Some of these polls have small samples despite being conducted over a five day period (although I guess it's possible a majority of the respondents in some polls were only polled 3/1 and 3/2?).
"Data for Progress conducted a poll of 300 likely TX voters from 02/27/2020 to 03/02/2020. Likely voters were identified from the voter file and weighted to a likely electorate. The survey was conducted via text-to-web and web panel, with 70% from text to web and 30% from panel data. The margin of error is +/- 5.7."
https://www.dataforprogress.org/memos/super-tuesday-final-polling