Fix patches

airootfs/root/customize_airootfs/patches/0000-automatic-drop-pstate-in-idle.patch

@@ -4,7 +4,7 @@
  import subprocess
  from functools import lru_cache
  import installer
-+from modules.pstates import enter_pstate_low
++from pstates import enter_pstate_high, enter_pstate_low
 
 
  commandline_args = os.environ.get('COMMANDLINE_ARGS', "")
@@ -16,36 +16,13 @@
      installer.ensure_base_requirements()
      init_args() # setup argparser and default folders
      installer.args = args
-diff --git a/modules/pstates.py b/modules/pstates.py
-new file mode 100644
-index 00000000..44865a87
---- /dev/null
-+++ b/modules/pstates.py
-@@ -0,0 +1,17 @@
-+import os
-+import subprocess
-+
-+def _set_pstate(pstate):
-+  visible_devices = os.getenv("CUDA_VISIBLE_DEVICES")
-+
-+  if visible_devices:
-+    for device in visible_devices.split(","):
-+      subprocess.check_call([ "nvidia-pstate", "-i", device, "-ps", pstate ], stdout=subprocess.DEVNULL)
-+  else:
-+    subprocess.check_call([ "nvidia-pstate", "-ps", pstate ], stdout=subprocess.DEVNULL)
-+
-+def enter_pstate_low():
-+  _set_pstate("8")
-+
-+def enter_pstate_high():
-+  _set_pstate("16")
 --- a/modules/shared_state.py
 +++ b/modules/shared_state.py
 @@ -3,6 +3,7 @@ import sys
  import time
  import datetime
  from modules.errors import log
-+from modules.pstates import enter_pstate_high, enter_pstate_low
++from pstates import enter_pstate_high, enter_pstate_low
 
 
  class State:

airootfs/root/customize_airootfs/patches/0000-koboldcpp-drop-pstate-in-idle.patch

@@ -8,47 +8,27 @@
 
  sampler_order_max = 7
  stop_token_max = 16
-@@ -301,6 +302,7 @@ def load_model(model_filename):
+@@ -302,6 +303,7 @@ def load_model(model_filename):
      return ret
 
- def generate(prompt, memory="", max_length=32, max_context_length=512, temperature=0.7, top_k=100, top_a=0.0, top_p=0.92, min_p=0.0, typical_p=1.0, tfs=1.0, rep_pen=1.1, rep_pen_range=128, mirostat=0, mirostat_tau=5.0, mirostat_eta=0.1, sampler_order=[6,0,1,3,4,2,5], seed=-1, stop_sequence=[], use_default_badwordsids=False, stream_sse=False, grammar='', grammar_retain_state=False, genkey='', trimstop=False, quiet=False):
+ def generate(prompt, memory="", max_length=32, max_context_length=512, temperature=0.7, top_k=100, top_a=0.0, top_p=0.92, min_p=0.0, typical_p=1.0, tfs=1.0, rep_pen=1.1, rep_pen_range=128, presence_penalty=0.0, mirostat=0, mirostat_tau=5.0, mirostat_eta=0.1, sampler_order=[6,0,1,3,4,2,5], seed=-1, stop_sequence=[], use_default_badwordsids=False, stream_sse=False, grammar='', grammar_retain_state=False, genkey='', trimstop=False, quiet=False):
 +    enter_pstate_high()
      global maxctx, args, currentusergenkey, totalgens
      inputs = generation_inputs()
      outputs = ctypes.create_unicode_buffer(ctypes.sizeof(generation_outputs))
-@@ -364,6 +366,7 @@ def generate(prompt, memory="", max_length=32, max_context_length=512, temperatu
+@@ -366,6 +368,7 @@ def generate(prompt, memory="", max_length=32, max_context_length=512, temperatu
              sindex = outstr.find(trim_str)
              if sindex != -1 and trim_str!="":
                  outstr = outstr[:sindex]
 +    enter_pstate_low()
      return outstr
 
  def utfprint(str):
-@@ -2031,6 +2034,7 @@ def sanitize_string(input_string):
+@@ -2176,6 +2179,7 @@ def sanitize_string(input_string):
      return sanitized_string
 
  def main(launch_args,start_server=True):
 +    enter_pstate_low()
      global args, friendlymodelname
      args = launch_args
      embedded_kailite = None
---- /dev/null
-+++ b/pstates.py
-@@ -0,0 +1,17 @@
-+import os
-+import subprocess
-+
-+def _set_pstate(pstate):
-+  visible_devices = os.getenv("CUDA_VISIBLE_DEVICES")
-+
-+  if visible_devices:
-+    for device in visible_devices.split(","):
-+      subprocess.check_call([ "nvidia-pstate", "-i", device, "-ps", pstate ], stdout=subprocess.DEVNULL)
-+  else:
-+    subprocess.check_call([ "nvidia-pstate", "-ps", pstate ], stdout=subprocess.DEVNULL)
-+
-+def enter_pstate_low():
-+  _set_pstate("8")
-+
-+def enter_pstate_high():
-+  _set_pstate("16")

airootfs/usr/lib/python3.11/pstates.py

@@ -0,0 +1,17 @@
+import os
+import subprocess
+
+def _set_pstate(pstate):
+  visible_devices = os.getenv("CUDA_VISIBLE_DEVICES")
+
+  if visible_devices:
+    for device in visible_devices.split(","):
+      subprocess.check_call([ "nvidia-pstate", "-i", device, "-ps", pstate ], stdout=subprocess.DEVNULL)
+  else:
+    subprocess.check_call([ "nvidia-pstate", "-ps", pstate ], stdout=subprocess.DEVNULL)
+
+def enter_pstate_low():
+  _set_pstate("8")
+
+def enter_pstate_high():
+  _set_pstate("16")