Rationalization of BBS steps after benchmarks

The useful has been added, the useless has been removed.

Author: Nexesenex

koboldcpp.py

@@ -2650,10 +2650,10 @@ def hide_tooltip(event):
 
     tabcontent = {}
     # slider data
-    blasbatchsize_values = ["-1", "1", "2", "4", "8", "16", "32", "40", "48", "64", "80", "96", "128", "160", "192", "256", "320", "384", "512", "640", "768", "1024", "1280", "1536", "2048", "2560", "3072", "4096"]
-    blasbatchsize_text = ["Don't Batch BLAS","1","2","4","8","16","32","40","48","64","80","96","128","160","192","256","320","384","512","640","768","1024","1280","1536","2048","2560","3072","4096"]
-    blasubatchsize_values = ["0", "1", "2", "4", "8", "16", "32", "40", "48", "64", "80", "96", "128", "160", "192", "256", "320", "384", "512", "640", "768", "1024", "1280", "1536", "2048", "2560", "3072", "4096"]
-    blasubatchsize_text = ["0 - Physical Blas Batch same as Logical","1","2","4","8","16","32","40","48","64","80","96","128","160","192","256","320","384","512","640","768","1024","1280","1536","2048","2560","3072","4096"]
+    blasbatchsize_values = ["-1", "1", "2", "4", "8", "16", "20", "24", "28", "32", "40", "48", "56", "64", "80", "96", "112", "128", "160", "192", "224", "256", "384", "512", "768", "1024", "1536", "2048", "3072", "4096"]
+    blasbatchsize_text = ["Don't Batch BLAS","1","2","4","8","16","20","24","28","32","40","48","56","64","80","96","112","128","160","192","224","256","384","512","768","1024","1536","2048","3072","4096"]
+    blasubatchsize_values = ["0", "1", "2", "4", "8", "16", "20", "24", "28", "32", "40", "48", "56", "64", "80", "96", "112", "128", "160", "192", "224", "256", "384", "512", "768", "1024", "1536", "2048", "3072", "4096"]
+    blasubatchsize_text = ["0 - Physical Blas Batch same as Logical","1","2","4","8","16","20","24","28","32","40","48","56","64","80","96","112","128","160","192","224","256","384","512","768","1024","1536","2048","3072","4096"]
     contextsize_text = ["128", "256", "384", "512", "640", "768", "896", "1024", "1152", "1280", "1408", "1536", "1664", "1792", "1920", "2048", "2176", "2304", "2432", "2560", "2688", "2816", "2944", "3072", "3200", "3328", "3456", "3584", "3712", "3840", "3968", "4096", "4224", "4352", "4480", "4608", "4736", "4864", "4992", "5120", "5248", "5376", "5504", "5632", "5760", "5888", "6016", "6144", "6272", "6400", "6528", "6656", "6784", "6912", "7040", "7168", "7296", "7424", "7552", "7680", "7808", "7936", "8064", "8192", "8320", "8448", "8576", "8704", "8832", "8960", "9088", "9216", "9344", "9472", "9600", "9728", "9856", "9984", "10112", "10240", "10368", "10496", "10624", "10752", "10880", "11008", "11136", "11264", "11392", "11520", "11648", "11776", "11904", "12032", "12160", "12288", "12416", "12544", "12672", "12800", "12928", "13056", "13184", "13312", "13440", "13568", "13696", "13824", "13952", "14080", "14208", "14336", "14464", "14592", "14720", "14848", "14976", "15104", "15232", "15360", "15488", "15616", "15744", "15872", "16