the ChatFAQ SDK, given that the backend supports tool use.
 
-ChatFAQ provide in its fixtures a default configuration for each of these components except for the Knowledge Base and the RAG Config. You can apply the fixtures by simply running the following command from the `back` directory:
+ChatFAQ provide in its fixtures a default configuration for each of these components. You can apply the fixtures by simply running the following command from the `back` directory:
 
 ```bash
 make apply_fixtures
@@ -327,7 +327,7 @@ from chatfaq_sdk import ChatFAQSDK
 from chatfaq_sdk.clients import retrieve
 from chatfaq_sdk.layers import Message
 
-async def send_rag_answer(sdk: ChatFAQSDK, ctx: dict):
+async def send_retrieval(sdk: ChatFAQSDK, ctx: dict):
     query = 'What is ChatFAQ?'
     items = await retrieve(sdk, 'chatfaq_retriever', query, top_k=3, bot_channel_name=ctx["bot_channel_name"])
     yield Message(
@@ -344,7 +344,7 @@ from chatfaq_sdk import ChatFAQSDK
 from chatfaq_sdk.clients import llm_request
 from chatfaq_sdk.layers import Message, StreamingMessage
 
-async def send_rag_answer(sdk: ChatFAQSDK, ctx: dict):
+async def send_llm_answer(sdk: ChatFAQSDK, ctx: dict):
     # Some messages
     messages = [
         {"role": "system", "content": "You are a helpful assistant."},
@@ -353,3 +353,38 @@ async def send_rag_answer(sdk: ChatFAQSDK, ctx: dict):
     generator = llm_request(sdk, 'gpt-4o', use_conversation_context=False, conversation_id=ctx["conversation_id"], bot_channel_name=ctx["bot_channel_name"], messages=messages)
     yield StreamingMessage(generator)
 ```
+
+Creating a simple RAG pipeline:
+
+```python
+from chatfaq_sdk import ChatFAQSDK
+from chatfaq_sdk.clients import retrieve, llm_request
+from chatfaq_sdk.layers import Message, StreamingMessage
+from chatfaq_sdk.utils import convert_mml_to_llm_format
+
+async def send_rag_answer(sdk: ChatFAQSDK, ctx: dict):
+
+    messages = convert_mml_to_llm_format(ctx["conv_mml"][1:])
+    last_user_message = messages[-1]["content"]
+    
+    # Retrieve context
+    contexts = await retrieve(sdk, 'chatfaq_retriever', last_user_message, top_k=3, bot_channel_name=ctx["bot_channel_name"])
+    
+    # Augment prompt with context
+    system_prompt = rag_system_prompt
+    context_content = "\n".join([f"- {context['content']}" for context in contexts['knowledge_items']])
+    system_prompt += f"\nInformation:\n{context_content}"
+    messages.insert(0, {"role": "system", "content": system_prompt})
+    
+    # Generate response
+    generator = llm_request(
+        sdk,
+        "gpt-4o",
+        use_conversation_context=False,
+        conversation_id=ctx["conversation_id"],
+        bot_channel_name=ctx["bot_channel_name"],
+        messages=messages,
+    )
+
+    yield StreamingMessage(generator, references=contexts)
+```
diff --git a/docker-compose.yaml b/docker-compose.yaml
@@ -53,7 +53,7 @@ services:
       args:
         - INSTALL_CHAT_RAG=true
     container_name: chatfaq_ray
-    command: ["/.venv/bin/ray", "start", "--address=back:6375", "--resources={\"tasks\": 100, \"rags\": 100}", "--block"]
+    command: ["/.venv/bin/ray", "start", "--address=back:6375", "--resources={\"tasks\": 100, \"ai_components\": 100}", "--block"]
     networks:
       - default
       - chatfaq_db_net
diff --git a/sdk/README.md b/sdk/README.md
@@ -77,25 +77,25 @@ import random
 from chatfaq_sdk import ChatFAQSDK
 from chatfaq_sdk.fsm import FSMDefinition, State, Transition
 from chatfaq_sdk.conditions import Condition
-from chatfaq_sdk.layers import Text
+from chatfaq_sdk.layers import Message
 ```
 
 Declare the 3 possible states of our FSM:
 
 ```python
 def send_greeting(ctx: dict):
-    yield Text("Hello!")
-    yield Text("How are you?", allow_feedback=False)
+    yield Message("Hello!")
+    yield Message("How are you?", allow_feedback=False)
 
 greeting_state = State(name="Greeting", events=[send_greeting], initial=True)
 
 
 def send_answer(ctx: dict):
-    last_payload = ctx["conv_mml"][-1]["stack"][0]["payload"]
-    yield Text(
+    last_payload = ctx["conv_mml"][-1]["stack"][0]["payload"]["content"]
+    yield Message(
         f'My answer to your message: "{last_payload}" is: {random.randint(0, 999)}'
     )
-    yield Text(f"Tell me more")
+    yield Message("Tell me more")
 
 answering_state = State(
     name="Answering",
@@ -104,7 +104,7 @@ answering_state = State(
 
