Local LLMs into main (#1351)

* fix * fixing the toolkit config in iteration workflow * List file s3 fix (#1076) * List File S3 * Unit Test Added --------- Co-authored-by: Taran <97586318+Tarraann@users.noreply.github.com> * workflow changes * minor seed file fix * frontend fixes (#1079) * fixed api_bug (#1080) * add one condition (#1082) * api fix (#1087) * Tools error fix (#1093) * webhooks frontend + api calls complete almost * Tool-LTM(Updated) (#1039) * Toolkit configuration fix (#1102) * webhooks compplete frontend * schedule agent fix (#1104) Co-authored-by: Rounak Bhatia <rounak@contlo.com> * Models superagi (#936) Models Superagi * Models superagi (#1108) Bug Fixes * Changes for no receiver address * made changes to github helper * Models superagi (#1112) * Models superagi (#1117) * Models fixes (#1118) * Models Frontend Changes * Models Frontend Changes * Models Frontend Changes * \n bug resolved (#1122) * PDF and DOCX support in Write File - Feature Improvement, close #548 (#928) * Added functions to write various file types and a file handler too * FileManager updated to handle and save HTMLs. * adding PDF + DOCX support to save images * Added Wkhtmltopdf package installation run commands in docker * Added get_all_responses feature for extractng the response for particular tools * Added Image embedding feature, this will extract and embed the images generated by the agent during the run * renaming functions and refactoring * renaming functions and refactoring * Update Dockerfile * removing unsused classmethods * Finding generated images and attached files in the write tool. Images are fetched inorder to be embedded in the respective file type. * Adding the filename and paths to the Resource Manager S3 storage * Code Cleanup * added logger: Fix for the failing TEST --------- Co-authored-by: Fluder-Paradyne <121793617+Fluder-Paradyne@users.noreply.github.com> * Revert "PDF and DOCX support in Write File - Feature Improvement, close #548 (#928)" (#1124) This reverts commit 8b01357. * expose port * latest safetensors breaking in macs (#1134) * Changes in save template (#1120) * fix * changes in save template * changes in save template * bug fixes * changes in save template * changes in save template --------- Co-authored-by: Rounak Bhatia <rounak@contlo.com> Co-authored-by: namansleeps <mandhani12@gmail.com> * Main to dev sync (#1139) * Models fixes (#1126) * Models Frontend Changes * Models Frontend Changes * Models Frontend Changes * Backend Compatibility for New/Existing users on local * DEV api key requirements * removing print statements * removing print statements * removing print statements * removing print statements * backend compatibility * backend compatibility * backend compatibility * added filters in the webhooks * fix * added filters in the webhooks * Models fixes (#1145) Fixes related to Models Feature * Jira Bug Fix * Jira Bug Fix 2.0 * Jira Bug Fix 3.0 * added filters in the webhooks * Models fixes (#1147) Model Feature Fixes * Bug fix model redirection (#1148) Bug Fix - Model URL Redirection * added tool config for dalle * removed model dependency on dalle tool * Remove hardcoded creds * fixed env error * removed refactoring from main * removed refactoring * removed refactoring * handled error * stop agent from executing if model is not found (#1156) * entity details (#1158) * Metric frontend (#1152) * added filters in webhooks * added filters in webhooks * minor changes * webhooks complete * minor changes for PR * minor changes for PR * Publish agent template