turbollm 1.7.2 → 1.7.3

package/dist/webdist/assets/ModelDetailDialog-M2tPWvP4.js

@@ -0,0 +1 @@
+import{g as $,X as ae,f as q,k as we,l as Se,m as Pe,n as Ae,o as Le,u as Fe,p as Re,A as D,B as G,t as H}from"./index-tVO2x-Rw.js";import{R as ze,j as e,P as Ge,a as Ee,C as Te,b as De,D as Ue}from"./radix-DVGeLgPN.js";import{r as g}from"./vendor-BE0HADM8.js";import{e as $e,g as qe,A as Ve,a as Be,b as Ke,c as He,d as Ie,f as Oe,h as _e,i as We}from"./vram-DVAmNgRC.js";import{E as Qe}from"./refresh-cw-BVSLL9xa.js";import{C as Je}from"./circle-check-ByN99MCf.js";import{S as Xe}from"./save-DI4M5kP-.js";import{R as Ze}from"./rotate-ccw-B7QiZxpr.js";const Ye=[["path",{d:"m12 14 4-4",key:"9kzdfg"}],["path",{d:"M3.34 19a10 10 0 1 1 17.32 0",key:"19p75a"}]],ee=$("gauge",Ye);const et=[["path",{d:"m21 21-4.34-4.34",key:"14j7rj"}],["circle",{cx:"11",cy:"11",r:"8",key:"4ej97u"}]],St=$("search",et);const tt=[["path",{d:"M10 5H3",key:"1qgfaw"}],["path",{d:"M12 19H3",key:"yhmn1j"}],["path",{d:"M14 3v4",key:"1sua03"}],["path",{d:"M16 17v4",key:"1q0r14"}],["path",{d:"M21 12h-9",key:"1o4lsq"}],["path",{d:"M21 19h-5",key:"1rlt1p"}],["path",{d:"M21 5h-7",key:"1oszz2"}],["path",{d:"M8 10v4",key:"tgpxqk"}],["path",{d:"M8 12H3",key:"a7s4jb"}]],Pt=$("sliders-horizontal",tt);const nt=[["path",{d:"M4 14a1 1 0 0 1-.78-1.63l9.9-10.2a.5.5 0 0 1 .86.46l-1.92 6.02A1 1 0 0 0 13 10h7a1 1 0 0 1 .78 1.63l-9.9 10.2a.5.5 0 0 1-.86-.46l1.92-6.02A1 1 0 0 0 11 14z",key:"1xq2db"}]],se=$("zap",nt);function te(){return{splitMode:"layer",tensorSplit:[],mainGpu:-1,tensorParallelSize:1}}function at(){return{maxModelLen:0,gpuMemoryUtilization:.9,maxNumSeqs:0,dtype:"auto",kvCacheDtype:"auto",enforceEager:!1,trustRemoteCode:!1}}const st=ze,le=g.forwardRef(({className:n,children:s,...l},r)=>e.jsx(Ge,{children:e.jsxs(Ee,{ref:r,className:q("tllm-sheet fixed z-50 flex flex-col border-border bg-panel shadow-[var(--shadow-2)] focus:outline-none",n),...l,children:[s,e.jsx(Te,{className:"absolute right-4 top-4 rounded-sm text-muted hover:text-ink","aria-label":"Close",children:e.jsx(ae,{size:16})})]})}));le.displayName="SheetContent";function lt({className:n,...s}){return e.jsx("div",{className:q("mb-4 flex flex-col gap-1",n),...s})}const oe=g.forwardRef(({className:n,...s},l)=>e.jsx(De,{ref:l,className:q("text-[16px] font-semibold tracking-[-0.01em] text-ink",n),...s}));oe.displayName="SheetTitle";const re=g.forwardRef(({className:n,...s},l)=>e.jsx(Ue,{ref:l,className:q("text-[13px] text-muted",n),...s}));re.displayName="SheetDescription";function ot(n){switch(n){case"llama-server":return"llamacpp";case"mlx":return"mlx";case"vllm":return"vllm";default:return"none"}}function At({modelKey:n,onClose:s,onViewRepo:l}){const r=we(),i=r.data?.engines.find(t=>t.id===r.data?.activeEngineId),c=Se(n,i?.id),d=Pe(),u=Ae(),f=Le(),[y,S]=g.useState(null),[a,j]=g.useState(null),[J,me]=g.useState(!1),[V,pe]=g.useState(!0),xe=Fe(),o=c.data,B=Re().data?.engine.state;g.useEffect(()=>{o&&j(structuredClone(o.profile))},[o]),g.useEffect(()=>{if(!n)return;const t=document.documentElement,m=parseInt(rt()??"",10);if(m>0){const h=Math.min(Math.max(m,Q),ce());t.style.setProperty("--tllm-config-w",`${h}px`)}return t.classList.add("tllm-config-open"),()=>t.classList.remove("tllm-config-open")},[n]),g.useEffect(()=>{y&&(B==="stopped"||B==="error")&&(u.start.mutate({key:y,base:a??void 0}),S(null))},[y,B,u.start]);const ue=i?.capabilities.kvTypes??["f16"],E=i?.capabilities.flags??[],A=t=>E.length===0||E.includes(t),X=t=>E.length===0||E.includes(`spec-type:${t}`),he=(o?.arch??"").toLowerCase(),fe=(o?.name??"").toLowerCase(),ge=/gemma4/.test(he)||/gemma[ _-]?4/.test(fe),ve=(o?.nextnLayers??0)>0,L=["off"];A("--spec-type")&&A("--mtp-head")&&ge&&L.push("mtp"),A("--spec-type")&&A("--model-draft")&&ve&&(X("nextn")||X("draft-mtp"))&&L.push("nextn"),A("--model-draft")&&L.push("draft");const T=g.useMemo(()=>!o||!a?null:$e(a,o,qe(o.gpus??