 
 def send_goodbye(ctx: dict):
-    yield Text("Byeeeeeeee!", allow_feedback=False)
+    yield Message("Byeeeeeeee!", allow_feedback=False)
 
 goodbye_state = State(
     name="Goodbye",
@@ -117,7 +117,7 @@ Declare the only computable condition for the transitions of our FSM:
 
 ```python
 def is_saying_goodbye(ctx: dict):
-    if ctx["conv_mml"][-1]["stack"][0]["payload"] == "goodbye":
+    if ctx["conv_mml"][-1]["stack"][0]["payload"]["content"] == "goodbye":
         return Condition(1)
     return Condition(0)
 ```
@@ -163,60 +163,4 @@ sdk = ChatFAQSDK(
 sdk.connect()
 ```
 
-The resulting FSM looks like this:
 
-## Other useful info
-### Build the docs
-
-go inside the `doc` directory and run:
-
-```
-poetry run make html
-```
-
-
-### Publish package
-
-#### PYPI test
-
-add repository to poetry config
-
-    poetry config repositories.chatfaq-sdk https://test.pypi.org/legacy/
-
-get token from https://test.pypi.org/manage/account/token/
-
-store token using
-
-    poetry config pypi-token.chatfaq-sdk pypi-YYYYYYYY
-
-#### PYPI production
-
-get token from https://pypi.org/manage/account/token/
-
-store token using
-
-    poetry config pypi-token.chatfaq-sdk pypi-XXXXXXXX
-
-Each time you need to publish
-
-Bump version
-
-    poetry version prerelease
-
-or
-
-    poetry version patch
-
-#### Poetry Build
-
-    poetry build
-
-#### Poetry Publish
-
-To TestPyPi
-
-    poetry publish -r chatfaq-sdk
-
-To PyPi
-
-    poetry publish

Original file line number	Diff line number	Diff line change
`@@ -14,7 +14,7 @@`
`14`	`14`	`ray_actor_options={`
`15`	`15`	`"num_cpus": 1,`
`16`	`16`	`"resources": {`
`17`		`- "rags": 1,`
	`17`	`+ "ai_components": 1,`
`18`	`18`	`},`
`19`	`19`	`},`
`20`	`20`	`)`
Original file line number	Diff line number	Diff line change
`@@ -13,7 +13,7 @@`
`13`	`13`	`ray_actor_options={`
`14`	`14`	`"num_cpus": 1,`
`15`	`15`	`"resources": {`
`16`		`- "rags": 1,`
	`16`	`+ "ai_components": 1,`
`17`	`17`	`}`
`18`	`18`	`}`
`19`	`19`	`)`
Original file line number	Diff line number	Diff line change
`@@ -10,7 +10,7 @@`
`10`	`10`	`ray_actor_options={`
`11`	`11`	`"num_cpus": 0.01,`
`12`	`12`	`"resources": {`
`13`		`- "rags": 1,`
	`13`	`+ "ai_components": 1,`
`14`	`14`	`},`
`15`	`15`	`},`
`16`	`16`	`)`
Original file line number	Diff line number	Diff line change
`@@ -393,8 +393,8 @@ def generate_suggested_intents_task(`
`393`	`393`	`Message.objects.filter(`
`394`	`394`	`stack__contains=[`
`395`	`395`	`{`
`396`		`- "payload": {"references": {"knowledge_base_id": "1"}},`
`397`		`- "type": "rag_generated_text",`
	`396`	`+ "payload": {"references": {"knowledge_base_id": knowledge_base_pk}},`
	`397`	`+ # "type": "rag_generated_text",`
`398`	`398`	`}`
`399`	`399`	`]`
`400`	`400`	`)`
`@@ -410,7 +410,7 @@ def generate_suggested_intents_task(`
`410`	`410`	`messages = Message.objects.filter(id__in=Subquery(subquery), sender__type="human")`
`411`	`411`
`412`	`412`	`messages_text = [`
`413`		`- Message.objects.get(id=item["message_id"]).stack[0]["payload"]`
	`413`	`+ Message.objects.get(id=item["message_id"]).stack[0]["payload"]["content"]`
`414`	`414`	`for item in messages`
`415`	`415`	`]`
`416`	`416`
`@@ -428,7 +428,7 @@ def generate_suggested_intents_task(`
`428`	`428`	`< new_intents_thresholds["max"]`
`429`	`429`	`]`
`430`	`430`
`431`		`- messages_text = [message.stack[0]["payload"] for message in messages]`
	`431`	`+ messages_text = [message.stack[0]["payload"]["content"] for message in messages]`
`432`	`432`
`433`	`433`	`logger.info(f"Number of messages after filtering: {messages.count()}")`
`434`	`434`
Original file line number	Diff line number	Diff line change
`@@ -1,3 +1,3 @@`
`1`	`1`	`export $(xargs <.env)`
`2`	`2`
`3`		`-ray start --address=back:6375 --resources='{"tasks": 100, "rags": 100}' --block`
	`3`	`+ray start --address=back:6375 --resources='{"tasks": 100, "ai_components": 100}' --block`