I am getting "Fetch Failed" error. My flowise is running in docker on localhost :3000 and ollama server is running on the machine (not docker) at localhost:11434? Can you please help me? Has flowise running in docker got to do anything with this issue? I can run ChatOpenAi without any problem, its just Ollama.
Leon, How to make Flow send Media (image, audio, video, PDF and other files) during the conversation and not just links? for example: Get information from a Google Drive.
Thanks for the video tutorial, but embedding is not working for me, the Ollama server is running, and embedding activate mmap, this is the error 2024-04-15 22:51:49 [ERROR]: fetch failed TypeError: fetch failed at Object.fetch (node:internal/deps/undici/undici:11372:11) at async OllamaEmbeddings._request (D:\flowise\Flowise ode_modules\.pnpm\@langchain+community@0.0.39_@aws-crypto+sha256-js@5.2.0_@aws-sdk+client-bedrock-runtime@3.422_sdjgbtbvm2dvzs44hyiv6rdbae ode_modules\@langchain\community\dist\embeddings\ollama.cjs:110:26) at async RetryOperation._fn (D:\flowise\Flowise ode_modules\.pnpm\p-retry@4.6.2 ode_modules\p-retry\index.js:50:12) 2024-04-15 22:51:49 [ERROR]: [server]: Error: TypeError: fetch failed Error: TypeError: fetch failed at buildFlow (D:\flowise\Flowise\packages\server\dist\utils\index.js:415:19) at async utilBuildChatflow (D:\flowise\Flowise\packages\server\dist\utils\buildChatflow.js:229:36) at async createInternalPrediction (D:\flowise\Flowise\packages\server\dist\controllers\internal-predictions\index.js:7:29) 2024-04-15 22:55:43 [INFO]: PUT /api/v1/chatflows/0d375ada-df1f-4d66-941e-1f495ea9f4e5 2024-04-15 22:55:48 [INFO]: POST /api/v1/vector/internal-upsert/0d375ada-df1f-4d66-941e-1f495ea9f4e5 2024-04-15 23:00:50 [ERROR]: TypeError: fetch failed Error: TypeError: fetch failed at InMemoryVectorStore_VectorStores.upsert (D:\flowise\Flowise\packages\components\dist odes\vectorstores\InMemory\InMemoryVectorStore.js:26:27) at async buildFlow (D:\flowise\Flowise\packages\server\dist\utils\index.js:352:37) at async upsertVector (D:\flowise\Flowise\packages\server\dist\utils\upsertVector.js:117:32) at async Object.upsertVectorMiddleware (D:\flowise\Flowise\packages\server\dist\services\vectors\index.js:9:16) at async createInternalUpsert (D:\flowise\Flowise\packages\server\dist\controllers\vectors\index.js:28:29) 2024-04-15 23:00:50 [ERROR]: [server]: Error: Error: TypeError: fetch failed Error: Error: TypeError: fetch failed at buildFlow (D:\flowise\Flowise\packages\server\dist\utils\index.js:415:19) at async upsertVector (D:\flowise\Flowise\packages\server\dist\utils\upsertVector.js:117:32)
i like this video cause it's free of nonsense. if you can mention how to cancel the ollama session and remove the call for action to subscribe will be perfect
Hi Leon. I have tried but I get the fetch failed error. I get the message "Ollama is running" when I run the ip localhost, but somehow flowise doesn't detect it. What could it be? Thanks!
I did everything exactly the same as you for the rag and made sure that ollama server is running (and talked to the model in the terminal and responded find immediately) also made sure mmap is on but still get Error: "Error: Request to Ollama server failed: 404 Not Found".
Yeah, these models are resource intensive. You could always try smaller models. Kaggle is a no-go. The point of this video, and Ollama, is to run these models locally. We will look at using hosted solutions as well though, like Huggingface. But, again, this will / could result in costs for the API usage OR infrastructure. There is a reason why OpenAI is so popular 😊
You're welcome 🤗. The folder loader should be easy enough to add. Take note that it only works when running Flowise on your local machine. Simply create a folder on your machine, copy the path to the folder and add it to the folder loader. If you wanted to do something in the cloud, then you'd need to use something like the S3 file / bucket loader. Hope this helps.
Great stuff. I tried some of the earlier tutorials using Ollama instead of OpenAI (the posiitve/negative review reply tutorial) and found the if/else didn't work with llama3. I'd like to learn more about how to make agents that are completely local and don't use any 3rd party services like open ai, pineapple, etc., but maybe that's not possible without too much missing functionality.
Hey Leon, amazing tutorials, thank you I'm trying to make a project on Flowise using your tutorials, maybe you could help with two questions: 1. Is there a way to make a Multi-Agent system where Agents with different roles and functions can give instructions or feedback to each other before executing? (like AutoGen) I saw your tutorial on how to make something like this using Conversation Chain, but is it possible to make more advanced system with Agents? 2. How do I load images from external URLs? I don't see such a template. If I upload a pdf with an image will it work? Thanks!
Your all Videos are great. I am creating small web applications using flowise. Can you create one video based on adding streaming effect after creating API and that should reflect in web application? This will be very useful.
