วีดีโอ

Hi @JeomonGeroge, thanks for your comments. I have to check for an hosted llama/groq provider with OpenAI API compatible service so I can point the langchain code to llama/groq. If you are interested in llamacpp implementation here is a sample: github.com/sravzpublic/ragai/blob/main/src/llama2.py#L74 The sample switches the LLM between chat gpt and llamacpp. github.com/ggerganov/llama.cpp

@@sravzllc I have made the multi agent hierarchial langraph with groq which is similar to yours but it keep on looping it's not going to Finish

@@JeomonGeorge Yes, I'm facing the same issue too, that is holding me from deploying this to Prod. Have you tried to connect the code with langsmith and check the graph: smith.langchain.com/. I noticed this issue: When the supervisor asks the LLM (ChatGPT) which agent next or FInish, ChatGPT seems to give incorrect agent name. So it ends up looping. I'm planning to check the code and see if there is a callback which can be called after an agent + tool run and the supervisor begins the call to LLM for next/FINISH. In that call back (if there is one), update the user message such that when the supervisor asks LLM for the next/FINISH the prompt is updated to just what work is left (do not send the complete initial prompt. I'll investigate further and let you know, in case you find some solution please let me know?

Hi @JeomonGeorge, I updated the system prompt for the supervisor to include description of each agent. Earlier we listed just the agent names, now I added agent descriptions: gist.github.com/sravzpublic/80ce4e6ad8168e9271195a1a5a433c09/revisions#diff-e45c715fe36cbe44080f21c4acf60588992492e52799d94a83bd9f588c77f31aR50. members_description = ["json_agent can handle json input and creates an updated json output", "python_repl_agent can run safe python code"] I also update the user prompt to make it more clear: run_workflow(QueryData(key = "Top_Holdings", query=f"""Using the json_agent get the json key Top_Holdings for the given JSONs then get the top 5 holdings complete JSON object with name, weights etc for each JSON, then flatten the list of list of JSON objects into a single list called flattned_list. Generate safe python code to load the flattned_list to pandas dataframe and save the dataframe to a html file at {output_file_name}. Run the generated python code so the output file {output_file_name} is created""")) In my last 10 tries I did not see a loop. Please check if this works for you and let me know your thoughts?

Hello, we will create another tutorial for that: Below is the docker command: # installing mongocxx driver - connects c++ to mongo RUN cd /usr/src \ && wget github.com/mongodb/mongo-cxx-driver/releases/download/r3.7.1/mongo-cxx-driver-r3.7.1.tar.gz \ && tar -xzf mongo-cxx-driver-r3.7.1.tar.gz \ && cd mongo-cxx-driver-r3.7.1/build \ && cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=/usr/local .. -DCMAKE_CXX_STANDARD=17 -DBSONCXX_POLY_USE_SYSTEM_MNMLSTC=0 -DBSONCXX_POLY_USE_MNMLSTC=0 \ && make \ && make install Here is the build command which include mongo headers and libs: sravz: $(INCLUDE_DIR) $(WEB_INCLUDES) $(INCLUDE_SERVICES_DIR) $(SRC_FILES) $(WEB_SOURCES) $(SERVICES_SOURCES) $(ROUTER_SOURCES) \ $(CONTROLLER_SOURCES) ./src/util.cpp ./include/util.hpp /usr/lib/ccache/g++ -pipe --std=c++17 -Wno-deprecated-declarations -DBOOST_LOG_DYN_LINK \ -I$(REDIS_PLUS_PLUS_MODULES_DIR) \ -I$(INCLUDE_DIR) \ -I$(INCLUDE_SERVICES_DIR) \ -I$(INCLUDE_CONTROLLERS_DIR) \ -I$(INCLUDE_ROUTER_DIR) \ -I/usr/local/include/mongocxx/v_noabi \ -I/usr/local/include/bsoncxx/v_noabi \ -I/usr/local/include/aws/include \ -L/usr/local/lib/ \ -o ./dist/backend-cpp \ $(SRC_FILES) $(WEB_SOURCES) $(SERVICES_SOURCES) $(ROUTER_SOURCES) $(CONTROLLER_SOURCES) \ -lhiredis -lhiredis_ssl -lhiredis_ssl -lredis++ -lpthread -lboost_thread -lboost_program_options \ -lcrypto -lssl -lboost_json -lboost_url -lboost_log -lmongocxx -lbsoncxx \ -lboost_filesystem -lboost_system -lboost_date_time -lboost_regex \ -lboost_log_setup -laws-cpp-sdk-core -laws-cpp-sdk-s3 We will create another video which will use CMAKE to install mongo dependencies. In this video I explained above Make command conversion to CMake: th-cam.com/video/Um_blxn7-4o/w-d-xo.html at a later point I'll update the CMake to install the dependencies.

Hi @rajendranselvaraj4434, Thanks for the kind words. Glad you found the videos helpful, if you have any request on the related topic, please let us know here, we'll try to create a video around that request.

Please post your questions here so the community can benefit

@@sampatp3598oh ok you know his discord link or telegram?

Hi @shubangisaxena2322 I don't have access to Ceph but per Ceph documentation: docs.ceph.com/en/quincy/radosgw/s3/python/ conn = boto.connect_s3( aws_access_key_id = access_key, aws_secret_access_key = secret_key, host = 'objects.dreamhost.com', #is_secure=False, # uncomment if you are not using ssl calling_format = boto.s3.connection.OrdinaryCallingFormat(), ) Please export below two values with your Ceph values // export AWS_ACCESS_KEY_ID=XXX // export AWS_SECRET_ACCESS_KEY=XXX Updated host at this line: clientConfig.endpointOverride = Aws::String("usc1.contabostorage.com"); to your Ceph host and give it a try.