[],a.gpu)),[o,a]),p=(t,m)=>j(h=>h&&{...h,[t]:m}),C=(t,m)=>j(h=>h&&{...h,sampling:{...h.sampling,[t]:m}}),Z=(t,m)=>j(h=>h&&{...h,gpu:{...h.gpu??te(),[t]:m}}),N=(t,m)=>j(h=>h&&{...h,vllm:{...h.vllm??at(),[t]:m}}),Y=d.load.error instanceof D?d.load.error.message:null,F=!!f&&f.modelKey===o?.key,je=!!f?.running&&F,R=!!f?.done&&F&&!f.running,be=u.start.error instanceof D?u.start.error.message:null,K=ot(i?.kind),P=K==="llamacpp",ye=K==="mlx",ke=K==="vllm",Ce=()=>{o&&(o.loaded?(S(o.key),d.eject.mutate()):u.start.mutate({key:o.key,base:a??void 0}))};g.useEffect(()=>{R&&o?.key&&xe.invalidateQueries({queryKey:["model",o.key]})},[R,o?.key]);const Ne=t=>{u.save.mutate(void 0,{onSuccess:()=>{if(H.success("Tuned settings saved"),t){const m=document.createElement("a");m.href="/api/v1/bench/log",m.click()}},onError:m=>H.error(m instanceof D?m.message:"Could not save tuned settings.")}),s()},Me=()=>{u.cancel.mutate(),s()};return e.jsxs(e.Fragment,{children:[e.jsx(st,{open:!!n,onOpenChange:t=>!t&&s(),modal:!1,children:e.jsxs(le,{className:"overflow-y-auto p-5",onPointerDownOutside:t=>t.preventDefault(),onInteractOutside:t=>t.preventDefault(),children:[e.jsx(it,{}),e.jsxs(lt,{children:[e.jsx(oe,{className:"truncate",children:o?.name??"Model"}),e.jsx(re,{children:o?`${o.arch} · ${o.quant} · ${gt(o.sizeBytes)}`:"Load settings"}),o?.sourceRepo&&l&&e.jsxs("button",{type:"button",onClick:()=>l(o.sourceRepo),className:"mt-1 inline-flex w-fit items-center gap-1 text-[12px] font-medium text-accent hover:underline",title:`View ${o.sourceRepo} on Hugging Face`,children:[e.jsx(Qe,{size:12})," Model card & quants on Hugging Face"]})]}),!o||!a?e.jsx("div",{className:"py-10 text-center text-[13px] text-muted",children:"Loading…"}):e.jsxs("div",{className:"flex flex-col gap-4",children:[T&&e.jsx(dt,{estMb:T.estMb,totalMb:T.totalVramMb,verdict:T.verdict}),P&&o.gpu&&e.jsx(mt,{running:je,done:R,step:F?f?.step:void 0,bestTps:F?f?.bestTps:void 0,resultError:R?f?.error:void 0,benchTps:o.benchTps,tuned:a.tunedBy==="bench",loaded:o.loaded,startError:be,pending:u.start.isPending||y!==null,onStart:Ce,onCancel:()=>u.cancel.mutate()}),ye&&e.jsxs("div",{className:"rounded-md border border-border bg-panel-2 px-3 py-2.5 text-[12px] text-muted",children:["MLX (Apple Silicon, mlx-lm) manages context and KV cache automatically — there are no context/GPU-layer/KV knobs to set. Only the ",e.jsx("span",{className:"text-ink",children:"sampling defaults"})," below apply at load; per-conversation overrides still work in chat."]}),ke&&e.jsxs(M,{children:[e.jsx(x,{label:"Max model length",hint:"vLLM --max-model-len. Max context tokens. Leave blank to derive from the model config.",children:e.jsx(k,{value:a.vllm?.maxModelLen||void 0,placeholder:"from model",min:1,max:o.nativeCtx||1048576,step:1024,onChange:t=>N("maxModelLen",t??0)})}),e.jsx(v,{label:"GPU memory utilization",hint:"vLLM --gpu-memory-utilization. Fraction of VRAM vLLM may reserve. Lower it to share the GPU.",value:Math.round((a.vllm?.gpuMemoryUtilization??.9)*100),min:10,max:100,step:5,onChange:t=>N("gpuMemoryUtilization",t/100),fmt:t=>`${t}%`}),e.jsx(x,{label:"Max concurrent sequences",hint:"vLLM --max-num-seqs. Requests served in parallel.",children:e.jsx(k,{value:a.vllm?.maxNumSeqs||void 0,placeholder:"auto",min:1,max:1024,step:1,onChange:t=>N("maxNumSeqs",t??0)})}),e.jsx(x,{label:"Compute dtype",hint:"vLLM --dtype. 'auto' follows the model's config.",children:e.jsx(_,{value:a.vllm?.dtype??"auto",options:["auto","bfloat16","float16","float32"],onChange:t=>N("dtype",t)})}),e.jsx(x,{label:"KV cache dtype",hint:"vLLM --kv-cache-dtype. fp8 roughly halves KV-cache memory.",children:e.jsx(_,{value:a.vllm?.kvCacheDtype??"auto",options:["auto","fp8"],onChange:t=>N("kvCacheDtype",t)})}),e.jsx(w,{label:"Enforce eager",hint:"vLLM --enforce-eager. Skips CUDA graphs: less VRAM, somewhat slower.",value:a.vllm?.enforceEager??!1,onChange:t=>N("enforceEager",t)}),e.jsx(w,{label:"Trust remote code",hint:"vLLM --trust-remote-code. Needed for models that ship custom modelling code.",value:a.vllm?.trustRemoteCode??!1,onChange:t=>N("trustRemoteCode",t)})]}),P&&e.jsxs(M,{children:[e.jsx(v,{label:"Context length",hint:"Tokens of history the model can use.",value:a.ctx,min:512,max:Math.max(512,o.nativeCtx||8192),step:512,onChange:t=>p("ctx",t),fmt:t=>t.toLocaleString()}),o.gpu&&e.jsx(v,{label:"GPU layers",hint:o.blockCount>0?