to marketplace (#1106) * publish agent to marketplace --------- Co-authored-by: Rounak Bhatia <rounak@contlo.com> Co-authored-by: namansleeps <mandhani12@gmail.com> * added filters in webhooks * added filters in webhooks * resolving conflicts * added filters in the webhooks * lint issue fixed * bug fix of prev PR * fix for new run and edit agent * error handling * added filters in webhooks * fix for knowledge search tool * Docker digitalocean deployment * changed branch name * added filters in the webhooks * changes * removed region * added button * change in branch * added filters in the webhooks * Update conftest.py * Added filters in the webhooks (#1140) * webhooks frontend + web hooks with filters --------- Co-authored-by: namansleeps <mandhani12@gmail.com> Co-authored-by: Fluder-Paradyne <121793617+Fluder-Paradyne@users.noreply.github.com> * Models calls logs dev (#1174) Call logs organisation level bug fix * models scroll fix, format of log timestamp fix, adding of loader to models, toolkit metrics dropdown bug fixed, publish agent dropdown bug (#1171) * Update app.yaml (#1179) * fixes related to webhooks * fixes for webhooks * Fixes for webhooks (#1181) * fixes for webhooks --------- Co-authored-by: namansleeps <mandhani12@gmail.com> Co-authored-by: Fluder-Paradyne <121793617+Fluder-Paradyne@users.noreply.github.com> * bugs by qa (#1178) * Fix for schedule agent (#1184) Co-authored-by: Tarraann <jot.taran15522@gmail.com> * Entity fix (#1185) * fixes for webhooks * fixes for webhooks * fix added for index state (#1188) * fix added for index state * Update KnowledgeTemplate.js --------- Co-authored-by: Tarraann <jot.taran15522@gmail.com> * API bug fixes for SDK (#1189) * fix api's for sdk * removed unused imports --------- Co-authored-by: jagtarcontlo <123375045+jagtarcontlo@users.noreply.github.com> * Main to dev sync v12 (#1193) sync back to dev ------ Co-authored-by: Taran <97586318+Tarraann@users.noreply.github.com> Co-authored-by: TransformerOptimus <muknrq@gmail.com> Co-authored-by: Fluder-Paradyne <121793617+Fluder-Paradyne@users.noreply.github.com> Co-authored-by: Maverick-F35 <138012351+Maverick-F35@users.noreply.github.com> Co-authored-by: BoundlessAsura <122777244+boundless-asura@users.noreply.github.com> Co-authored-by: Akshat Jain <92881074+Akki-jain@users.noreply.github.com> Co-authored-by: sayan1101 <139119661+sayan1101@users.noreply.github.com> Co-authored-by: Rounak Bhatia <f20201807@goa.bits-pilani.ac.in> Co-authored-by: Rounak Bhatia <rounak@contlo.com> Co-authored-by: Kalki <97698934+jedan2506@users.noreply.github.com> Co-authored-by: Tarraann <jot.taran15522@gmail.com> Co-authored-by: rakesh-krishna-a-s <akrishna@contlo.com> Co-authored-by: Captain Levi <123375045+CaptainLevi0007@users.noreply.github.com> Co-authored-by: andrew-kelly-neutralaiz <128111428+andrew-kelly-neutralaiz@users.noreply.github.com> Co-authored-by: James Wakelim <james.wakelim@neutralaiz.com> * added button * GitHub pull request tools (#1190) * adding github review tools * cleanup and adding code review prompt * fixing comments * PDF and DOCX support in Write File - Feature Improvement, close #548 (#1125) Co-authored-by: Fluder-Paradyne <121793617+Fluder-Paradyne@users.noreply.github.com> Co-authored-by: Abhijeet <129729795+luciferlinx101@users.noreply.github.com> * minor documentation fix * Design bugs (#1199) * fetching token limit from db * Revert "PDF and DOCX support in Write File - Feature Improvement, close #548 (#1125)" (#1202) This reverts commit 26f6a1d. * Unit Test Fix (#1203) * adding