xbpeng121 comments

Results 4 comments of


                                            xbpeng121

A simple image understanding workflow runs out of time

I'm interested in this issue.

API calls are not efficient for multiple qps

I think there is nothing about node.js，the workflow apis are on the port 5001 which is hosted by the api_server written by python on Flask framework

LLM node receives multiple base64-encoded images

I also need this feature that "llm node" can accept base64-image as vision content. not only from "HTTP node", but also other nodes, for example "start node". The scenario is...

when running a single LLM Node(with vision model), if the input image is self-define in the Start Node，the LLM Node cannot give expected output

the problem is Dify used hard code for "sys.files". to fix it just change one line in "api/core/workflow/nodes/llm/node.py" file, in the function "_extract_variable_selector_to_variable_mapping"， `if node_data.vision.enabled: variable_mapping["#files#"] = ["sys", SystemVariableKey.FILES.value]` change...