Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-06 13:01:21 +02:00
Code Issues Packages Projects Releases Wiki Activity

reordered some args

Browse Source
This commit is contained in:
Concedo
2026-06-05 17:27:17 +08:00
parent 9755b05556
commit a9d743634c
1 changed files with 100 additions and 99 deletions
Download Patch File Download Diff File Expand all files Collapse all files
koboldcpp.py
+100 -99
View File
@@ -11696,111 +11696,112 @@ if __name__ == '__main__':
return range_checker
parser = argparse.ArgumentParser(description=f'KoboldCpp Server - Version {KcppVersion}')
modelgroup = parser.add_mutually_exclusive_group() #we want to be backwards compatible with the unnamed positional args
modelgroup.add_argument("--model","-m", metavar=('[filenames]'), help="Model file to load. Accepts multiple values if they are URLs.", type=str, nargs='+', default=[])
modelgroup.add_argument("model_param", help="Model file to load (positional)", nargs="?")
portgroup = parser.add_mutually_exclusive_group() #we want to be backwards compatible with the unnamed positional args
portgroup.add_argument("--port", metavar=('[portnumber]'), help=f"Port to listen on. (Defaults to {defaultport})", default=defaultport, type=int, action='store')
portgroup.add_argument("port_param", help="Port to listen on (positional)", default=defaultport, nargs="?", type=int, action='store')
parser.add_argument("--host", metavar=('[ipaddr]'), help="Host IP to listen on. If this flag is not set, all routable interfaces are accepted.", default="")
parser.add_argument("--launch", help="Launches a web browser when load is completed.", action='store_true')
parser.add_argument("--config", metavar=('[filename]'), help="Load settings from a .kcpps file. Other arguments will be ignored", type=str, nargs=1)
parser.add_argument("--threads","-t", metavar=('[threads]'), help="Use a custom number of threads if specified. Otherwise, uses an amount based on CPU cores", type=int, default=get_default_threads())
compatgroup = parser.add_mutually_exclusive_group()
compatgroup.add_argument("--usecuda", "--usecublas", "--usehipblas", help="Use CUDA for GPU Acceleration. Requires CUDA. Enter a number afterwards to select and use 1 GPU. Leaving no number will use all GPUs.", nargs='*',metavar=('[main GPU ID]'), choices=['0','1','2','3','all', 'mmq','nommq','normal','lowvram','rowsplit'])
compatgroup.add_argument("--usevulkan", help="Use Vulkan for GPU Acceleration. Can optionally specify one or more GPU Device ID (e.g. --usevulkan 0), leave blank to autodetect.", metavar=('[Device IDs]'), nargs='*', type=int, default=None)
compatgroup.add_argument("--usecpu", help="Do not use any GPU acceleration (CPU Only)", action='store_true')
modelgroup = parser.add_mutually_exclusive_group() #we want to be backwards compatible with the unnamed positional args
modelgroup.add_argument("--model","-m", metavar=('[filenames]'), help="Model file to load. Accepts multiple values if they are URLs.", type=str, nargs='+', default=[])
modelgroup.add_argument("model_param", help="Model file to load (positional)", nargs="?")
parser.add_argument("--config", metavar=('[filename]'), help="Load settings from a .kcpps file. Other arguments will be ignored", type=str, nargs=1)
parser.add_argument("--contextsize","--ctx-size", "-c", help="Controls the memory allocated for maximum context size, only change if you need more RAM for big contexts. (default 8192).",metavar=('[256 to 262144]'), type=check_range(int,256,262144), default=8192)
parser.add_argument("--gpulayers","--gpu-layers","--n-gpu-layers","-ngl", help="Set number of layers to offload to GPU when using GPU. Requires GPU. Set to -1 to try autodetect, set to 0 to disable GPU offload.",metavar=('[GPU layers]'), nargs='?', const=1, type=int, default=-1)
parser.add_argument("--gpulayers","--gpu-layers","--n-gpu-layers","-ngl", help="Set number of layers to offload to GPU (when using GPU). Set to -1 to enable autofit (default), set to 0 to disable GPU offload.",metavar=('[GPU layers]'), nargs='?', const=1, type=int, default=-1)
parser.add_argument("--host", metavar=('[ipaddr]'), help="Host IP to listen on. If this flag is not set, all routable interfaces are accepted.", default="")
parser.add_argument("--launch", help="Launches a web browser when load is completed.", action='store_true')
parser.add_argument("--tensor_split","--tensorsplit","--tensor-split","-ts", help="For CUDA and Vulkan only, ratio to split tensors across multiple GPUs, space-separated list of proportions, e.g. 7 3", metavar=('[Ratios]'), type=float, nargs='+')
parser.add_argument("--autofit","--fit","-fit", help="Automatically attempt to fit the model in the best possible way. Overrides everything else. Experimental.", action='store_true')
parser.add_argument("--threads","-t", metavar=('[threads]'), help="Use a custom number of threads if specified. Otherwise, uses an amount based on CPU cores", type=int, default=get_default_threads())
portgroup = parser.add_mutually_exclusive_group() #we want to be backwards compatible with the unnamed positional args
portgroup.add_argument("--port", metavar=('[portnumber]'), help=f"Port to listen on. (Defaults to {defaultport})", default=defaultport, type=int, action='store')
portgroup.add_argument("port_param", help="Port to listen on (positional)", default=defaultport, nargs="?", type=int, action='store')
parser.add_argument("--version", help="Prints version and exits.", action='store_true')
#more advanced params
advparser = parser.add_argument_group('Advanced Commands')