`${o.blockCount} total layers.`:"All layers on GPU = max performance.",value:a.ngl,min:0,max:o.blockCount>0?o.blockCount:99,step:1,onChange:t=>p("ngl",t),fmt:t=>t>=(o.blockCount>0?o.blockCount:99)?"All":String(t)}),o.moe&&o.blockCount>0&&e.jsx(v,{label:"MoE experts on CPU",hint:"Higher = less VRAM, slower. Lower = faster if it fits.",value:a.nCpuMoe,min:0,max:o.blockCount,step:1,onChange:t=>p("nCpuMoe",t)}),e.jsx(x,{label:"Context overflow",hint:"What to do when the context window fills up.",children:e.jsx(z,{value:a.contextOverflow,options:["shift","keep"],onChange:t=>p("contextOverflow",t)})}),a.contextOverflow==="keep"&&e.jsx(x,{label:"Tokens to keep",hint:"Preserve this many tokens from the start (e.g. system prompt length) when shifting.",children:e.jsx(U,{value:a.nKeep,min:0,max:a.ctx,onChange:t=>p("nKeep",t)})}),e.jsx(x,{label:"Batch size",hint:"--batch-size: logical prompt-processing batch. Larger = faster prefill, more VRAM.",children:e.jsx(k,{value:a.batchSize,placeholder:"2048",min:1,max:65536,step:128,onChange:t=>p("batchSize",t)})}),e.jsx(x,{label:"Micro-batch size",hint:"--ubatch-size: physical micro-batch. Must be ≤ batch size.",children:e.jsx(k,{value:a.uBatchSize,placeholder:"512",min:1,max:65536,step:128,onChange:t=>p("uBatchSize",t)})})]}),o.gpus&&o.gpus.length>1&&(i?.kind==="llama-server"||i?.kind==="vllm")&&e.jsxs(e.Fragment,{children:[e.jsxs(O,{children:["Multi-GPU · ",o.gpus.length," GPUs"]}),e.jsx(M,{children:i.kind==="vllm"?e.jsx(x,{label:"Tensor parallel size",hint:"Shard the model across this many GPUs.",children:e.jsx(U,{value:a.gpu?.tensorParallelSize??1,min:1,max:o.gpus.length,onChange:t=>Z("tensorParallelSize",t)})}):e.jsx(pt,{gpus:o.gpus,gpu:a.gpu??te(),setG:Z})})]}),P&&e.jsxs(M,{children:[e.jsx(x,{label:"Parallel slots",children:e.jsx(U,{value:a.parallel,min:1,max:16,onChange:t=>p("parallel",t)})}),e.jsx(x,{label:"KV cache type",hint:"turbo* are TurboQuant-fork exclusive.",children:e.jsx(_,{value:a.kvTypeK,options:ue,onChange:t=>j(m=>m&&{...m,kvTypeK:t,kvTypeV:t})})}),e.jsx(x,{label:"Flash attention",children:e.jsx(z,{value:a.flashAttn,options:["auto","on","off"],onChange:t=>p("flashAttn",t)})}),e.jsx(x,{label:"KV cache",hint:"VRAM is fastest; RAM frees VRAM for bigger models.",children:e.jsx(z,{value:a.kvOffload===!1?"RAM":"GPU",options:["GPU","RAM"],onChange:t=>p("kvOffload",t==="GPU")})}),e.jsx(w,{label:"Use Jinja chat template",value:a.useJinja,onChange:t=>p("useJinja",t)}),a.useJinja&&e.jsx(W,{label:"Template file path (optional)",hint:"Leave empty to use model's built-in template.",value:a.chatTemplateFile,placeholder:"Path to .jinja template file",onChange:t=>p("chatTemplateFile",t)})]}),e.jsx(O,{children:"Sampling"}),e.jsxs(M,{children:[e.jsx(v,{label:"Temperature",value:a.sampling.temp,min:0,max:2,step:.05,onChange:t=>C("temp",t),fmt:t=>t.toFixed(2)}),e.jsx(v,{label:"Top P",value:a.sampling.topP,min:0,max:1,step:.01,onChange:t=>C("topP",t),fmt:t=>t.toFixed(2)}),e.jsx(v,{label:"Top K",value:a.sampling.topK,min:0,max:200,step:1,onChange:t=>C("topK",t)}),e.jsx(v,{label:"Min P",value:a.sampling.minP,min:0,max:1,step:.01,onChange:t=>C("minP",t),fmt:t=>t.toFixed(2)}),P&&e.jsxs(e.Fragment,{children:[e.jsx(v,{label:"Repeat penalty",hint:"Penalise tokens that appeared earlier. 1.0 = off.",value:a.sampling.repeatPenalty,min:1,max:2,step:.05,onChange:t=>C("repeatPenalty",t),fmt:t=>t.toFixed(2)}),e.jsx(v,{label:"Presence penalty",hint:"Flat penalty for any token that appeared. 0 = off.",value:a.sampling.presencePenalty,min:0,max:2,step:.05,onChange:t=>C("presencePenalty",t),fmt:t=>t.toFixed(2)}),e.jsx(v,{label:"Frequency penalty",hint:"Penalty proportional to how often a token appeared. 