File tree Expand file tree Collapse file tree
Expand file tree Collapse file tree Original file line number Diff line number Diff line change @@ -907,6 +907,90 @@ const docTemplate = `{
907907 }
908908 }
909909 },
910+ "/api/nodes/{id}/max-replicas-per-model": {
911+ "put": {
912+ "tags": [
913+ "Nodes"
914+ ],
915+ "summary": "Update a node's max replicas per model",
916+ "parameters": [
917+ {
918+ "type": "string",
919+ "description": "Node ID",
920+ "name": "id",
921+ "in": "path",
922+ "required": true
923+ },
924+ {
925+ "description": "New value",
926+ "name": "request",
927+ "in": "body",
928+ "required": true,
929+ "schema": {
930+ "$ref": "#/definitions/localai.UpdateMaxReplicasPerModelRequest"
931+ }
932+ }
933+ ],
934+ "responses": {
935+ "200": {
936+ "description": "OK",
937+ "schema": {
938+ "type": "object",
939+ "additionalProperties": {
940+ "type": "integer"
941+ }
942+ }
943+ },
944+ "400": {
945+ "description": "value must be \u003e= 1",
946+ "schema": {
947+ "type": "object",
948+ "additionalProperties": true
949+ }
950+ },
951+ "404": {
952+ "description": "node not found",
953+ "schema": {
954+ "type": "object",
955+ "additionalProperties": true
956+ }
957+ }
958+ }
959+ },
960+ "delete": {
961+ "tags": [
962+ "Nodes"
963+ ],
964+ "summary": "Reset a node's max replicas per model to the worker default",
965+ "parameters": [
966+ {
967+ "type": "string",
968+ "description": "Node ID",
969+ "name": "id",
970+ "in": "path",
971+ "required": true
972+ }
973+ ],
974+ "responses": {
975+ "200": {
976+ "description": "OK",
977+ "schema": {
978+ "type": "object",
979+ "additionalProperties": {
980+ "type": "boolean"
981+ }
982+ }
983+ },
984+ "404": {
985+ "description": "node not found",
986+ "schema": {
987+ "type": "object",
988+ "additionalProperties": true
989+ }
990+ }
991+ }
992+ }
993+ },
910994 "/api/p2p": {
911995 "get": {
912996 "tags": [
@@ -2951,6 +3035,15 @@ const docTemplate = `{
29513035 }
29523036 }
29533037 },
3038+ "localai.UpdateMaxReplicasPerModelRequest": {
3039+ "type": "object",
3040+ "properties": {
3041+ "value": {
3042+ "description": "Value is the new per-model replica cap on this node. Must be \u003e= 1.",
3043+ "type": "integer"
3044+ }
3045+ }
3046+ },
29543047 "localai.vramEstimateRequest": {
29553048 "type": "object",
29563049 "properties": {
Original file line number Diff line number Diff line change 904904 }
905905 }
906906 },
907+ "/api/nodes/{id}/max-replicas-per-model" : {
908+ "put" : {
909+ "tags" : [
910+ " Nodes"
911+ ],
912+ "summary" : " Update a node's max replicas per model" ,
913+ "parameters" : [
914+ {
915+ "type" : " string" ,
916+ "description" : " Node ID" ,
917+ "name" : " id" ,
918+ "in" : " path" ,
919+ "required" : true
920+ },
921+ {
922+ "description" : " New value" ,
923+ "name" : " request" ,
924+ "in" : " body" ,
925+ "required" : true ,
926+ "schema" : {
927+ "$ref" : " #/definitions/localai.UpdateMaxReplicasPerModelRequest"
928+ }
929+ }
930+ ],
931+ "responses" : {
932+ "200" : {
933+ "description" : " OK" ,
934+ "schema" : {
935+ "type" : " object" ,
936+ "additionalProperties" : {
937+ "type" : " integer"
938+ }
939+ }
940+ },
941+ "400" : {
942+ "description" : " value must be \u003e = 1" ,
943+ "schema" : {
944+ "type" : " object" ,
945+ "additionalProperties" : true