Hi Leon, how are you? I'm from Brazil and I'm advancing my studies in the area of automation in general using AI resources and when looking for videos about Flowise I found your channel. Since yesterday I've been watching all the videos on the playlist about Flowise. I want to express my gratitude for your dedication and teachings. Thank you very much!
Hi Leon, very useful tutorial!!! I'm running this on CPU (8 vCPUs + 30Gb of RAM) and it is extremely slow. In fact Ollama uses just a few resources and I can't make it use all available CPUs or RAM. I know that GPU is the way to go with LLMs, but perhaps you have some suggestions on how to make this configuration perform a little better. Thank you!!!
@@leonvanzyl Nope I havent tried but will try and get back to you. Thanks for the response i thought I should not use flowise but please help us we will make great llms. Hope you will solve our problems
Hi Leon! Thank you for sharing another excellent content. I´m struggling to figure out how to insert a system prompt along with my own prompt into the flow of a chatbot I'm creating on Flowise. I'm using an Open Source uncensored LLM via Replicate, and it needs a system prompt to behave the way I'd like. I would be very grateful if you could help me in any way.
You can set the system prompt by clicking on Additional Parameters on the chain, or you can assign a Chat Prompt Template. Apologies if is misunderstood ☺️
Hi Leon, great content. Really love your contents and presentation. Can you please make a video of RAG, where you embed different types of files, like CSV, pdf, doc and do the chats with those. Thanks
Very few models support function calling. It's actually limited to OpenAI and Mistral at the moment. You could therefore simply download the Mistral model in Ollama 👍. Be warned, the hardware requirements for Mistral function calling is steep 😄
nice tutorial leon, few questions 1. can we use these Open Source Models to create a chatbot and give to clients, if yes then where will it reside ?? 2. For data privacy its a good option, but how to make it to production OR production ready deployment keeping privacy factor.
Thanks! Then point of the video is to run the bots locally. If you want to use these models in the cloud you would need to use hosted services like Huggingface or AWS Bedrock. I'll definitely release a video on these. These is a cost involved in using these services of course, so I just wanted to give you guys a free local alternative.
You don't need to always host everything in cloud. You can also have your own server at home, garage, office.. All you need is good enough hardware for the model and public IP with well-configured networking. But local hosting is a whole different topic to learn. So you either go: - Self-hosted server - Cloud - Combination of both (e.g. hosting Flowise, files and databases on your own server and calling your model hosted on stateless cloud compute endpoint from it)
I've been watching your work on youtube for a very long time and wanted to say thank you very much for what you do and wanted to wish you a lot of luck for your wishes!
Hi Leon, thanks a lot for the video. For my purpose i used nomic-embed-text as llm for embedding, it s faster. I managed to connect directly my ollama+flowise customtool to my crm api .. but it worked only with llama2 model and not mistral ... i struggled a while then found the trick ! no need to interface to make or n8n ... i'm still working on it ..Cheers
@@leonvanzyl for now i have good results with llava:13b or llama2, they are the only one which use the custom tool on flowise with ollama (with the others no results : nexusraven, orca2,gemma,phi,openchat,mistral ....) i'll keep you in touch
Been mind-blown by pretty much every single one of your videos / tutorials 👏, I'm super grateful! 🙏 Just a thought, how about an entirely open source and local stack running on Docker - I mean, for example, Flowise, ChromaDB, Mixtral (a quantized model running through Ollama?) running in different containers locally - And then, wow, a deployment pipeline to, say, Google Cloud with a CI/CD script via Github Actions - So, dev locally with the Docker containers, push to Github when satisfied, automatically build and deploy on Github Actions and boom, app available on Google Cloud - That would be incredible! I'm going to try it right now, lots of research and trials and errors ahead... 😁
@@leonvanzyl Looks like ```ollama run mixtral:8x7b``` is going to be a little challenging for my modest hardware 😁 Gonna try with ``` ollama run gemma:2b```, may be even ```gemma:7b```, it is still open source and apparently it is lightweight...
Thank you for your video!!!!! just wonder if we host everything locally on Flowise, after we set It up, can we generate the Python API and get to use it somewhere else? or we can only use API generated by flowise locally as well in VS code?
If you're hosting it locally, then you can only access it locally. I have a video on deploying Flowise in this series, but I'm guessing you want to use Open Source models in the cloud? Your best option is to use Huggingface (video coming soon).
thank you so much, you literally save my life. I been considering hugging face as well, but open source models like Mixtral model , it doesn't work with outputparaser of langchain, only openai model
do you think i can deploy the Ollama locally and flowise locally and build a outputparser framework, and eventually get to use it through local API generated by local flowise ? my goal is to find a cheaper way to run outputparser with a decent performance , since openai will cost so much @@leonvanzyl
i juste have an issue with openai api key it shows me this error when i try to run the flowise.. => "InsufficientQuotaError: 429 You exceeded your current quota...." should i pay tokens ? @@leonvanzyl
great video as always. thx for the premium content you deliver. 🦾💙 is this only for local models or could I use open router credentials here, or would I do dedicated API agent instead of chatbot? sorry for confusion =D