of docs and and discord link correction (#1205) * openai error handling * error_handling * api call only when agent is running * Feature : Wait block for agent workflow (#1186) Agent Wait Block Step * minor changes (#1213) Co-authored-by: Jagtar Saggu <jagtarsaggu@Jagtars-MacBook-Pro.local> * error handling * error handling * error handling * error handling * error handling * fix * fix * fix * error handling * models changes (#1207) Model related frontend changes. * error handling * models marketplace changes (#1219) Co-authored-by: Abhijeet <129729795+luciferlinx101@users.noreply.github.com> * minor changes * error handling * error handling * removing single qoutes (#1224) Co-authored-by: namansleeps <mandhani12@gmail.com> * apm changes (#1222) APM Bug Fixes * list tool fix * list tool fix * PR CHANGES * entity fix for dev (#1230) Entity Fix for Dev * frontend changes (#1231) * read_tool_fix * fix * waiting block frontend (#1233) Waiting Block Changes and Frontend Addition * Dev Fixes (#1242) * read tool fix * Maintaining dev (#1244) Dev Fix * added logs (#1246) * error_handling fix (#1247) Error handling fix. * Feature first login src (#1241) Adding source in user database for analytics. * apollo NoneType bug fix (#1238) Bug Fix * Mixpanel integration (#1256) Mix Panel * Models Marketplace bug fix for dev (#1266) * Fix 1257 dev (#1269) Bug Fix * add cache layer (#1275) Added package caching for github actions workflow * Fix api dev (#1283) * save other config to agent_execution config * add config * mixpanel changes (#1285) * rename error_handling.py to error_handler.py (#1287) * Analytics login (#1258) * calendar issues fixed * append fle tool bug fixed (#1294) Co-authored-by: Tarraann <jot.taran15522@gmail.com> * adding cookie in access token (#1301) * local_llms * local_llms * local_llms * local_llms * local_llms * fixes * models error fixed (#1308) * local_llms * local_llms * local_llms * local_llms * local_llms * frontend_changes * local_llms * local_llms * local_llms * local_llms * local_llms_frontend * fixes * fixes * fixes * fixes * merged main into local_llm_final * merged main into local_llm_final * local llms --------- Co-authored-by: TransformerOptimus <muknrq@gmail.com> Co-authored-by: Abhijeet <129729795+luciferlinx101@users.noreply.github.com> Co-authored-by: Taran <97586318+Tarraann@users.noreply.github.com> Co-authored-by: Fluder-Paradyne <121793617+Fluder-Paradyne@users.noreply.github.com> Co-authored-by: Maverick-F35 <138012351+Maverick-F35@users.noreply.github.com> Co-authored-by: namansleeps <mandhani12@gmail.com> Co-authored-by: Aditya Sharma <138581531+AdityaSharma13064@users.noreply.github.com> Co-authored-by: Sayan Samanta <139119661+sayan1101@users.noreply.github.com> Co-authored-by: Kalki <97698934+jedan2506@users.noreply.github.com> Co-authored-by: Tarraann <jot.taran15522@gmail.com> Co-authored-by: Arkajit Datta <61142632+Arkajit-Datta@users.noreply.github.com> Co-authored-by: rakesh-krishna-a-s <akrishna@contlo.com> Co-authored-by: jagtarcontlo <123375045+jagtarcontlo@users.noreply.github.com> Co-authored-by: Captain Levi <123375045+CaptainLevi0007@users.noreply.github.com> Co-authored-by: namansleeps <122260931+namansleeps@users.noreply.github.com> Co-authored-by: I’m <133493246+TransformerOptimus@users.noreply.github.com> Co-authored-by: Jagtar Saggu <jagtarsaggu@Jagtars-MacBook-Pro.local> Co-authored-by: Ubuntu <ubuntu@ip-10-14-2-4.ec2.internal>
TransformerOptimus · Oct 25, 2023 · b96fdab · b96fdab
1 parent 862a701
commit b96fdab
Show file tree