advparser.add_argument("--version", help="Prints version and exits.", action='store_true')
advparser.add_argument("--analyze", metavar=('[filename]'), help="Reads the metadata, weight types and tensor names in any GGUF file.", default="")
advparser.add_argument("--maingpu","--main-gpu","-mg", help="Only used in a multi-gpu setup. Sets the index of the main GPU that will be used.",metavar=('[Device ID]'), type=int, default=-1)
advparser.add_argument("--autofit","--fit","-fit", help="Forces autofit, which attempts to fit the model in the best possible way. Overrides everything else.", action='store_true')
advparser.add_argument("--autofitpadding", metavar=('[padding in MB]'), help="How much spare allowance in MB should autofit reserve? If it's too little, the load might fail.", type=int, default=default_autofit_padding)
advparser.add_argument("--batchsize","--blasbatchsize","--batch-size","-b", help="Sets the batch size used in batched processing (default 512). Setting it to -1 disables batched mode, but keeps other benefits like GPU offload.", type=int,choices=[-1,16,32,64,128,256,512,1024,2048,4096], default=512)
advparser.add_argument("--parallelrequests","--continuous-batching","--contbatch", help="Allows multiple requests to be batched and executed in parallel. Only works for basic text generation requests (Experimental, No media)", metavar=('[slots]'), type=check_range(int,0,32), default=1)
advparser.add_argument("--blasthreads","--batchthreads","--threadsbatch","--threads-batch", help="Use a different number of threads during batching if specified. Otherwise, has the same value as --threads",metavar=('[threads]'), type=int, default=0)
advparser.add_argument("--splitmode","-sm","--split-mode", help="How to split the model across multiple GPUs", metavar=('[split mode]'), type=str, choices=splitmode_choices, default=splitmode_choices[0])
advparser.add_argument("--nommq", help="Disables MMQ, only used for cuda backend. This flag may be removed in future.", action='store_true')
advparser.add_argument("--lora", help="GGUF models only, applies a lora file on top of model.", metavar=('[lora_filename]'), nargs='+')
advparser.add_argument("--loramult", metavar=('[amount]'), help="Multiplier for the Text LORA model to be applied.", type=float, default=1.0)
advparser.add_argument("--noshift","--no-context-shift", help="If set, do not attempt to Trim and Shift the GGUF context.", action='store_true')
advparser.add_argument("--nofastforward", help="If set, do not attempt to fast forward GGUF context (always reprocess). Will also enable noshift", action='store_true')
advparser.add_argument("--noswa","--swa-full", help="If set, uses full-size SWA KV Cache. Otherwise, SWA will be enabled automatically on models that support it. SWA saves memory but cannot be used with context shifting.", action='store_true')
advparser.add_argument("--swapadding", help="How much extra to pad the SWA KV cache, this affects the rewind limit before reprocessing is forced.", type=int, default=swa_padding_default)
advparser.add_argument("--smartcache", help="Enables intelligent context switching by saving KV cache snapshots to RAM. Requires fast forwarding.", metavar=('limit'), nargs='?', const=1, type=int, default=0)
advparser.add_argument("--ropeconfig", help="If set, uses customized RoPE scaling from configured frequency scale and frequency base (e.g. --ropeconfig 0.25 10000). Otherwise, uses NTK-Aware scaling set automatically based on context size. For linear rope, simply set the freq-scale and ignore the freq-base",metavar=('[rope-freq-scale]', '[rope-freq-base]'), default=[0.0, 10000.0], type=float, nargs='+')
advparser.add_argument("--overridenativecontext", help="Overrides the native trained context of the loaded model with a custom value to be used for Rope scaling.",metavar=('[trained context]'), type=int, default=0)
compatgroup3 = advparser.add_mutually_exclusive_group()
compatgroup3.add_argument("--usemmap", help="If set, uses mmap to load model.", action='store_true')
advparser.add_argument("--usemlock","--mlock", help="Enables mlock, preventing the RAM used to load the model from being paged out. Not usually recommended.", action='store_true')
advparser.add_argument("--noavx2", help="Do not use AVX2 instructions, a slower compatibility mode for older devices.", action='store_true')
advparser.add_argument("--failsafe", help="Use failsafe mode, extremely old CPU compatibility mode that should work on all devices.", action='store_true')
advparser.add_argument("--debugmode", help="Shows additional debug info in the terminal. Levels: -1 (Horde-quiet, suppresses non-essential prints; auto-applied when Horde args are set), 0 (default, normal output), 1 (verbose: extra slot/cache info, larger print buffers, retains horde-debug prefix). Passing the flag without a value implies 1.", nargs='?', const=1, type=int, default=0)
advparser.add_argument("--onready", help="An optional shell command to execute after the model has been loaded.", metavar=('[shell command]'), type=str, default="")
advparser.add_argument("--benchmark", help="Do not start server, instead run benchmarks. If filename is provided, appends results to provided file.", metavar=('[filename]'), nargs='?', const="stdout", type=str, default=None)
advparser.add_argument("--prompt","-p", metavar=('[prompt]'), help="Passing a prompt string triggers a direct inference, loading the model, outputs the response to stdout and exits. Can be used alone or with benchmark.", type=str, default="")