0 = off.",value:a.sampling.frequencyPenalty,min:0,max:2,step:.05,onChange:t=>C("frequencyPenalty",t),fmt:t=>t.toFixed(2)}),e.jsx(x,{label:"Stop strings",hint:"Halt generation when any of these sequences is produced.",children:e.jsx("div",{className:"w-full"})}),e.jsx(ht,{value:a.sampling.stop,onChange:t=>j(m=>m&&{...m,sampling:{...m.sampling,stop:t}})})]})]}),P&&L.length>1&&e.jsxs(e.Fragment,{children:[e.jsx(O,{children:"Speculative decoding"}),e.jsxs(M,{children:[e.jsx(x,{label:"Mode",hint:"Predict extra tokens per step — faster generation.",children:e.jsx(ut,{value:a.speculative,options:L,onChange:t=>p("speculative",t)})}),a.speculative==="mtp"&&e.jsx(W,{label:"MTP head GGUF",hint:"Gemma-4 assistant model file.",value:a.mtpHeadPath,placeholder:"Path to gemma-4-mtp-assistant.gguf",onChange:t=>p("mtpHeadPath",t)}),a.speculative==="nextn"&&e.jsx("p",{className:"text-[11px] text-faint",children:"Uses this model's built-in NextN head — no extra file needed."}),a.speculative==="draft"&&e.jsx(W,{label:"Draft model GGUF",hint:"A small same-family model.",value:a.draftModelPath,placeholder:"Path to small draft model",onChange:t=>p("draftModelPath",t)}),a.speculative!=="off"&&e.jsxs(e.Fragment,{children:[e.jsx(x,{label:"Max drafts",hint:"Tokens the draft head proposes per step before verification (--draft-max). Default 16.",children:e.jsx(k,{value:a.draftMax,placeholder:"16",min:1,max:64,onChange:t=>p("draftMax",t)})}),e.jsx(x,{label:"Min drafts",hint:"Minimum tokens drafted per step before the main model verifies (--draft-min). Default 1.",children:e.jsx(k,{value:a.draftMin,placeholder:"1",min:0,max:8,onChange:t=>p("draftMin",t)})})]})]})]}),P&&e.jsxs(e.Fragment,{children:[e.jsxs("button",{type:"button",onClick:()=>me(t=>!t),className:"flex items-center gap-1 text-[13px] font-medium text-muted hover:text-ink",children:[e.jsx(Je,{size:14,className:J?"rotate-180 transition-transform":"transition-transform"}),"Advanced"]}),J&&e.jsxs(M,{children:[e.jsx(v,{label:"CPU threads",hint:a.threads===0?`Auto — half your cores (${ft(o.cores)} of ${o.cores||"?"})`:`${a.threads} of ${o.cores||"?"} cores`,value:a.threads,min:0,max:Math.max(1,o.cores||64),step:1,onChange:t=>p("threads",t),fmt:t=>t===0?"Auto":String(t)}),e.jsx(w,{label:"Reuse cached prompt",hint:"Skips re-processing the unchanged start of a prompt on repeated or continued requests — faster first token. Recommended on.",value:a.cacheReuse>0,onChange:t=>p("cacheReuse",t?256:0)}),o.vision&&e.jsx(w,{label:"Vision encoder on GPU",value:a.mmprojGpu,onChange:t=>p("mmprojGpu",t)}),a.parallel>1&&e.jsx(w,{label:"Unified KV across slots",value:a.kvUnified,onChange:t=>p("kvUnified",t)}),e.jsx(x,{label:"RoPE scaling",hint:"Extend context beyond the model's trained limit. 'none' = model native.",children:e.jsx(z,{value:a.ropeScalingType,options:["none","linear","yarn"],onChange:t=>p("ropeScalingType",t)})}),a.ropeScalingType!=="none"&&e.jsxs(e.Fragment,{children:[e.jsx(x,{label:"RoPE freq base",hint:"Base frequency override. Leave blank for model native.",children:e.jsx(k,{value:a.ropeFreqBase||void 0,placeholder:"model native",min:1,max:1e7,onChange:t=>p("ropeFreqBase",t??0)})}),e.jsx(x,{label:"RoPE freq scale",hint:"Frequency scale — e.g. 0.25 for 4× context. Leave blank for model native.",children:e.jsx(k,{value:a.ropeFreqScale||void 0,placeholder:"model native",min:.01,max:10,step:.01,onChange:t=>p("ropeFreqScale",t??0)})})]})]})]}),Y&&e.jsx("p",{className:"text-[12px]",style:{color:"var(--err)"},children:Y}),e.jsxs("label",{className:"flex cursor-pointer items-center gap-2 pt-1 text-[13px] text-ink",children:[e.jsx("input",{type:"checkbox",checked:V,onChange:t=>pe(t.target.checked),className:"h-4 w-4 accent-[var(--accent)]"}),"Remember these settings",e.jsx("span",{className:"text-[11px] text-faint",children:V?"· saved as this model's default":"· this load only"})]}),e.jsxs("div",{className:"flex items-center gap-2",children:[e.jsxs(G,{className:"flex-1",onClick:()=>{const t=()=>d.load.mutate({key:o.key,overrides:a},{onError:m=>H.error(m instanceof D?`Could not load model: ${m.message}`:"Could not load model — check the engine logs on the Engines screen.")});V?d.save.mutate({key:o.key,profile:a,engineId:i?.id??"*"},{onSuccess:t,onError:t}):t(),s()},disabled:d.load.isPending,children:[e.jsx(se,{size:14}),o.loaded?"Reload":"Load model"]}),e.jsxs(G,{variant:"outline",onClick:()=>d.save.mutate({key:o.key,profile:a,engineId:i?.id??"