Comment on page
Pipelines
This may happen when your input is too large. Each LLM has a maximum number of input and output tokens it can process:
- GPT-3.5 can support up to 4096 tokens
- GPT-3.5-16k can support up to 16384 tokens
- GPT-4 can support up to 8192 tokens
If your input is too large, the model will be constrained to significantly shorten the output. To avoid running into the input limit, you can pass your inputs through a VectorDB Loader / VectorDB Reader or load them into a Vector Store before passing them into your LLM to reduce your token count.
VectorDB Readers take databases (i.e., VectorDB Loaders) and queries as inputs and return the the parts of the database that are most relevant to your query. For example, if you load a VectorDB Loader with your website's documentation and ask a question about a specific feature in your query, the VectorDB Reader will return the parts of your documentation relevant to the feature you queried.
Click the three dots on the right hand side of a pipeline and click "Share Pipeline". You will be brought to a pop up where you can input the email address of the recipient you want to share the pipeline with. You can adjust whether the recipient has "run" or "edit" access, or remove run / edit access of a recipient you have shared with previously. Click " save" to save the settings.
In the LLM node within the pipeline, click on the gear and then click "Stream Response".
Last modified 2d ago