advparser.add_argument("--blasthreads","--batchthreads","--threadsbatch","--threads-batch", help="Use a different number of threads during batching if specified. Otherwise, has the same value as --threads",metavar=('[threads]'), type=int, default=0)
advparser.add_argument("--chatcompletionsadapter", metavar=('[filename]'), help="Select an optional ChatCompletions Adapter JSON file to force custom instruct tags.", default="AutoGuess")
advparser.add_argument("--cli", help="Does not launch KoboldCpp HTTP server. Instead, enables KoboldCpp from the command line, accepting interactive console input and displaying responses to the terminal.", action='store_true')
advparser.add_argument("--genlimit","--promptlimit", help="Sets the maximum number of generated tokens, it will restrict all generations to this or lower. Also usable with --prompt or --benchmark.",metavar=('[token limit]'), type=int, default=0)
advparser.add_argument("--multiuser", help="Set maximum number of queued incoming requests allowed.", metavar=('limit'), type=int, nargs='?', const=multiuser_concurrent_limit, default=multiuser_concurrent_limit)
advparser.add_argument("--multiplayer", help="Hosts a shared multiplayer session that others can join.", action='store_true')
advparser.add_argument("--websearch", help="Enable the local search engine proxy so Web Searches can be done.", action='store_true')
advparser.add_argument("--remotetunnel", help="Uses Cloudflare to create a remote tunnel, allowing you to access koboldcpp remotely over the internet even behind a firewall.", action='store_true')
advparser.add_argument("--highpriority", help="Experimental flag. If set, increases the process CPU priority, potentially speeding up generation. Use caution.", action='store_true')
advparser.add_argument("--foreground", help="Windows only. Sends the terminal to the foreground every time a new prompt is generated. This helps avoid some idle slowdown issues.", action='store_true')
advparser.add_argument("--preloadstory", metavar=('[savefile]'), help="Configures a prepared story json save file to be hosted on the server, which frontends (such as KoboldAI Lite) can access over the API.", default="")
advparser.add_argument("--savedatafile", metavar=('[savefile]'), help="If enabled, creates or opens a persistent database file on the server, that allows users to save and load their data remotely. A new file is created if it does not exist.", default="")
advparser.add_argument("--quiet", help="Enable quiet mode, which hides generation inputs and outputs in the terminal. Quiet mode is automatically enabled when running a horde worker.", action='store_true')
advparser.add_argument("--ssl", help="Allows all content to be served over SSL instead. A valid UNENCRYPTED SSL cert and key .pem files must be provided", metavar=('[cert_pem]', '[key_pem]'), nargs='+')
advparser.add_argument("--nocertify", help="Allows insecure SSL connections. Use this if you have cert errors and need to bypass certificate restrictions.", action='store_true')
advparser.add_argument("--mmproj", metavar=('[filename]'), help="Select a multimodal projector file for vision models like LLaVA.", default="")
advparser.add_argument("--mmprojcpu","--no-mmproj-offload", help="Force CLIP for Vision mmproj always on CPU.", action='store_true')
advparser.add_argument("--visionmaxres", metavar=('[max px]'), help="Clamp MMProj vision maximum allowed resolution. Allowed values are between 512 to 2048 px (default 1024).", type=int, default=default_visionmaxres)
advparser.add_argument("--visionmintokens","--image-min-tokens", metavar=('[tokens]'), help="Override the minimum tokens for the MMProj embedding (default -1).", type=int, default=-1)
advparser.add_argument("--visionmaxtokens","--image-max-tokens", metavar=('[tokens]'), help="Override the maximum tokens for the MMProj embedding (default -1).", type=int, default=-1)
advparser.add_argument("--draftmodel","--model-draft","-md", metavar=('[filename]'), help="Load a small draft model for speculative decoding. It will be fully offloaded. Vocab must match the main model.", default="")
advparser.add_argument("--debugmode", help="Shows additional debug info in the terminal. Levels: -1 (Horde-quiet, suppresses non-essential prints; auto-applied when Horde args are set), 0 (default, normal output), 1 (verbose: extra slot/cache info, larger print buffers, retains horde-debug prefix). Passing the flag without a value implies 1.", nargs='?', const=1, type=int, default=0)
advparser.add_argument("--defaultgenamt", help="How many tokens to generate by default, if not specified. Must be smaller than context size. Usually, your frontend GUI will override this.", type=check_range(int,64,8192), default=default_genlen)
advparser.add_argument("--device", "-dev", metavar=('<dev1,dev2,..>'), help="Set llama.cpp compatible device selection override. Comma separated. Overrides normal device choices.", default="")
advparser.add_argument("--downloaddir", metavar=('[directory]'), help="Specify a directory that models will be downloaded to or searched from, if unset uses the working directory.", default="")
advparser.add_argument("--draftamount","--draft-max","--draft-n", metavar=('[tokens]'), help="How many tokens to draft per chunk before verifying results", type=int, default=default_draft_amount)
advparser.add_argument("--draftgpulayers","--gpu-layers-draft","--n-gpu-layers-draft","-ngld", metavar=('[layers]'), help="How many layers to offload to GPU for the draft model (default=full offload)", type=int, default=999)
