hi, is this just me or anyone else or we are getting inconsistent results with deepgram flux model i have given

"model": "flux-general-en",
                    "eot_threshold": 0.9,
                    "eot_timeout_ms": 9000,
                    "mip_opt_out": true

livekit agents and plugins versions - 1.2.18 basically in some cases the transcript itself is wrong or incomplete and even though the eot_timeout_ms is 9 seconds the delay is at max 6 seconds, sometimes 3 and sometimes 1 for clear incomplete sentences

Not sure if it will solve your issue but there were improvements to deepgram in version 1.3.6

@refined-appointment-81829 ok will update and test