000", "16128", "16256", "16384", "16512", "16640", "16768", "16896", "17024", "17152", "17280", "17408", "17536", "17664", "17792", "17920", "18048", "18176", "18304", "18432", "18560", "18688", "18816", "18944", "19072", "19200", "19328", "19456", "19584", "19712", "19840", "19968", "20096", "20224", "20352", "20480", "20608", "20736", "20864", "20992", "21120", "21248", "21376", "21504", "21632", "21760", "21888", "22016", "22144", "22272", "22400", "22528", "22656", "22784", "22912", "23040", "23168", "23296", "23424", "23552", "23680", "23808", "23936", "24064", "24192", "24320", "24448", "24576", "24704", "24832", "24960", "25088", "25216", "25344", "25472", "25600", "25728", "25856", "25984", "26112", "26240", "26368", "26496", "26624", "26752", "26880", "27008", "27136", "27264", "27392", "27520", "27648", "27776", "27904", "28032", "28160", "28288", "28416", "28544", "28672", "28800", "28928", "29056", "29184", "29312", "29440", "29568", "29696", "29824", "29952", "30080", "30208", "30336", "30464", "30592", "30720", "30848", "30976", "31104", "31232", "31360", "31488", "31616", "31744", "31872", "32000", "32128", "32256", "32384", "32512", "32640", "32768", "32896", "33024", "33152", "33280", "33408", "33536", "33664", "33792", "33920", "34048", "34176", "34304", "34432", "34560", "34688", "34816", "34944", "35072", "35200", "35328", "35456", "35584", "35712", "35840", "35968", "36096", "36224", "36352", "36480", "36608", "36736", "36864", "36992", "37120", "37248", "37376", "37504", "37632", "37760", "37888", "38016", "38144", "38272", "38400", "38528", "38656", "38784", "38912", "39040", "39168", "39296", "39424", "39552", "39680", "39808", "39936", "40064", "40192", "40320", "40448", "40576", "40704", "40832", "40960", "41088", "41216", "41344", "41472", "41600", "41728", "41856", "41984", "42112", "42240", "42368", "42496", "42624", "42752", "42880", "43008", "43136", "43264", "43392", "43520", "43648", "43776", "43904", "44032", "44160", "44288", "44416", "44544", "44672", "44800", "44928", "45056", "45184", "45312", "45440", "45568", "45696", "45824", "45952", "46080", "46208", "46336", "46464", "46592", "46720", "46848", "46976", "47104", "47232", "47360", "47488", "47616", "47744", "47872", "48000", "48128", "48256", "48384", "48512", "48640", "48768", "48896", "49024", "49152", "49408", "49664", "49920", "50176", "50432", "50688", "50944", "51200", "51456", "51712", "51968", "52224", "52480", "52736", "52992", "53248", "53504", "53760", "54016", "54272", "54528", "54784", "55040", "55296", "55552", "55808", "56064", "56320", "56576", "56832", "57088", "57344", "57600", "57856", "58112", "58368", "58624", "58880", "59136", "59392", "59648", "59904", "60160", "60416", "60672", "60928", "61184", "61440", "61696", "61952", "62208", "62464", "62720", "62976", "63232", "63488", "63744", "64000", "64256", "64512", "64768", "65024", "65280", "65536", "66048", "66560", "67072", "67584", "68096", "68608", "69120", "69632", "70144", "70656", "71168", "71680", "72192", "72