advparser.add_argument("--draftgpusplit", help="GPU layer distribution ratio for draft model (default=same as main). Only works if multi-GPUs selected for MAIN model and tensor_split is set!", metavar=('[Ratios]'), type=float, nargs='+')
advparser.add_argument("--password", metavar=('[API key]'), help="Enter a password required to use this instance. This key will be required for all text endpoints. Image endpoints are not secured. Can also be set with env var KCPP_PASSWORD", default=os.getenv('KCPP_PASSWORD',None))
advparser.add_argument("--ratelimit", metavar=('[seconds]'), help="If enabled, rate limit generative request by IP address. Each IP can only send a new request once per X seconds.", type=int, default=0)
advparser.add_argument("--ignoremissing", help="Ignores all missing non-essential files, just skipping them instead.", action='store_true')
advparser.add_argument("--chatcompletionsadapter", metavar=('[filename]'), help="Select an optional ChatCompletions Adapter JSON file to force custom instruct tags.", default="AutoGuess")
advparser.add_argument("--jinja", help="Enables using jinja chat template formatting for chat completions endpoint. Other endpoints are unaffected. Tool calls are done without jinja.", action='store_true')
advparser.add_argument("--jinja_tools","--jinja-tools","--jinjatools", help="Enables using jinja chat template formatting for chat completions endpoint. Other endpoints are unaffected. Tool calls are done with jinja.", action='store_true')
advparser.add_argument("--jinja_kwargs","--jinja-kwargs","--jinjakwargs","--chat-template-kwargs", metavar=('{"parameter":"value",...}'), help="Set additional fields for Jinja JSON template parser, must be a valid JSON object.", default="")
advparser.add_argument("--jinjatemplate","--chat-template-file", metavar=('[filename]'), help="Select a custom Jinja chat template, will overwrite model jinja chat template", default="")
advparser.add_argument("--jinjathink", help="A quick way to enable or disable thinking in the jinja template.", type=str, choices=['default','true','false'], default="default")
advparser.add_argument("--noflashattention","--no-flash-attn","-nofa", help="Disables flash attention.", action='store_true')
advparser.add_argument("--lowvram","-nkvo","--no-kv-offload", help="If supported by the backend, do not offload KV to GPU (lowvram mode). Not recommended, will be slow.", action='store_true')
advparser.add_argument("--quantkv", help="Sets the KV cache data type quantization, options are f16/bf16/q8_0/q5_1/q4_0. Requires Flash Attention for full effect, otherwise only K cache is quantized.",metavar=('[quantization level f16/bf16/q8_0/q5_1/q4_0]'), type=str, choices=["f16","bf16","q8_0","q5_1","q4_0","0","1","2","3"], default="f16")
advparser.add_argument("--smartcontext", help="Reserving a portion of context to try processing less frequently. Outdated. Not recommended.", action='store_true')
advparser.add_argument("--unpack", help="Extracts the file contents of the KoboldCpp binary into a target directory.", metavar=('destination'), type=str, default="")
advparser.add_argument("--draftmodel","--model-draft","-md", metavar=('[filename]'), help="Load a small draft model for speculative decoding. It will be fully offloaded. Vocab must match the main model.", default="")
advparser.add_argument("--enableguidance", help="Enables the use of Classifier-Free-Guidance, which allows the use of negative prompts. Has performance and memory impact.", action='store_true')
advparser.add_argument("--exportconfig", help="Exports the current selected arguments as a .kcpps settings file", metavar=('[filename]'), type=str, default="")
advparser.add_argument("--exporttemplate", help="Exports the current selected arguments as a .kcppt template file", metavar=('[filename]'), type=str, default="")
advparser.add_argument("--nomodel", help="Allows you to launch the GUI alone, without selecting any model.", action='store_true')
advparser.add_argument("--moeexperts", metavar=('[num of experts]'), help="How many experts to use for MoE models (default=follow gguf)", type=int, default=-1)
advparser.add_argument("--moecpu","--n-cpu-moe", "-ncmoe", metavar=('[layers affected]'), help="Keep the Mixture of Experts (MoE) weights of the first N layers in the CPU. If no value is provided, applies to all layers.", nargs='?', const=999, type=int, default=0)
advparser.add_argument("--defaultgenamt", help="How many tokens to generate by default, if not specified. Must be smaller than context size. Usually, your frontend GUI will override this.", type=check_range(int,64,8192), default=default_genlen)
advparser.add_argument("--nobostoken", help="Prevents BOS token from being added at the start of any prompt. Usually NOT recommended for most models.", action='store_true')
advparser.add_argument("--enableguidance", help="Enables the use of Classifier-Free-Guidance, which allows the use of negative prompts. Has performance and memory impact.", action='store_true')
advparser.add_argument("--maxrequestsize", metavar=('[size in MB]'), help="Specify a max request payload size. Any requests to the server larger than this size will be dropped. Do not change if unsure.", type=int, default=32)
advparser.add_argument("--overridekv","--override-kv", metavar=('[name=type:value]'), help="Override metadata value by key. Separate multiple values with commas. Format is name=type:value. Types: int, float, bool, str", default="")