*"}),disabled:d.save.isPending,title:"Save without reloading",children:[e.jsx(Xe,{size:14}),"Save"]}),e.jsx(G,{variant:"ghost",onClick:()=>d.reset.mutate({key:o.key,engineId:i?.id??"*"}),disabled:d.reset.isPending,title:"Reset to defaults",children:e.jsx(Ze,{size:14})})]})]})]})}),e.jsx(ct,{result:R&&F?f?.result:void 0,modelName:o?.name,onSave:Ne,onCancel:Me})]})}const ie="tllm-config-w",Q=360;function ce(){return Math.max(Q,Math.min(760,window.innerWidth-220))}function rt(){try{return localStorage.getItem(ie)}catch{return null}}function it(){const n=s=>{if(s.button!==0)return;s.preventDefault();const l=document.documentElement;l.classList.add("tllm-resizing");const r=c=>{const d=Math.min(Math.max(window.innerWidth-c.clientX,Q),ce());l.style.setProperty("--tllm-config-w",`${Math.round(d)}px`)},i=()=>{l.classList.remove("tllm-resizing"),window.removeEventListener("pointermove",r),window.removeEventListener("pointerup",i);try{localStorage.setItem(ie,getComputedStyle(l).getPropertyValue("--tllm-config-w").trim())}catch{}};window.addEventListener("pointermove",r),window.addEventListener("pointerup",i)};return e.jsx("div",{className:"tllm-config-resizer",onPointerDown:n,role:"separator","aria-orientation":"vertical","aria-label":"Resize panel"})}function b({label:n,value:s,tag:l}){return e.jsxs("div",{className:"flex items-center justify-between gap-3 py-[3px]",children:[e.jsx("span",{className:"text-muted",children:n}),e.jsxs("span",{className:"flex items-center gap-1.5",children:[l&&e.jsx("span",{className:"rounded px-1 py-0.5 text-[10px] font-medium uppercase tracking-wide",style:{color:"var(--accent)",background:"color-mix(in srgb, var(--accent) 12%, transparent)"},children:l}),e.jsx("span",{className:"font-mono text-ink",children:s})]})]})}function I({title:n}){return e.jsx("div",{className:"mt-2 pb-1 pt-1 text-[11px] font-semibold uppercase tracking-wide text-faint first:mt-0 first:pt-0",children:n})}function ct({result:n,modelName:s,onSave:l,onCancel:r}){const[i,c]=g.useState(!0),d=n?.recommendedSampling,u=n?.sampling,f=a=>d?.[a]!=null,y=!!d&&(d.temp!=null||d.topP!=null||d.topK!=null||d.minP!=null),S=[{label:"Temperature",key:"temp"},{label:"Top-K",key:"topK"},{label:"Top-P",key:"topP"},{label:"Min-P",key:"minP"}];return e.jsx(Ve,{open:!!n,onOpenChange:a=>{a||r()},children:e.jsxs(Be,{children:[e.jsxs(Ke,{children:[e.jsx(He,{children:"Auto-tune complete"}),e.jsx(Ie,{asChild:!0,children:e.jsxs("div",{className:"flex flex-col gap-2.5 text-[13px]",children:[e.jsxs("span",{className:"text-muted",children:["Fastest config found on your machine —"," ",e.jsxs("span",{className:"font-mono font-medium",style:{color:"var(--ok)"},children:[n?.tps.toFixed(1)," tok/s"]}),". Save applies this complete config to ",s??"this model",":"]}),n&&e.jsxs("div",{className:"rounded-md border border-border bg-panel-2 px-3 py-1.5",children:[e.jsx(I,{title:"Runtime"}),e.jsx(b,{label:"GPU layers",value:n.params.ngl}),n.params.nCpuMoe>0&&e.jsx(b,{label:"MoE experts on CPU",value:n.params.nCpuMoe}),e.jsx(b,{label:"Context length",value:`${n.params.ctx.toLocaleString()} tok`}),e.jsx(b,{label:"KV cache type",value:n.params.kvTypeK}),e.jsx(b,{label:"Flash attention",value:n.params.flashAttn}),n.params.parallel>1&&e.jsx(b,{label:"Parallel slots",value:n.params.parallel}),e.jsx(I,{title:"Sampling"}),u?S.map(({label:a,key:j})=>e.jsx(b,{label:a,value:u[j]??"—",tag:f(j)?"from card":void 0},j)):e.jsx(b,{label:"Sampling",value:"model defaults"}),e.jsx(I,{title:"Measured"}),e.jsx(b,{label:"Speed",value:`${n.tps.toFixed(1)} tok/s`}),n.vramMb!=null&&e.jsx(b,{label:"VRAM used",value:`~${n.vramMb.toLocaleString()} MB`}),n.ttftMs?e.jsx(b,{label:"First token",value:`${Math.round(n.ttftMs)} ms`}):null]}),e.jsxs("span",{className:"text-faint",children:[y?e.jsxs(e.Fragment,{children:["Values tagged ",e.jsx("span",{style:{color:"var(--accent)"},children:"from card"})," were read from the model's Hugging Face page. "]}):e.jsx(e.Fragment,{children:"No sampling recommendation was found on the model's card — sampling stays at your current values. "}),"Change any of these later in Sampling."]