704", "73216", "73728", "74240", "74752", "75264", "75776", "76288", "76800", "77312", "77824", "78336", "78848", "79360", "79872", "80384", "80896", "81408", "81920", "82432", "82944", "83456", "83968", "84480", "84992", "85504", "86016", "86528", "87040", "87552", "88064", "88576", "89088", "89600", "90112", "90624", "91136", "91648", "92160", "92672", "93184", "93696", "94208", "94720", "95232", "95744", "96256", "96768", "97280", "97792", "98304", "99328", "100352", "101476", "102400", "103424", "104448", "105472", "106496", "107520", "108544", "109568", "110592", "111616", "112640", "113664", "114688", "115712", "116736", "117760", "118784", "119808", "120832", "121856", "122880", "123904", "124928", "125952", "126976", "128000", "129024", "130048", "131072", "132096", "133120", "134144", "135168", "136192", "137216", "138240", "139264", "140288", "141312", "142336", "143360", "144384", "145408", "146432", "147456", "148480", "149504", "150528", "151552", "152576", "153600", "154624", "155648", "156672", "157696", "158720", "159744", "160768", "161792", "162816", "163840", "164864", "165888", "166912", "167936", "168960", "169984", "171008", "172032", "173056", "174088", "175112", "176128", "177152", "178176", "179200", "180224", "181248", "182272", "183296", "184320", "185344", "186368", "187392", "188416", "189440", "190464", "191488", "192512", "193536", "194560", "195584", "196608", "198656", "200704", "202752", "204800", "206848", "208896", "210944", "212992", "215040", "217088", "219136", "221184", "223232", "225280", "227328", "229376", "231424", "233472", "235520", "237568", "239616", "241664", "243712", "245760", "247808", "249856", "251904", "253952", "256000", "258048", "260096", "262144", "266140", "270336", "274432", "278528", "282524", "286720", "290816", "294912", "299008", "303104", "307200", "311296", "315392", "319488", "323584", "327680", "331776", "335872", "339968", "344064", "348160", "352256", "356352", "360448", "364544", "368640", "372736", "376832", "380928", "385024", "389120", "393216", "401408", "409600", "417792", "425984", "434176", "442368", "450560", "458752", "466944", "475136", "483328", "491520", "499712", "507904", "516096", "524288", "540672", "557056", "573440", "589824", "606208", "622592", "638976", "655360", "671744", "688128", "704512", "720896", "737280", "753664", "770048", "786432", "802816", "819200", "835584", "851968", "868352", "884736", "901120", "917504", "933888", "950272", "966656", "983040", "999424", "1015808", "1032192", "1048576"]
     antirunopts = [opt.replace("Use ", "") for lib, opt in lib_option_pairs if not (opt in runopts)]
     quantkv_values = ["0", "1", "2", "3", "4", "5", "6", "7", "8", "9", "10", "11", "12", "13", "14", "15", "16", "17", "18", "19", "20", "21", "22"]
@@ -3111,8 +3111,8 @@ def changerunmode(a,b,c):
     makelabelentry(quick_tab, "BLAS threads:" , blas_threads_var, 14, 50,tooltip="How many threads to use during BLAS processing.\nIf left blank, uses same value as regular thread count.")
 