advparser.add_argument("--overridetensors","--override-tensor","-ot", metavar=('[tensor name pattern=buffer type]'), help="Override selected backend for specific tensors matching tensor_name_regex_pattern=buffer_type, same as in llama.cpp.", default="")
compatgroup2 = parser.add_mutually_exclusive_group()
compatgroup2.add_argument("--showgui", help="Always show the GUI instead of launching the model right away when loading settings from a .kcpps file.", action='store_true')
compatgroup2.add_argument("--skiplauncher", help="Doesn't display or use the GUI launcher. Overrides showgui.", action='store_true')
advparser.add_argument("--singleinstance", help="Allows this KoboldCpp instance to be shut down by any new instance requesting the same port, preventing duplicate servers from clashing on a port.", action='store_true')
advparser.add_argument("--nopipelineparallel", help="Disable Pipeline Parallelism. Pipeline Parallelism provides faster multigpu speeds but using more memory, only active for multigpu.", action='store_true')
advparser.add_argument("--failsafe", help="Use failsafe mode, extremely old CPU compatibility mode that should work on all devices.", action='store_true')
advparser.add_argument("--foreground", help="Windows only. Sends the terminal to the foreground every time a new prompt is generated. This helps avoid some idle slowdown issues.", action='store_true')
advparser.add_argument("--gendefaults", metavar=('{"parameter":"value",...}'), help="Sets extra default parameters for some fields in API requests, as a JSON string.", default="")
advparser.add_argument("--gendefaultsoverwrite", help="Allow the gendefaults parameters to overwrite the original value in API payloads.", action='store_true')
advparser.add_argument("--genlimit","--promptlimit", help="Sets the maximum number of generated tokens, it will restrict all generations to this or lower. Also usable with --prompt or --benchmark.",metavar=('[token limit]'), type=int, default=0)
advparser.add_argument("--highpriority", help="Experimental flag. If set, increases the process CPU priority, potentially speeding up generation. Use caution.", action='store_true')
advparser.add_argument("--ignoremissing", help="Ignores all missing non-essential files, just skipping them instead.", action='store_true')
advparser.add_argument("--jinja", help="Enables using jinja chat template formatting for chat completions endpoint. Other endpoints are unaffected. Tool calls are done without jinja.", action='store_true')
advparser.add_argument("--jinja_kwargs","--jinja-kwargs","--jinjakwargs","--chat-template-kwargs", metavar=('{"parameter":"value",...}'), help="Set additional fields for Jinja JSON template parser, must be a valid JSON object.", default="")
advparser.add_argument("--jinja_tools","--jinja-tools","--jinjatools", help="Enables using jinja chat template formatting for chat completions endpoint. Other endpoints are unaffected. Tool calls are done with jinja.", action='store_true')
advparser.add_argument("--jinjatemplate","--chat-template-file", metavar=('[filename]'), help="Select a custom Jinja chat template, will overwrite model jinja chat template", default="")
advparser.add_argument("--jinjathink", help="A quick way to enable or disable thinking in the jinja template.", type=str, choices=['default','true','false'], default="default")
advparser.add_argument("--lora", help="GGUF models only, applies a lora file on top of model.", metavar=('[lora_filename]'), nargs='+')
advparser.add_argument("--loramult", metavar=('[amount]'), help="Multiplier for the Text LORA model to be applied.", type=float, default=1.0)
advparser.add_argument("--lowvram","-nkvo","--no-kv-offload", help="If supported by the backend, do not offload KV to GPU (lowvram mode). Not recommended, will be slow.", action='store_true')
advparser.add_argument("--maingpu","--main-gpu","-mg", help="Only used in a multi-gpu setup. Sets the index of the main GPU that will be used.",metavar=('[Device ID]'), type=int, default=-1)
advparser.add_argument("--maxrequestsize", metavar=('[size in MB]'), help="Specify a max request payload size. Any requests to the server larger than this size will be dropped. Do not change if unsure.", type=int, default=32)
advparser.add_argument("--mcpfile", metavar=('[mcp json file]'), help="Specify path to mcp.json which contains the Cladue Desktop compatible MCP server config.", default="")
advparser.add_argument("--device", "-dev", metavar=('<dev1,dev2,..>'), help="Set llama.cpp compatible device selection override. Comma separated. Overrides normal device choices.", default="")
advparser.add_argument("--downloaddir", metavar=('[directory]'), help="Specify a directory that models will be downloaded to or searched from, if unset uses the working directory.", default="")
advparser.add_argument("--autofitpadding", metavar=('[padding in MB]'), help="How much spare allowance in MB should autofit reserve? If it's too little, the load might fail.", type=int, default=default_autofit_padding)
advparser.add_argument("--mmproj", metavar=('[filename]'), help="Select a multimodal projector file for vision models like LLaVA.", default="")
advparser.add_argument("--mmprojcpu","--no-mmproj-offload", help="Force CLIP for Vision mmproj always on CPU.", action='store_true')
advparser.add_argument("--moecpu","--n-cpu-moe", "-ncmoe", metavar=('[layers affected]'), help="Keep the Mixture of Experts (MoE) weights of the first N layers in the CPU. If no value is provided, applies to all layers.", nargs='?', const=999, type=int, default=0)
