Update repositories & migrate nvidia-pstate to other repo

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

@@ -4,25 +4,23 @@
  import subprocess
  from functools import lru_cache
  import installer
-+from pstates import enter_pstate_high, enter_pstate_low
++from nvidia_pstate import set_pstate_high, set_pstate_low
 
 
- commandline_args = os.environ.get('COMMANDLINE_ARGS', "")
-@@ -175,6 +176,7 @@ def start_server(immediate=True, server=None):
+ debug_install = installer.log.debug if os.environ.get('SD_INSTALL_DEBUG', None) is not None else lambda *args, **kwargs: None
+@@ -260,4 +261,5 @@ def main():
 
 
  if __name__ == "__main__":
-+    enter_pstate_low()
-     installer.ensure_base_requirements()
-     init_args() # setup argparser and default folders
-     installer.args = args
++    set_pstate_low()
+     main()
 --- 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 pstates import enter_pstate_high, enter_pstate_low
++from nvidia_pstate import set_pstate_high, set_pstate_low
 
 
  class State:
@@ -31,7 +29,7 @@
          if self.debug_output:
              log.debug(f'State begin: {self.job}')
 +        if self.job != "load":
-+            enter_pstate_high()
++            set_pstate_high()
          modules.devices.torch_gc()
 
      def end(self, api=None):
@@ -40,7 +38,7 @@
          if self.debug_output:
              log.debug(f'State end: {self.job} time={time.time() - self.time_start:.2f}')
 +        if self.job != "load":
-+            enter_pstate_low()
++            set_pstate_low()
          self.job = ""
          self.job_count = 0
          self.job_no = 0

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

@@ -4,31 +4,31 @@
  import argparse
  import json, sys, http.server, time, asyncio, socket, threading
  from concurrent.futures import ThreadPoolExecutor
-+from pstates import enter_pstate_high, enter_pstate_low
++from nvidia_pstate import set_pstate_high, set_pstate_low
 
  sampler_order_max = 7
  stop_token_max = 16
-@@ -339,6 +340,7 @@ def load_model(model_filename):
+@@ -380,6 +381,7 @@ 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=1.0, 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, dynatemp_range=0.0, dynatemp_exponent=1.0, smoothing_factor=0.0, logit_biases={}):
-+    enter_pstate_high()
++    set_pstate_high()
      global maxctx, args, currentusergenkey, totalgens, pendingabortkey
      inputs = generation_inputs()
-     outputs = ctypes.create_unicode_buffer(ctypes.sizeof(generation_outputs))
-@@ -438,6 +440,7 @@ def generate(prompt, memory="", max_length=32, max_context_length=512, temperatu
+     inputs.prompt = prompt.encode("UTF-8")
+@@ -478,6 +480,7 @@ 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()
++        set_pstate_low()
          return outstr
 
- def utfprint(str):
-@@ -2337,6 +2340,7 @@ def sanitize_string(input_string):
+
+@@ -2545,6 +2548,7 @@ def sanitize_string(input_string):
      return sanitized_string
 
  def main(launch_args,start_server=True):
-+    enter_pstate_low()
-     global args, friendlymodelname
++    set_pstate_low()
+     global args, friendlymodelname, friendlysdmodelname, fullsdmodelpath
      args = launch_args
      embedded_kailite = None

airootfs/root/customize_airootfs/patches/0100-llamacpp-enable-prompt-cache.patch

@@ -1,17 +1,26 @@
 --- a/examples/server/oai.hpp
 +++ b/examples/server/oai.hpp
-@@ -35,7 +35,7 @@ inline static json oaicompat_completion_params_parse(
+@@ -33,7 +33,7 @@ inline static json oaicompat_completion_params_parse(
      llama_sampling_params default_sparams;
      llama_params["model"]             = json_value(body, "model", std::string("unknown"));
-     llama_params["prompt"]            = formatted_prompt;
+     llama_params["prompt"]            = format_chat(model, chat_template, body["messages"]);
 -    llama_params["cache_prompt"]      = json_value(body, "cache_prompt", false);
 +    llama_params["cache_prompt"]      = json_value(body, "cache_prompt", true);
      llama_params["temperature"]       = json_value(body, "temperature", 0.0);
      llama_params["top_k"]             = json_value(body, "top_k", default_sparams.top_k);
      llama_params["top_p"]             = json_value(body, "top_p", 1.0);
 --- a/examples/server/server.cpp
 +++ b/examples/server/server.cpp
-@@ -522,7 +522,7 @@ struct llama_server_context
+@@ -68,7 +68,7 @@ enum slot_command {
+
+ struct slot_params {
+     bool stream       = true;
+-    bool cache_prompt = false; // remember the prompt to avoid reprocessing all prompt
++    bool cache_prompt = true; // remember the prompt to avoid reprocessing all prompt
+
+     uint32_t seed      = -1; // RNG seed
+     int32_t  n_keep    =  0; // number of tokens to keep from initial prompt
+@@ -552,7 +552,7 @@ struct llama_server_context
          }
 
          slot->params.stream             = json_value(data, "stream",            false);
@@ -20,14 +29,3 @@
          slot->params.n_predict          = json_value(data, "n_predict",         default_params.n_predict);
          slot->sparams.top_k             = json_value(data, "top_k",             default_sparams.top_k);
          slot->sparams.top_p             = json_value(data, "top_p",             default_sparams.top_p);
---- a/examples/server/utils.hpp
-+++ b/examples/server/utils.hpp
-@@ -93,7 +93,7 @@ enum slot_command
- struct slot_params
- {
-     bool stream       = true;
--    bool cache_prompt = false; // remember the prompt to avoid reprocessing all prompt
-+    bool cache_prompt = true; // remember the prompt to avoid reprocessing all prompt
-
-     uint32_t seed      = -1; // RNG seed
-     int32_t  n_keep    =  0; // number of tokens to keep from initial prompt

airootfs/root/customize_airootfs/patches/0100-vllm-enable-other-archs.patch

@@ -6,7 +6,7 @@
  # Supported NVIDIA GPU architectures.
 -NVIDIA_SUPPORTED_ARCHS = {"7.0", "7.5", "8.0", "8.6", "8.9", "9.0"}
 +NVIDIA_SUPPORTED_ARCHS = {"6.0", "6.1", "7.0", "7.5", "8.0", "8.6", "8.9", "9.0"}
- ROCM_SUPPORTED_ARCHS = {"gfx90a", "gfx942", "gfx1100"}
+ ROCM_SUPPORTED_ARCHS = {"gfx908", "gfx90a", "gfx942", "gfx1100"}
  # SUPPORTED_ARCHS = NVIDIA_SUPPORTED_ARCHS.union(ROCM_SUPPORTED_ARCHS)
 
 @@ -222,9 +222,9 @@ if _is_cuda() and not compute_capabilities:

airootfs/root/customize_airootfs/scripts/0050-install-nvidia-pstate.sh.jinja2

@@ -0,0 +1,6 @@
+#!/bin/bash
+set -eu
+
+{% if CUDA %}
+pip3 install --break-system-packages nvidia-pstate
+{% endif %}