     # blas batch size
-    makeslider(quick_tab, "BLAS Logical Batch Size - optimum of 128 if not filled :", blasbatchsize_text, blasbatchsize_var, 0, 27, 16, width=391, set=12,tooltip="How many tokens to process at once per batch.\nLarger values use more memory unless Physical Batch supersedes it.")
-    makeslider(quick_tab, "BLAS Physical Batch Size - same as Logical if not filled :", blasubatchsize_text, blasubatchsize_var, 0, 27, 18, width=391, set=0,tooltip="How many tokens to process at once per batch.\nLarger values use more memory.")
+    makeslider(quick_tab, "BLAS Logical Batch Size - optimum of 128 if not filled :", blasbatchsize_text, blasbatchsize_var, 0, 29, 16, width=391, set=17,tooltip="How many tokens to process at once per batch.\nLarger values use more memory unless Physical Batch supersedes it.")
+    makeslider(quick_tab, "BLAS Physical Batch Size - same as Logical if not filled :", blasubatchsize_text, blasubatchsize_var, 0, 29, 18, width=391, set=0,tooltip="How many tokens to process at once per batch.\nLarger values use more memory.")
 
     # load model
     makefileentry(quick_tab, "Model:", "Select GGML or GGML Model File", model_var, 40, 576, onchoosefile=on_picked_model_file,tooltiptxt="Select a GGUF or GGML model file on disk to be loaded.")
@@ -3168,8 +3168,8 @@ def changerunmode(a,b,c):
     # blas thread specifier
     makelabelentry(hardware_tab, "BLAS threads:" , blas_threads_var, 14, 50,tooltip="How many threads to use during BLAS processing.\nIf left blank, uses same value as regular thread count.")
     # blas batch size
-    makeslider(hardware_tab, "BLAS Logical Batch Size - optimum of 128 if not filled :", blasbatchsize_text, blasbatchsize_var, 0, 27, 16, width=391, set=12 ,tooltip="How many tokens to process at once per batch.\nLarger values use more memory unless Physical Batch supersedes it.")
-    makeslider(hardware_tab, "BLAS Physical Batch Size - same as Logical if not filled :", blasubatchsize_text, blasubatchsize_var, 0, 27, 18, width=391, set=0,tooltip="How many tokens to process at once per batch.\nLarger values use more memory.")
+    makeslider(hardware_tab, "BLAS Logical Batch Size - optimum of 128 if not filled :", blasbatchsize_text, blasbatchsize_var, 0, 29, 16, width=391, set=17 ,tooltip="How many tokens to process at once per batch.\nLarger values use more memory unless Physical Batch supersedes it.")
+    makeslider(hardware_tab, "BLAS Physical Batch Size - same as Logical if not filled :", blasubatchsize_text, blasubatchsize_var, 0, 29, 18, width=391, set=0,tooltip="How many tokens to process at once per batch.\nLarger values use more memory.")
     blasbatchsize_var.trace("w", changed_gpulayers_estimate)
 
     # force version
@@ -5013,8 +5013,8 @@ def range_checker(arg: str):
     #more advanced params
     advparser = parser.add_argument_group('Advanced Commands')
     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 NTK Rope, a rule of thumb is to double the base frequency to go 50% beyond the base context, and to triple the base frequency to double the context. Beyond, the NTK calculations are more complex and you might need to use frequency scale as well. To use only 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("--blasbatchsize", help="Sets the Logical batch size used in BLAS processing (default 128 for VRAM savings, optimal speed is 512, 256 is a great compromise). Setting it to -1 disables BLAS mode, but keeps other benefits like GPU offload.", type=int,choices=[-1,1,2,4,8,16,32,40,48,64,80,96,128,160,192,256,320,384,512,640,768,1024,1280,1536,2048,2560,3072,4096], default=128)
-    advparser.add_argument("--blasubatchsize", help="Sets the Physical batch size used in BLAS processing (default 128 for VRAM savings, optimal speed is 512, 256 is a great compromise). Setting it to 0 alignes Physical BLAS batch on logical BLAS.", type=int,choices=[0,1,2,4,8,16,32,40,48,64,80,96,128,160,192,256,320,384,512,640,768,1024,1280,1536,2048,2560,3072,4096], default=0)
+    advparser.add_argument("--blasbatchsize", help="Sets the Logical batch size used in BLAS processing (default 128 for VRAM savings, optimal speed is 512, 256 is a great compromise). Setting it to -1 disables BLAS mode, but keeps other benefits like GPU offload. Max 4096.", type=check_range(int,-1,4096), default=128)
+    advparser.add_argument("--blasubatchsize", help="Sets the Physical batch size used in BLAS processing (default 128 for VRAM savings, optimal speed is 512, 256 is a great compromise). Setting it to 0 alignes Physical BLAS batch on logical BLAS. Max 4096.", type=check_range(int,0,4096), default=0)
     advparser.add_argument("--blasthreads", help="Use a different number of threads during BLAS if specified. Otherwise, has the same value as --threads",metavar=('[threads]'), type=int, default=0)
     advparser.add_argument("--lora", help="LLAMA models only, applies a lora file on top of model. Experimental.", metavar=('[lora_filename]', '[lora_base]'), nargs='+')
     advparser.add_argument("--noshift", help="If set, do not attempt to Trim and Shift the GGUF context without reprocessing everything once the max context is reached. If you disable it (or need to use Quantized KV cache (KVQ) with FlashAttention, aka. modes 1 to 20, which are incompatible with Context Shift), you can eventually use --smartcontext instead.", action='store_true')