advparser.add_argument("--moeexperts", metavar=('[num of experts]'), help="How many experts to use for MoE models (default=follow gguf)", type=int, default=-1)
advparser.add_argument("--multiuser", help="Set maximum number of queued incoming requests allowed.", metavar=('limit'), type=int, nargs='?', const=multiuser_concurrent_limit, default=multiuser_concurrent_limit)
advparser.add_argument("--multiplayer", help="Hosts a shared multiplayer session that others can join.", action='store_true')
advparser.add_argument("--nocertify", help="Allows insecure SSL connections. Use this if you have cert errors and need to bypass certificate restrictions.", action='store_true')
advparser.add_argument("--nofastforward", help="If set, do not attempt to fast forward GGUF context (always reprocess). Will also enable noshift", action='store_true')
advparser.add_argument("--noflashattention","--no-flash-attn","-nofa", help="Disables flash attention.", action='store_true')
advparser.add_argument("--noavx2", help="Do not use AVX2 instructions, a slower compatibility mode for older devices.", action='store_true')
advparser.add_argument("--nobostoken", help="Prevents BOS token from being added at the start of any prompt. Usually NOT recommended for most models.", action='store_true')
advparser.add_argument("--nommq", help="Disables MMQ, only used for cuda backend. This flag may be removed in future.", action='store_true')
advparser.add_argument("--nomodel", help="Allows you to launch the GUI alone, without selecting any model.", action='store_true')
advparser.add_argument("--nopipelineparallel", help="Disable Pipeline Parallelism. Pipeline Parallelism provides faster multigpu speeds but using more memory, only active for multigpu.", action='store_true')
advparser.add_argument("--noshift","--no-context-shift", help="If set, do not attempt to Trim and Shift the GGUF context.", action='store_true')
advparser.add_argument("--noswa","--swa-full", help="If set, uses full-size SWA KV Cache. Otherwise, SWA will be enabled automatically on models that support it. SWA saves memory but cannot be used with context shifting.", action='store_true')
advparser.add_argument("--onready", help="An optional shell command to execute after the model has been loaded.", metavar=('[shell command]'), type=str, default="")
advparser.add_argument("--overridekv","--override-kv", metavar=('[name=type:value]'), help="Override metadata value by key. Separate multiple values with commas. Format is name=type:value. Types: int, float, bool, str", default="")
advparser.add_argument("--overridenativecontext", help="Overrides the native trained context of the loaded model with a custom value to be used for Rope scaling.",metavar=('[trained context]'), type=int, default=0)
advparser.add_argument("--overridetensors","--override-tensor","-ot", metavar=('[tensor name pattern=buffer type]'), help="Override selected backend for specific tensors matching tensor_name_regex_pattern=buffer_type, same as in llama.cpp.", default="")
advparser.add_argument("--parallelrequests","--continuous-batching","--contbatch", help="Allows multiple requests to be batched and executed in parallel. Only works for basic text generation requests (Experimental, No media)", metavar=('[slots]'), type=check_range(int,0,32), default=1)
advparser.add_argument("--password", metavar=('[API key]'), help="Enter a password required to use this instance. This key will be required for all text endpoints. Image endpoints are not secured. Can also be set with env var KCPP_PASSWORD", default=os.getenv('KCPP_PASSWORD',None))
advparser.add_argument("--preloadstory", metavar=('[savefile]'), help="Configures a prepared story json save file to be hosted on the server, which frontends (such as KoboldAI Lite) can access over the API.", default="")
advparser.add_argument("--prompt","-p", metavar=('[prompt]'), help="Passing a prompt string triggers a direct inference, loading the model, outputs the response to stdout and exits. Can be used alone or with benchmark.", type=str, default="")
advparser.add_argument("--quantkv", help="Sets the KV cache data type quantization, options are f16/bf16/q8_0/q5_1/q4_0. Requires Flash Attention for full effect, otherwise only K cache is quantized.",metavar=('[quantization level f16/bf16/q8_0/q5_1/q4_0]'), type=str, choices=["f16","bf16","q8_0","q5_1","q4_0","0","1","2","3"], default="f16")
advparser.add_argument("--quiet", help="Enable quiet mode, which hides generation inputs and outputs in the terminal. Quiet mode is automatically enabled when running a horde worker.", action='store_true')
advparser.add_argument("--ratelimit", metavar=('[seconds]'), help="If enabled, rate limit generative request by IP address. Each IP can only send a new request once per X seconds.", type=int, default=0)
advparser.add_argument("--remotetunnel", help="Uses Cloudflare to create a remote tunnel, allowing you to access koboldcpp remotely over the internet even behind a firewall.", action='store_true')
advparser.add_argument("--ropeconfig", help="If set, uses customized RoPE scaling from configured frequency scale and frequency base (e.g. --ropeconfig 0.25 10000). Otherwise, uses NTK-Aware scaling set automatically based on context size. For linear rope, simply set the freq-scale and ignore the freq-base",metavar=('[rope-freq-scale]', '[rope-freq-base]'), default=[0.0, 10000.0], type=float, nargs='+')
advparser.add_argument("--savedatafile", metavar=('[savefile]'), help="If enabled, creates or opens a persistent database file on the server, that allows users to save and load their data remotely. A new file is created if it does not exist.", default="")