Hide file tree

Showing 10 changed files with 198 additions and 2 deletions.
diff --git a/gui/pages/Content/APM/ApmDashboard.js b/gui/pages/Content/APM/ApmDashboard.js
@@ -76,7 +76,7 @@ export default function ApmDashboard() {
     const fetchData = async () => {
       try {
         const [metricsResponse, agentsResponse, activeRunsResponse, toolsUsageResponse] = await Promise.all([getMetrics(), getAllAgents(), getActiveRuns(), getToolsUsage()]);
-        const models = ['gpt-4', 'gpt-3.5-turbo', 'gpt-3.5-turbo-16k', 'gpt-4-32k', 'google-palm-bison-001'];
+        const models = ['gpt-4', 'gpt-3.5-turbo', 'gpt-3.5-turbo-16k', 'gpt-4-32k', 'google-palm-bison-001', 'replicate-llama13b-v2-chat'];
 
         assignDefaultDataPerModel(metricsResponse.data.agent_details.model_metrics, models);
         assignDefaultDataPerModel(metricsResponse.data.tokens_details.model_metrics, models);

diff --git a/gui/pages/_app.js b/gui/pages/_app.js
@@ -61,6 +61,7 @@ export default function App() {
       });
   }
 
+
   const installFromMarketplace = () => {
     const toolkitName = localStorage.getItem('toolkit_to_install') || null;
     const agentTemplateId = localStorage.getItem('agent_to_install') || null;

diff --git a/main.py b/main.py
@@ -50,6 +50,7 @@
 from superagi.llms.replicate import Replicate
 from superagi.llms.hugging_face import HuggingFace
 from superagi.models.agent_template import AgentTemplate
+from superagi.models.models_config import ModelsConfig
 from superagi.models.organisation import Organisation
 from superagi.models.types.login_request import LoginRequest
 from superagi.models.types.validate_llm_api_key_request import ValidateAPIKeyRequest
@@ -215,6 +216,13 @@ def register_toolkit_for_master_organisation():
             Organisation.id == marketplace_organisation_id).first()
         if marketplace_organisation is not None:
             register_marketplace_toolkits(session, marketplace_organisation)
+
+    def local_llm_model_config():
+        existing_models_config = session.query(ModelsConfig).filter(ModelsConfig.org_id == default_user.organisation_id, ModelsConfig.provider == 'Local LLM').first()
+        if existing_models_config is None:
+            models_config = ModelsConfig(org_id=default_user.organisation_id, provider='Local LLM', api_key="EMPTY")
+            session.add(models_config)
+            session.commit()
 
     IterationWorkflowSeed.build_single_step_agent(session)
     IterationWorkflowSeed.build_task_based_agents(session)
@@ -238,7 +246,8 @@ def register_toolkit_for_master_organisation():
     # AgentWorkflowSeed.doc_search_and_code(session)
     # AgentWorkflowSeed.build_research_email_workflow(session)
     replace_old_iteration_workflows(session)
-
+    local_llm_model_config()
+
     if env != "PROD":
         register_toolkit_for_all_organisation()
     else:

diff --git a/migrations/versions/9270eb5a8475_local_llms.py b/migrations/versions/9270eb5a8475_local_llms.py
@@ -0,0 +1,28 @@
+"""local_llms
+
+Revision ID: 9270eb5a8475
+Revises: 3867bb00a495
+Create Date: 2023-10-04 09:26:33.865424
+
+"""
+from alembic import op
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision = '9270eb5a8475'
+down_revision = '3867bb00a495'
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.add_column('models', sa.Column('context_length', sa.Integer(), nullable=True))
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.drop_column('models', 'context_length')
+    # ### end Alembic commands ###
diff --git a/requirements.txt b/requirements.txt
@@ -158,3 +158,4 @@ google-generativeai==0.1.0
 unstructured==0.8.1
 ai21==1.2.6
 typing-extensions==4.5.0
+llama_cpp_python==0.2.7
diff --git a/superagi/helper/llm_loader.py b/superagi/helper/llm_loader.py
@@ -0,0 +1,38 @@
+from llama_cpp import Llama
+from llama_cpp import LlamaGrammar
+from superagi.config.config import get_config
+from superagi.lib.logger import logger
+
+
+class LLMLoader:
+    _instance = None
+    _model = None
+    _grammar = None
+
+    def __new__(cls, *args, **kwargs):
+        if cls._instance is None:
+            cls._instance = super(LLMLoader, cls).__new__(cls)
+        return cls._instance
+
+    def __init__(self, context_length):
+        self.context_length = context_length
+
+    @property
+    def model(self):
+        if self._model is None:
+            try:
+                self._model = Llama(
+                    model_path="/app/local_model_path", n_ctx=self.context_length)
+            except Exception as e:
+                logger.error(e)
+        return self._model
+
+    @property
+    def grammar(self):
+        if self._grammar is None:
+            try:
+                self._grammar = LlamaGrammar.from_file(
+                    "superagi/llms/grammar/json.gbnf")
+            except Exception as e:
+                logger.error(e)
+        return self._grammar
diff --git a/superagi/jobs/agent_executor.py b/superagi/jobs/agent_executor.py
@@ -1,6 +1,7 @@
 from datetime import datetime, timedelta
 
 from sqlalchemy.orm import sessionmaker
+from superagi.llms.local_llm import LocalLLM
 
 import superagi.worker
 from superagi.agent.agent_iteration_step_handler import AgentIterationStepHandler

