clore
/
onboarding
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

report clock locks

This commit is contained in:
FellowJacob 2024-10-31 03:21:49 +01:00
parent 2f62997012
commit f170a0fdd5
1 changed files with 37 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
clore_onboarding.py
View File

@ -194,6 +194,21 @@ async def async_read_file(path):
except Exception as e: except Exception as e:
#print(f"Error reading file {path}: {e}") #print(f"Error reading file {path}: {e}")
return None return None
def extract_last_setcore_setmem(input_string):
try:
setcore_pattern = r'--setcore\s+((?:\d+\s*)+)(?=\D|$)'
setmem_pattern = r'--setmem\s+((?:\d+\s*)+)(?=\D|$)'
setcore_matches = re.findall(setcore_pattern, input_string)
setmem_matches = re.findall(setmem_pattern, input_string)
last_setcore = [int(num) for num in setcore_matches[-1].split()] if setcore_matches else []
last_setmem = [int(num) for num in setmem_matches[-1].split()] if setmem_matches else []
return last_setcore, last_setmem
except Exception as e:
return [[], []]
async def hive_parse_config(file_content): async def hive_parse_config(file_content):
conf = {} conf = {}
@ -228,7 +243,10 @@ async def hive_load_configs(default_power_limits, static_config):
# Parse wallet config # Parse wallet config
clore_config = None clore_config = None
get_oc_config = False get_oc_config = False
fs_mem_lock = []
fs_core_lock = []
if wallet_conf_content: if wallet_conf_content:
fs_core_lock, fs_mem_lock = extract_last_setcore_setmem(wallet_conf_content)
for wallet_conf_line in wallet_conf_content.split('\n'): for wallet_conf_line in wallet_conf_content.split('\n'):
wallet_conf_line = wallet_conf_line.strip() wallet_conf_line = wallet_conf_line.strip()
if wallet_conf_line[:1] != "#" and '=' in wallet_conf_line: if wallet_conf_line[:1] != "#" and '=' in wallet_conf_line:
@ -259,6 +277,8 @@ async def hive_load_configs(default_power_limits, static_config):
try: try:
core_offset = None core_offset = None
mem_offset = None mem_offset = None
core_lock = None
mem_lock = None
pl_static = None pl_static = None
for nvidia_conf_line in nvidia_oc.split('\n'): for nvidia_conf_line in nvidia_oc.split('\n'):
nvidia_conf_line = nvidia_conf_line.strip() nvidia_conf_line = nvidia_conf_line.strip()
@ -272,14 +292,30 @@ async def hive_load_configs(default_power_limits, static_config):
mem_offset = [math.floor(num / 2) for num in validate_and_convert(value, -2000, 6000, adjust_bounds=True)] mem_offset = [math.floor(num / 2) for num in validate_and_convert(value, -2000, 6000, adjust_bounds=True)]
elif key=="PLIMIT": elif key=="PLIMIT":
pl_static = validate_and_convert(value, 1, 1500, adjust_bounds=True) pl_static = validate_and_convert(value, 1, 1500, adjust_bounds=True)
elif key=="LCLOCK":
core_lock = validate_and_convert(value, 0, 12000, adjust_bounds=True)
elif key=="LMEM":
mem_lock = validate_and_convert(value, 0, 32000, adjust_bounds=True)
if core_offset or mem_offset or pl_static: if len(fs_mem_lock)>0:
mem_lock = fs_mem_lock
if len(fs_core_lock)>0:
core_lock = fs_core_lock
#print(mem_lock, core_lock)
if core_offset or mem_offset or pl_static or mem_lock or core_lock:
for gpu_idx, default_pl in enumerate(default_power_limits): for gpu_idx, default_pl in enumerate(default_power_limits):
out_oc_config[str(gpu_idx)] = { out_oc_config[str(gpu_idx)] = {
"core": get_number_or_last(core_offset, gpu_idx) if core_offset else 0, "core": get_number_or_last(core_offset, gpu_idx) if core_offset else 0,
"mem": get_number_or_last(mem_offset, gpu_idx) if mem_offset else 0, "mem": get_number_or_last(mem_offset, gpu_idx) if mem_offset else 0,
"pl": get_number_or_last(pl_static, gpu_idx) if pl_static else default_pl "pl": get_number_or_last(pl_static, gpu_idx) if pl_static else default_pl
} }
if type(core_lock)==list and len(core_lock)>0:
out_oc_config[str(gpu_idx)]["core_lock"] = get_number_or_last(core_lock, gpu_idx)
if type(mem_lock)==list and len(mem_lock)>0:
out_oc_config[str(gpu_idx)]["mem_lock"] = get_number_or_last(mem_lock, gpu_idx)
except Exception as oc_info_e: except Exception as oc_info_e:
pass pass