})]})})]}),e.jsxs(Oe,{children:[e.jsxs("label",{className:"mr-auto flex cursor-pointer items-center gap-1.5 text-[12px] text-muted",children:[e.jsx("input",{type:"checkbox",checked:i,onChange:a=>c(a.target.checked),className:"h-3.5 w-3.5 accent-[var(--accent)]"}),"Download run log"]}),e.jsx(_e,{onClick:r,children:"Cancel"}),e.jsx(We,{onClick:()=>l(i),children:"Save"})]})]})})}function dt({estMb:n,totalMb:s,verdict:l}){if(l==="cpu"||s===0)return e.jsx("div",{className:"rounded-md border border-border bg-panel-2 px-3 py-2 text-[12px] text-muted",children:"CPU-only — no GPU detected."});const r=l==="fits"?"var(--ok)":l==="tight"?"var(--warn)":"var(--err)",i=Math.min(100,Math.round(n/s*100)),c=l==="overflow"?"will spill to system RAM — expect severe slowdown":l==="tight"?"may slow under desktop load":"fits comfortably";return e.jsxs("div",{className:"rounded-md border border-border bg-panel-2 px-3 py-2.5",children:[e.jsxs("div",{className:"mb-1.5 flex items-center justify-between text-[12px]",children:[e.jsx("span",{className:"text-muted",children:"Estimated VRAM"}),e.jsxs("span",{style:{color:r},children:["~",(n/1e3).toFixed(1)," / ",(s/1e3).toFixed(1)," GB · ",c]})]}),e.jsx("div",{className:"h-2 overflow-hidden rounded-full",style:{background:"var(--border)"},children:e.jsx("div",{className:"h-full rounded-full transition-all",style:{width:`${i}%`,background:r}})})]})}function mt({running:n,done:s,step:l,bestTps:r,resultError:i,benchTps:c,tuned:d,loaded:u,startError:f,pending:y,onStart:S,onCancel:a}){return e.jsx("div",{className:"rounded-lg border border-border bg-panel p-3",children:n?e.jsxs("div",{className:"flex flex-col gap-2",children:[e.jsxs("div",{className:"flex items-center justify-between gap-3",children:[e.jsxs("div",{className:"flex items-center gap-2 text-[13px] text-ink",children:[e.jsx(ee,{size:14,className:"animate-pulse",style:{color:"var(--accent)"}}),"Auto-tuning…"]}),e.jsxs(G,{variant:"outline",onClick:a,children:[e.jsx(ae,{size:14}),"Cancel"]})]}),e.jsx("p",{className:"font-mono text-[12px] text-muted",children:l??"Preparing…"}),r!==void 0&&e.jsxs("p",{className:"text-[12px] text-ink",children:["Best so far: ",e.jsxs("span",{className:"font-mono",style:{color:"var(--ok)"},children:[r.toFixed(1)," tok/s"]})]})]}):e.jsxs("div",{className:"flex flex-col gap-2",children:[e.jsxs("div",{className:"flex items-center justify-between gap-3",children:[e.jsxs("div",{className:"min-w-0",children:[e.jsxs("div",{className:"flex items-center gap-1.5 text-[13px] font-medium text-ink",children:[e.jsx(ee,{size:14,style:{color:"var(--accent)"}}),"Auto-tune"]}),e.jsx("p",{className:"text-[11px] text-faint",children:c!==null?`Tuned: ${c.toFixed(1)} tok/s on your machine${d?" · settings applied below":""}`:"Measures real speed on your hardware and saves the fastest settings (~3 min)."})]}),e.jsxs(G,{variant:"outline",onClick:S,disabled:y,children:[e.jsx(se,{size:14}),u?"Stop & benchmark":"Auto-tune"]})]}),s&&r!==void 0&&!i&&e.jsxs("p",{className:"text-[12px] text-ink",children:["Done — ",e.jsxs("span",{className:"font-mono",style:{color:"var(--ok)"},children:[r.toFixed(1)," tok/s"]}),". Saved as this model's default."]}),s&&i&&e.jsx("p",{className:"text-[12px]",style:{color:"var(--warn)"},children:i}),f&&e.jsx("p",{className:"text-[12px]",style:{color:"var(--err)"},children:f})]})})}function M({children:n}){return e.jsx("div",{className:"flex flex-col gap-3 rounded-lg border border-border bg-panel p-3",children:n})}function O({children:n}){return e.jsx("div",{className:"text-[12px] font-semibold uppercase tracking-wide text-faint",children:n})}function v({label:n,hint:s,value:l,min:r,max:i,step:c,onChange:d,fmt:u}){return e.jsxs("div",{children:[e.jsxs("div",{className:"mb-1 flex items-baseline justify-between",children:[e.jsx("span",{className:"text-[13px] text-ink",children:n}),e.jsx("span",{className:"font-mono text-[12px] text-muted",children:u?u(l):l})]}),e.jsx("input",{type:"range",min:r,max:i,step:c,value:l,onChange:f=>d(Number(f.target.value)),className:"w-full accent-[var(--accent)]"}),s&&e.jsx("p",{className:"mt-0.5 text-[11px] text-faint",children:s})]})}function x({label:n,hint:s,children:l}){return e.jsxs("div",{className:"flex