diff --git a/superagi/llms/grammar/json.gbnf b/superagi/llms/grammar/json.gbnf
@@ -0,0 +1,25 @@
+root   ::= object
+value  ::= object | array | string | number | ("true" | "false" | "null") ws
+
+object ::=
+  "{" ws (
+            string ":" ws value
+    ("," ws string ":" ws value)*
+  )? "}" ws
+
+array  ::=
+  "[" ws (
+            value
+    ("," ws value)*
+  )? "]" ws
+
+string ::=
+  "\"" (
+    [^"\\] |
+    "\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F]) # escapes
+  )* "\"" ws
+
+number ::= ("-"? ([0-9] | [1-9] [0-9]*)) ("." [0-9]+)? ([eE] [-+]? [0-9]+)? ws
+
+# Optional space: by convention, applied in this grammar after literal chars when allowed
+ws ::= ([ \t\n] ws)?
diff --git a/superagi/llms/llm_model_factory.py b/superagi/llms/llm_model_factory.py
@@ -1,4 +1,5 @@
 from superagi.llms.google_palm import GooglePalm
+from superagi.llms.local_llm import LocalLLM
 from superagi.llms.openai import OpenAi
 from superagi.llms.replicate import Replicate
 from superagi.llms.hugging_face import HuggingFace

diff --git a/superagi/llms/local_llm.py b/superagi/llms/local_llm.py
@@ -0,0 +1,92 @@
+from superagi.config.config import get_config
+from superagi.lib.logger import logger
+from superagi.llms.base_llm import BaseLlm
+from superagi.helper.llm_loader import LLMLoader
+
+
+class LocalLLM(BaseLlm):
+    def __init__(self, temperature=0.6, max_tokens=get_config("MAX_MODEL_TOKEN_LIMIT"), top_p=1,
+                 frequency_penalty=0,
+                 presence_penalty=0, number_of_results=1, model=None, api_key='EMPTY', context_length=4096):
+        """
+        Args:
+            model (str): The model.
+            temperature (float): The temperature.
+            max_tokens (int): The maximum number of tokens.
+            top_p (float): The top p.
+            frequency_penalty (float): The frequency penalty.
+            presence_penalty (float): The presence penalty.
+            number_of_results (int): The number of results.
+        """
+        self.model = model
+        self.api_key = api_key
+        self.temperature = temperature
+        self.max_tokens = max_tokens
+        self.top_p = top_p
+        self.frequency_penalty = frequency_penalty
+        self.presence_penalty = presence_penalty
+        self.number_of_results = number_of_results
+        self.context_length = context_length
+
+        llm_loader = LLMLoader(self.context_length)
+        self.llm_model = llm_loader.model
+        self.llm_grammar = llm_loader.grammar
+
+    def chat_completion(self, messages, max_tokens=get_config("MAX_MODEL_TOKEN_LIMIT")):
+        """
+        Call the chat completion.
+
+        Args:
+            messages (list): The messages.
+            max_tokens (int): The maximum number of tokens.
+
+        Returns:
+            dict: The response.
+        """
+        try:
+            if self.llm_model is None or self.llm_grammar is None:
+                logger.error("Model not found.")
+                return {"error": "Model loading error", "message": "Model not found. Please check your model path and try again."}
+            else:
+                response = self.llm_model.create_chat_completion(messages=messages, functions=None, function_call=None, temperature=self.temperature, top_p=self.top_p,
+                                                                 max_tokens=int(max_tokens), presence_penalty=self.presence_penalty, frequency_penalty=self.frequency_penalty, grammar=self.llm_grammar)
+                content = response["choices"][0]["message"]["content"]
+                logger.info(content)
+                return {"response": response, "content": content}
+
+        except Exception as exception:
+            logger.info("Exception:", exception)
+            return {"error": "ERROR", "message": "Error: "+str(exception)}
+
+    def get_source(self):
+        """
+        Get the source.
+
+        Returns:
+            str: The source.
+        """
+        return "Local LLM"
+
+    def get_api_key(self):
+        """
+        Returns:
+            str: The API key.
+        """
+        return self.api_key
+
+    def get_model(self):
+        """
+        Returns:
+            str: The model.
+        """
+        return self.model
+
+    def get_models(self):
+        """
+        Returns:
+            list: The models.
+        """
+        return self.model
+
+    def verify_access_key(self, api_key):
+        return True
-Original file line number
+Diff line change
@@ Expand Up / @@ -61,6 +61,7 @@ export default function App() { @@
           });
       }
       const installFromMarketplace = () => {
         const toolkitName = localStorage.getItem('toolkit_to_install') || null;
         const agentTemplateId = localStorage.getItem('agent_to_install') || null;
@@ Expand Down @@