advparser.add_argument("--singleinstance", help="Allows this KoboldCpp instance to be shut down by any new instance requesting the same port, preventing duplicate servers from clashing on a port.", action='store_true')
advparser.add_argument("--smartcache", help="Enables intelligent context switching by saving KV cache snapshots to RAM. Requires fast forwarding.", metavar=('limit'), nargs='?', const=1, type=int, default=0)
advparser.add_argument("--smartcontext", help="Reserving a portion of context to try processing less frequently. Outdated. Not recommended.", action='store_true')
advparser.add_argument("--splitmode","-sm","--split-mode", help="How to split the model across multiple GPUs", metavar=('[split mode]'), type=str, choices=splitmode_choices, default=splitmode_choices[0])
advparser.add_argument("--ssl", help="Allows all content to be served over SSL instead. A valid UNENCRYPTED SSL cert and key .pem files must be provided", metavar=('[cert_pem]', '[key_pem]'), nargs='+')
advparser.add_argument("--swapadding", help="How much extra to pad the SWA KV cache, this affects the rewind limit before reprocessing is forced.", type=int, default=swa_padding_default)
advparser.add_argument("--unpack", help="Extracts the file contents of the KoboldCpp binary into a target directory.", metavar=('destination'), type=str, default="")
advparser.add_argument("--usemlock","--mlock", help="Enables mlock, preventing the RAM used to load the model from being paged out. Not usually recommended.", action='store_true')
compatgroup3 = advparser.add_mutually_exclusive_group()
compatgroup3.add_argument("--usemmap", help="If set, uses mmap to load model.", action='store_true')
advparser.add_argument("--visionmaxres", metavar=('[max px]'), help="Clamp MMProj vision maximum allowed resolution. Allowed values are between 512 to 2048 px (default 1024).", type=int, default=default_visionmaxres)
advparser.add_argument("--visionmaxtokens","--image-max-tokens", metavar=('[tokens]'), help="Override the maximum tokens for the MMProj embedding (default -1).", type=int, default=-1)
advparser.add_argument("--visionmintokens","--image-min-tokens", metavar=('[tokens]'), help="Override the minimum tokens for the MMProj embedding (default -1).", type=int, default=-1)
advparser.add_argument("--websearch", help="Enable the local search engine proxy so Web Searches can be done.", action='store_true')
compatgroup2 = advparser.add_mutually_exclusive_group()
compatgroup2.add_argument("--showgui", help="Always show the GUI instead of launching the model right away when loading settings from a .kcpps file.", action='store_true')
compatgroup2.add_argument("--skiplauncher", help="Doesn't display or use the GUI launcher. Overrides showgui.", action='store_true')
hordeparsergroup = parser.add_argument_group('Horde Worker Commands')
hordeparsergroup.add_argument("--hordemodelname", metavar=('[name]'), help="Sets your AI Horde display model name.", default="")
@@ -11810,30 +11811,30 @@ if __name__ == '__main__':
hordeparsergroup.add_argument("--hordegenlen", metavar=('[amount]'), help="Sets the maximum number of tokens your worker will generate from an AI horde job.", type=int, default=0)
sdparsergroup = parser.add_argument_group('Image Generation Commands')
sdparsergroup.add_argument("--sdmodel", metavar=('[filename]'), help="Specify an image generation safetensors or gguf model to enable image generation.", default="")
sdparsergroup.add_argument("--sdthreads", metavar=('[threads]'), help="Use a different number of threads for image generation if specified. Otherwise, has the same value as --threads.", type=int, default=0)
sdparsergroup.add_argument("--sdaudiovae", metavar=('[filename]'), help="Specify an image generation audio VAE for LTX2.3 video generation.", default="")
sdparsergroup.add_argument("--sdclamped", metavar=('[maxres]'), help="If specified, limit generation steps and image size for shared use. Accepts an extra optional parameter that indicates maximum resolution (eg. 768 clamps to 768x768, min 512px, disabled if 0).", nargs='?', const=512, type=int, default=0)
sdparsergroup.add_argument("--sdclampedsoft", metavar=('[maxres]'), help="If specified, limit max image size to curb memory usage. Similar to --sdclamped, but less strict, allows trade-offs between width and height (e.g. 640 would allow 640x640, 512x768 and 768x512 images).", type=int, default=0)
sdparsergroup.add_argument("--sdt5xxl", metavar=('[filename]'), help="Specify a T5-XXL safetensors model. Leave blank if prebaked or unused.", default="")
sdparsergroup.add_argument("--sdclip1", "--sdclipl", metavar=('[filename]'), help="Specify first safetensors Clip model (SD3 or Flux Clip-L, WAN or QwenImg vision). Leave blank if prebaked or unused.", default="")
sdparsergroup.add_argument("--sdclip2", "--sdclipg", metavar=('[filename]'), help="Specify second safetensors Clip model (SD3 Clip-G). Leave blank if prebaked or unused.", default="")
sdparsergroup.add_argument("--sdphotomaker", metavar=('[filename]'), help="PhotoMaker is a model that allows face cloning. Specify a PhotoMaker safetensors model which will be applied replacing img2img. SDXL models only. Leave blank if unused.", default="")
sdparsergroup.add_argument("--sdupscaler", metavar=('[filename]'), help="You can use ESRGAN as an upscaling model to resize images. Leave blank if unused.", default="")
sdparsergroup.add_argument("--sdflashattention", help="Enables Flash Attention for image generation.", action='store_true')