items-center justify-between gap-3",children:[e.jsxs("div",{className:"min-w-0",children:[e.jsx("div",{className:"text-[13px] text-ink",children:n}),s&&e.jsx("div",{className:"text-[11px] text-faint",children:s})]}),e.jsx("div",{className:"shrink-0",children:l})]})}function U({value:n,min:s,max:l,step:r=1,onChange:i}){return e.jsx("input",{type:"number",min:s,max:l,step:r,value:n,onChange:c=>i(Math.max(s,Math.min(l,Number(c.target.value)||s))),className:"w-20 rounded-md border border-border bg-bg px-2 py-1 text-right text-[13px] text-ink outline-none"})}function k({value:n,placeholder:s,min:l,max:r,step:i=1,onChange:c}){return e.jsxs("div",{className:"flex items-center gap-1",children:[e.jsx("input",{type:"number",min:l,max:r,step:i,value:n??"",placeholder:s,onChange:d=>{const u=d.target.value;c(u===""?void 0:Math.max(l,Math.min(r,Number(u))))},className:"w-20 rounded-md border border-border bg-bg px-2 py-1 text-right text-[13px] text-ink outline-none placeholder:text-faint"}),n!==void 0&&e.jsx("button",{type:"button",onClick:()=>c(void 0),className:"shrink-0 text-faint hover:text-ink",title:"Reset to default",children:"×"})]})}function _({value:n,options:s,onChange:l}){return e.jsx("select",{value:n,onChange:r=>l(r.target.value),className:"rounded-md border border-border bg-bg px-2 py-1 text-[13px] text-ink outline-none",children:s.map(r=>e.jsx("option",{value:r,children:r},r))})}function z({value:n,options:s,onChange:l}){return e.jsx("div",{className:"flex overflow-hidden rounded-md border border-border",children:s.map(r=>e.jsx("button",{type:"button",onClick:()=>l(r),className:"px-2.5 py-1 text-[12px] capitalize transition-colors",style:{background:n===r?"var(--accent)":"transparent",color:n===r?"var(--on-accent)":"var(--muted)"},children:r},r))})}function de(n,s){const l=n.vramMb?` (${Math.round(n.vramMb/1024)} GB)`:"";return`GPU ${s}${n.name?` · ${n.name}`:""}${l}`}function ne({gpus:n,value:s,allowAuto:l,onChange:r}){return e.jsxs("select",{value:String(s),onChange:i=>r(Number(i.target.value)),className:"max-w-[260px] rounded-md border border-border bg-bg px-2 py-1 text-[13px] text-ink outline-none",children:[l&&e.jsx("option",{value:"-1",children:"Auto"}),n.map((i,c)=>e.jsx("option",{value:String(c),children:de(i,c)},c))]})}function pt({gpus:n,gpu:s,setG:l}){const r=s.tensorSplit.length>0;return e.jsxs(e.Fragment,{children:[e.jsx(x,{label:"Split mode",hint:"layer = by layers (default) · row = tensor-parallel · none = single GPU.",children:e.jsx(z,{value:s.splitMode,options:["layer","row","none"],onChange:i=>l("splitMode",i)})}),s.splitMode==="none"?e.jsx(x,{label:"GPU",hint:"Which GPU to load the model on.",children:e.jsx(ne,{gpus:n,value:s.mainGpu>=0?s.mainGpu:0,allowAuto:!1,onChange:i=>l("mainGpu",i)})}):e.jsxs(e.Fragment,{children:[e.jsx(x,{label:"Main GPU",hint:"Holds the KV cache / small tensors. Auto lets the engine choose.",children:e.jsx(ne,{gpus:n,value:s.mainGpu,allowAuto:!0,onChange:i=>l("mainGpu",i)})}),e.jsx(w,{label:"Custom GPU split",hint:"Off = even split across GPUs. On = set each GPU's share.",value:r,onChange:i=>l("tensorSplit",i?n.map(()=>1):[])}),r&&e.jsx("div",{className:"flex flex-col gap-2 pl-1",children:n.map((i,c)=>e.jsx(x,{label:de(i,c),children:e.jsx(U,{value:s.tensorSplit[c]??0,min:0,max:100,onChange:d=>l("tensorSplit",n.map((u,f)=>f===c?d:s.tensorSplit[f]??0))})},c))})]})]})}const xt={off:"Off",mtp:"MTP",nextn:"NextN",draft:"Draft"};function ut({value:n,options:s,onChange:l}){return e.jsx("div",{className:"flex overflow-hidden rounded-md border border-border",children:s.map(r=>e.jsx("button",{type:"button",onClick:()=>l(r),className:"px-2.5 py-1 text-[12px] transition-colors",style:{background:n===r?"var(--accent)":"transparent",color:n===r?"var(--on-accent)":"var(--muted)"},children:xt[r]},r))})}function W({label:n,hint:s,value:l,placeholder:r,onChange:i}){return e.jsxs("div",{className:"flex flex-col gap-1",children:[e.jsxs("div",{className:"text-[12px] text-ink",children:[n,s&&e.jsx("span",{className:"ml-1 text-faint",children:s})]}),e.jsx("input",{type:"text",value:l,placeholder:r,onChange:c=>i(c.target.value),className:"w-full rounded-md border border-border bg-bg px-2 py-1 font-mono text-[12px] text-ink outline-none placeholder:text-faint"})]})}function ht({value:n,onChange:s}){const[l,r]=g.useState(""),i=c=>{const d=c.trim();d&&!n.includes(d)&&s([...n,d]),r("")};return e.jsxs("div",{className:"flex flex-col gap-1.5",children:[n.length>0&&e.jsx("div",{className:"flex flex-wrap gap-1",children:n.map(c=>e.jsxs("span",{className:"flex items-center gap-0.5 rounded border border-border bg-panel-2 px-1.5 py-0.5 font-mono text-[11px] text-ink",children:[c,e.jsx("button",{type:"button",onClick:()=>s(n.filter(d=>d!