Basically, the streaming response for gpt3.5 turbo works fine (via Python), however, words are always sent separately, as seen in this example:
For example, the word “Embrace”:
“choices”: [
{
“delta”: {
“content”: “Em”
},
“finish_reason”: null,
“index”: 0
}
}, “choices”: [
{
“delta”: {
}, “content”: { “brace”.
},
“finish_reason”: null,
“index”: 0
Does anyone have a similar problem and know a solution for this?
1 Like
It’s responding with tokens, not words. This is normal behavior.
1 Like
I messed with the app. it does hamburgers with hamburg ers.
isnt that interesting?
Ok thanks for the answer. If I now want to display the response in real time, what is the best way to put the separated words back together? If there is no blank, then simply concatenate?
Im not sure, try unparsing them, or just make the parse an “”.
But I am not sure that is what you are talking about.
You should be able to simple concentrate everything directly—spaces will be included in tokens as needed.
1 Like