sdparsergroup.add_argument("--sdoffloadcpu", help="Offload image weights in RAM to save VRAM, swap into VRAM when needed.", action='store_true')
sdparsergroup.add_argument("--sdvaedevice", metavar=('[Device ID]'), help=f"VAE device for image generation. GPU index, -1 or 'main' for the main GPU, or 'CPU' (default: {default_sdvaedevice}).", type=sd_get_device_number, default=None)
sdparsergroup.add_argument("--sdclipdevice", metavar=('[Device ID]'), help=f"CLIP / T5 / LLM device for image generation. GPU index, -1 or 'main' for the main GPU, or 'CPU' (default: {default_sdclipdevice}).", type=sd_get_device_number, default=None)
sdparsergroup.add_argument("--sdconvdirect", help="Enables Conv2D Direct. May improve performance or reduce memory usage. Might crash if not supported by the backend. Can be 'off' (default) to disable, 'full' to turn it on for all operations, or 'vaeonly' to enable only for the VAE.", type=sd_convdirect_option, choices=sd_convdirect_choices, default=sd_convdirect_choices[0])
sdparsergroup.add_argument("--sdflashattention", help="Enables Flash Attention for image generation.", action='store_true')
sdparsergrouplora = sdparsergroup.add_mutually_exclusive_group()
sdparsergrouplora.add_argument("--sdlora", metavar=('[filename]'), help="Specify image generation LoRAs safetensors models to be applied. Multiple LoRAs are accepted.", nargs='+')
sdparsergroup.add_argument("--sdloramult", metavar=('[amounts]'), help="Multipliers for the image LoRA model to be applied.", type=float, nargs='+', default=[1.0])
sdparsergroup.add_argument("--sdmaingpu", metavar=('[Device ID]'), help="If specified, Image Generation weights will be placed on the selected GPU index. GPU index, -1 or 'main' for the main GPU, or 'CPU' (default: 'main')", type=sd_get_device_number, default=None)
sdparsergroup.add_argument("--sdmodel", metavar=('[filename]'), help="Specify an image generation safetensors or gguf model to enable image generation.", default="")
sdparsergroup.add_argument("--sdoffloadcpu", help="Offload image weights in RAM to save VRAM, swap into VRAM when needed.", action='store_true')
sdparsergroup.add_argument("--sdphotomaker", metavar=('[filename]'), help="PhotoMaker is a model that allows face cloning. Specify a PhotoMaker safetensors model which will be applied replacing img2img. SDXL models only. Leave blank if unused.", default="")
sdparsergrouplora.add_argument("--sdquant", metavar=('[quantization level 0/1/2]'), help="If specified, loads the model quantized to save memory. 0=off, 1=q8, 2=q4", type=int, choices=[0,1,2], nargs="?", const=2, default=0)
sdparsergroup.add_argument("--sdt5xxl", metavar=('[filename]'), help="Specify a T5-XXL safetensors model. Leave blank if prebaked or unused.", default="")
sdparsergroup.add_argument("--sdthreads", metavar=('[threads]'), help="Use a different number of threads for image generation if specified. Otherwise, has the same value as --threads.", type=int, default=0)
sdparsergroup.add_argument("--sdtiledvae", metavar=('[maxres]'), help="Adjust the automatic VAE tiling trigger for images above this size. 0 disables vae tiling.", type=int, default=default_vae_tile_threshold)
sdparsergroup.add_argument("--sdupscaler", metavar=('[filename]'), help="You can use ESRGAN as an upscaling model to resize images. Leave blank if unused.", default="")
sdparsergroupvae = sdparsergroup.add_mutually_exclusive_group()
sdparsergroupvae.add_argument("--sdvae", metavar=('[filename]'), help="Specify an image generation safetensors VAE which replaces the one in the model.", default="")
sdparsergroupvae.add_argument("--sdvaeauto", help="Uses a built-in tiny VAE via TAE SD, which is very fast, and fixed bad VAEs.", action='store_true')
sdparsergroup.add_argument("--sdaudiovae", metavar=('[filename]'), help="Specify an image generation audio VAE for LTX2.3 video generation.", default="")
sdparsergrouplora = sdparsergroup.add_mutually_exclusive_group()
sdparsergrouplora.add_argument("--sdquant", metavar=('[quantization level 0/1/2]'), help="If specified, loads the model quantized to save memory. 0=off, 1=q8, 2=q4", type=int, choices=[0,1,2], nargs="?", const=2, default=0)
sdparsergrouplora.add_argument("--sdlora", metavar=('[filename]'), help="Specify image generation LoRAs safetensors models to be applied. Multiple LoRAs are accepted.", nargs='+')
sdparsergroup.add_argument("--sdloramult", metavar=('[amounts]'), help="Multipliers for the image LoRA model to be applied.", type=float, nargs='+', default=[1.0])
sdparsergroup.add_argument("--sdtiledvae", metavar=('[maxres]'), help="Adjust the automatic VAE tiling trigger for images above this size. 0 disables vae tiling.", type=int, default=default_vae_tile_threshold)
sdparsergroup.add_argument("--sdmaingpu", metavar=('[Device ID]'), help="If specified, Image Generation weights will be placed on the selected GPU index. GPU index, -1 or 'main' for the main GPU, or 'CPU' (default: 'main')", type=sd_get_device_number, default=None)
sdparsergroup.add_argument("--sdvaedevice", metavar=('[Device ID]'), help=f"VAE device for image generation. GPU index, -1 or 'main' for the main GPU, or 'CPU' (default: {default_sdvaedevice}).", type=sd_get_device_number, default=None)
sdparsergroup.add_argument("--sdvramlimit", metavar=('[limit MB]'), help="If set, prevent VRAM usage from image gen from using more than this many MB.", type=int, default=0)
whisperparsergroup = parser.add_argument_group('Whisper Transcription Commands')