==c)),className:"ml-0.5 text-muted hover:text-[var(--err)]",children:"×"})]},c))}),e.jsx("input",{type:"text",value:l,placeholder:n.length?"Add another…":"Type a stop string, press Enter",onChange:c=>r(c.target.value),onKeyDown:c=>{c.key==="Enter"&&(c.preventDefault(),i(l)),c.key==="Backspace"&&!l&&n.length&&s(n.slice(0,-1))},onBlur:()=>{l.trim()&&i(l)},className:"w-full rounded-md border border-border bg-bg px-2 py-1 font-mono text-[12px] text-ink outline-none placeholder:text-faint"})]})}function w({label:n,hint:s,value:l,onChange:r}){return e.jsxs("label",{className:"flex cursor-pointer items-center justify-between gap-3",children:[e.jsxs("div",{className:"min-w-0",children:[e.jsx("span",{className:"text-[13px] text-ink",children:n}),s&&e.jsx("p",{className:"text-[11px] text-faint",children:s})]}),e.jsx("input",{type:"checkbox",checked:l,onChange:i=>r(i.target.checked),className:"h-4 w-4 shrink-0 accent-[var(--accent)]"})]})}function ft(n){return Math.max(1,Math.floor((n||2)/2))}function gt(n){return n>=1e9?`${(n/1e9).toFixed(1)} GB`:`${Math.round(n/1e6)} MB`}export{At as M,St as S,se as Z,Pt as a,st as b,le as c,lt as d,oe as e,re as f};

package/dist/webdist/assets/ModelDirs-CaIf1TUh.js

@@ -0,0 +1 @@
+import{j as e}from"./radix-DVGeLgPN.js";import{r as c}from"./vendor-BE0HADM8.js";import{A as m,X as x,I as p,B as u}from"./index-tVO2x-Rw.js";import{S as f}from"./star-BCxsXqcv.js";import{F as g}from"./save-DI4M5kP-.js";function D({dirs:a,primaryDir:s,mut:t}){const[o,i]=c.useState(""),n=t.addDir.error instanceof m?t.addDir.error.message:null,l=()=>{const r=o.trim();r&&t.addDir.mutate(r,{onSuccess:()=>i("")})};return e.jsxs("div",{className:"mb-5 rounded-lg border border-border bg-panel-2 p-4",children:[e.jsx("div",{className:"mb-2 text-[13px] font-medium text-ink",children:"Model folders"}),a.length>0&&e.jsx("div",{className:"mb-3 flex flex-col gap-1.5",children:a.map(r=>{const d=r===s;return e.jsxs("div",{className:"group/dir flex items-center gap-2 text-[13px]",children:[e.jsx("span",{className:"flex-1 truncate font-mono text-muted",children:r}),d?e.jsx(h,{tone:"ok",children:"Primary"}):e.jsxs("button",{type:"button",onClick:()=>t.setPrimaryDir.mutate(r),disabled:t.setPrimaryDir.isPending,title:"Downloads and imports will land in this folder",className:"flex items-center gap-1 rounded px-1.5 py-0.5 text-[11px] text-faint opacity-0 transition-opacity hover:text-ink focus:opacity-100 group-hover/dir:opacity-100",children:[e.jsx(f,{size:12}),"Set as primary"]}),e.jsx("button",{type:"button","aria-label":`Remove ${r}`,onClick:()=>t.removeDir.mutate(r),className:"rounded p-1 text-muted transition-colors hover:text-ink",children:e.jsx(x,{size:14})})]},r)})}),e.jsxs("div",{className:"flex items-center gap-2",children:[e.jsx(p,{value:o,onChange:r=>i(r.target.value),onKeyDown:r=>r.key==="Enter"&&l(),placeholder:"Paste an absolute folder path, e.g. D:\\\\models",className:"flex-1 font-mono text-[13px]"}),e.jsxs(u,{size:"sm",onClick:l,disabled:t.addDir.isPending||!o.trim(),children:[e.jsx(g,{size:14}),"Add folder"]})]}),n&&e.jsx("p",{className:"mt-2 text-[12px]",style:{color:"var(--err)"},children:n})]})}function h({children:a,tone:s}){const t=s==="ok"?"var(--ok)":s==="warn"?"var(--warn)":s==="err"?"var(--err)":s==="spec"?"var(--accent)":"var(--muted)";return e.jsx("span",{className:"rounded px-1.5 py-0.5 text-[10px] font-medium uppercase tracking-wide",style:{color:t,background:`color-mix(in srgb, ${t} 12%, transparent)`},children:a})}export{D as M};