946+ }
947+ },
948+ "404" : {
949+ "description" : " node not found" ,
950+ "schema" : {
951+ "type" : " object" ,
952+ "additionalProperties" : true
953+ }
954+ }
955+ }
956+ },
957+ "delete" : {
958+ "tags" : [
959+ " Nodes"
960+ ],
961+ "summary" : " Reset a node's max replicas per model to the worker default" ,
962+ "parameters" : [
963+ {
964+ "type" : " string" ,
965+ "description" : " Node ID" ,
966+ "name" : " id" ,
967+ "in" : " path" ,
968+ "required" : true
969+ }
970+ ],
971+ "responses" : {
972+ "200" : {
973+ "description" : " OK" ,
974+ "schema" : {
975+ "type" : " object" ,
976+ "additionalProperties" : {
977+ "type" : " boolean"
978+ }
979+ }
980+ },
981+ "404" : {
982+ "description" : " node not found" ,
983+ "schema" : {
984+ "type" : " object" ,
985+ "additionalProperties" : true
986+ }
987+ }
988+ }
989+ }
990+ },
907991 "/api/p2p" : {
908992 "get" : {
909993 "tags" : [
29483032 }
29493033 }
29503034 },
3035+ "localai.UpdateMaxReplicasPerModelRequest" : {
3036+ "type" : " object" ,
3037+ "properties" : {
3038+ "value" : {
3039+ "description" : " Value is the new per-model replica cap on this node. Must be \u003e = 1." ,
3040+ "type" : " integer"
3041+ }
3042+ }
3043+ },
29513044 "localai.vramEstimateRequest" : {
29523045 "type" : " object" ,
29533046 "properties" : {
Original file line number Diff line number Diff line change @@ -306,6 +306,13 @@ definitions:
306306 success :
307307 type : boolean
308308 type : object
309+ localai.UpdateMaxReplicasPerModelRequest :
310+ properties :
311+ value :
312+ description : Value is the new per-model replica cap on this node. Must be
313+ >= 1.
314+ type : integer
315+ type : object
309316 localai.vramEstimateRequest :
310317 properties :
311318 context_size :
@@ -2714,6 +2721,62 @@ paths:
27142721 summary : Estimate VRAM usage for a model
27152722 tags :
27162723 - config
2724+ /api/nodes/{id}/max-replicas-per-model :
2725+ delete :
2726+ parameters :
2727+ - description : Node ID
2728+ in : path
2729+ name : id
2730+ required : true
2731+ type : string
2732+ responses :
2733+ " 200 " :
2734+ description : OK
2735+ schema :
2736+ additionalProperties :
2737+ type : boolean
2738+ type : object
2739+ " 404 " :
2740+ description : node not found
2741+ schema :
2742+ additionalProperties : true
2743+ type : object
2744+ summary : Reset a node's max replicas per model to the worker default
2745+ tags :
2746+ - Nodes
2747+ put :
2748+ parameters :
2749+ - description : Node ID
2750+ in : path
2751+ name : id
2752+ required : true
2753+ type : string
2754+ - description : New value
2755+ in : body
2756+ name : request
2757+ required : true
2758+ schema :
2759+ $ref : ' #/definitions/localai.UpdateMaxReplicasPerModelRequest'
2760+ responses :
2761+ " 200 " :
2762+ description : OK
2763+ schema :
2764+ additionalProperties :
2765+ type : integer
2766+ type : object
2767+ " 400 " :
2768+ description : value must be >= 1
2769+ schema :
2770+ additionalProperties : true
2771+ type : object
2772+ " 404 " :
2773+ description : node not found
2774+ schema :
2775+ additionalProperties : true
2776+ type : object
2777+ summary : Update a node's max replicas per model
2778+ tags :
2779+ - Nodes
27172780 /api/p2p :
27182781 get :
27192782 responses :
You can’t perform that action at this time.
0 commit comments