cuda use wmma flash attention for turing (+1 squashed commits)

Squashed commits:

[3c5112398] 117 (+10 squashed commit)

Squashed commit:

[4f01bb2d4] 117 graphs 80v

[7549034ea] 117 graphs

[dabf9cb99] checking if cuda 11.5.2 works

[ba7ccdb7a] another try cu11.7 only

[752cf2ae5] increase aria2c download log rate

[dc4f198fd] test send turing to wmma flash attention

[496a22e83] temp build test cu11.7.0

[ca759c424] temp build test cu11.7

[c46ada17c] test build: enable virtual80 for oldcpu

[3ccfd939a] test build: with cuda graphs for all
This commit is contained in:
Concedo 2025-05-31 18:08:50 +08:00
parent b08dca65ed
commit f3bb947a13
5 changed files with 5 additions and 6 deletions

View file

@ -47,7 +47,7 @@ net_save_slots = 10
# abuse prevention
stop_token_max = 256
ban_token_max = 512
ban_token_max = 768
logit_bias_max = 512
dry_seq_break_max = 128
@ -5912,7 +5912,7 @@ def downloader_internal(input_url, output_filename, capture_output, min_file_siz
a2cexe = (os.path.join(basepath, "aria2c-win.exe"))
if os.path.exists(a2cexe): #on windows try using embedded a2cexe
rc = subprocess.run([
a2cexe, "-x", "16", "-s", "16", "--summary-interval=30", "--console-log-level=error", "--log-level=error",
a2cexe, "-x", "16", "-s", "16", "--summary-interval=20", "--console-log-level=error", "--log-level=error",
"--download-result=default", "--allow-overwrite=true", "--file-allocation=none", "--max-tries=3", "-o", output_filename, input_url
], capture_output=capture_output, text=True, check=True, encoding='utf-8')
dl_success = (rc.returncode == 0 and os.path.exists(output_filename) and os.path.getsize(output_filename) > min_file_size)