llmxray 0.4.9 → 0.5.1

package/dist/assets/stream-handler-DHKXR5XU.js

@@ -1,4 +0,0 @@
-import{n as u,A as z}from"./index-CE4LM5vl.js";import{u as G}from"./session-store-Dt5jLOgV.js";import{a as R,b as E,u as J}from"./agent-store-CPoOtsp1.js";import{u as $}from"./metrics-store--HW_Yktg.js";import{u as q}from"./toolcall-store-C3ghhPw3.js";const B=1e6;function F(t,n,a,e,c){const s=(e.total_duration??0)/B,o=(e.load_duration??0)/B,i=(e.prompt_eval_duration??0)/B,l=(e.eval_duration??0)/B,p=e.prompt_eval_count??0,r=e.eval_count??0,h=l>0?r/l*1e3:0,f=i>0?p/i*1e3:0,d=c.length>0?c[0]:0;return{sessionId:t,model:n,startedAt:a,completedAt:Date.now(),ttftMs:d,totalDurationMs:s,loadDurationMs:o,promptEvalDurationMs:i,evalDurationMs:l,tokensPerSecond:h,promptTokensPerSecond:f,promptTokenCount:p,completionTokenCount:r,totalTokenCount:p+r,tokenLatencies:c}}class V{constructor(n){this.stepIndex=0,this.activeStep=null,this.inThinkBlock=!1,this.completedSteps=[],this.lastProcessedLineIndex=-1,this.thinkBlockProcessed=!1,this.enablePatternDetection=n?.patternDetection??!0}get isThinking(){return this.inThinkBlock}get thinkingContent(){return this.inThinkBlock&&this.activeStep?this.activeStep.content:""}processToken(n,a){if(!this.inThinkBlock&&!this.thinkBlockProcessed&&a.includes("<think>")){this.inThinkBlock=!0;const i=a.split("<think>").pop()??"";return this.activeStep={type:"thought",content:i,startTokenIndex:n.index,startTimestamp:n.timestamp},null}if(this.inThinkBlock&&a.includes("</think>")){if(this.inThinkBlock=!1,this.thinkBlockProcessed=!0,this.activeStep){const i=a.match(/<think>([\s\S]*?)<\/think>/);if(i?this.activeStep.content=i[1]?.trim()??this.activeStep.content.trim():this.activeStep.content=this.activeStep.content.replace(/<\/think>[\s\S]*$/,"").trim(),this.activeStep.content)return this.finalizeStep(n);this.activeStep=null}return null}if(this.inThinkBlock&&this.activeStep)return this.activeStep.content+=n.text,null;if(!this.enablePatternDetection)return null;const e=a.split(`
-`);if(e.length<2)return null;const c=e.length-2;if(c<=this.lastProcessedLineIndex)return null;this.lastProcessedLineIndex=c;const s=e[c]?.trim()??"";if(!s)return null;const o=K(s);if(o){let i=null;return this.activeStep&&(i=this.finalizeStep(n)),this.activeStep={type:o.type,content:o.content,startTokenIndex:n.index,startTimestamp:n.timestamp},i}return this.activeStep&&(this.activeStep.content+=n.text),null}finalize(){if(this.activeStep){const n={id:u(),index:this.stepIndex++,type:this.activeStep.type,content:this.activeStep.content.trim(),startTokenIndex:this.activeStep.startTokenIndex,endTokenIndex:this.activeStep.startTokenIndex,timestamp:this.activeStep.startTimestamp,durationMs:Date.now()-this.activeStep.startTimestamp};if(this.activeStep=null,n.content)return this.completedSteps.push(n),n}return null}finalizeStep(n){const a={id:u(),index:this.stepIndex++,type:this.activeStep.type,content:this.activeStep.content.trim(),startTokenIndex:this.activeStep.startTokenIndex,endTokenIndex:n.index,timestamp:this.activeStep.startTimestamp,durationMs:n.timestamp-this.activeStep.startTimestamp};return this.completedSteps.push(a),this.activeStep=null,a}}const H=[{regex:/^(?:Step\s+\d+[:.]\s*)(.+)/i,type:"thought"},{regex:/^(?:Thought[:.]\s*)(.+)/i,type:"thought"},{regex:/^(?:Observation[:.]\s*)(.+)/i,type:"observation"},{regex:/^(?:Action[:.]\s*)(.+)/i,type:"action"},{regex:/^(?:Conclusion[:.]\s*)(.+)/i,type:"conclusion"},{regex:/^(?:Therefore[,:.]\s*)(.+)/i,type:"conclusion"}];function K(t){const n=t.trim();for(const{regex:a,type:e}of H){const c=n.match(a);if(c)return{type:e,content:c[1]??n}}return null}async function I(t,n,a){const e=t.getReader(),c=new TextDecoder;let s="";try{for(;!a?.aborted;){const{done:o,value:i}=await e.read();if(o)break;s+=c.decode(i,{stream:!0});const l=s.split(`
-`);s=l.pop()??"";for(const p of l){const r=p.trim();if(!r||!r.startsWith("data: "))continue;const h=r.slice(6);if(h==="[DONE]")return;try{const f=JSON.parse(h);n(f)}catch{}}}}finally{e.releaseLock()}}async function W(t,n,a){const e=t.getReader(),c=new TextDecoder;let s="";try{for(;!a?.aborted;){const{done:o,value:i}=await e.read();if(o)break;s+=c.decode(i,{stream:!0});const l=s.split(`
-`);s=l.pop()??"";for(const p of l){const r=p.trim();if(r)try{const h=JSON.parse(r);n(h)}catch{}}}if(s.trim())try{const o=JSON.parse(s.trim());n(o)}catch{}}finally{e.releaseLock()}}async function tt(t,n,a){const e=G(),c=R(),s=$(),o=E(),i=J(),l=Date.now();let p=0,r=l,h="";const f=[],d=z(),b=e.sessionById(t),y=b?d.isThinkingModel(b.model):!1,x=new V({patternDetection:y});i.initGraph(t),i.addNode(t,{id:u(),type:"start",label:"Generation Start",sessionId:t,stepIndex:0,state:{},timestamp:l}),e.updateSessionStatus(t,"streaming"),await W(n,g=>{const M=Date.now(),D=M-r;h+=g.response;const C=j(D,f),v={id:u(),index:p++,text:g.response,timestamp:M,confidence:C,cumulativeText:h,interTokenLatencyMs:D};c.pushToken(t,v),e.appendOutput(t,g.response),f.push(D),r=M;const k=x.processToken(v,h);if(o.setThinking(t,x.isThinking,x.thinkingContent),k&&(o.addStep(t,k),i.addNode(t,{id:u(),type:"decision",label:`${k.type}: ${k.content.slice(0,40)}...`,sessionId:t,stepIndex:k.index,state:{reasoning:k.type},timestamp:k.timestamp})),g.done){x.finalize(),o.setThinking(t,!1,""),g.done_reason&&e.setDoneReason(t,g.done_reason);const m=e.sessionById(t);if(m){const T=F(t,m.model,l,g,f);s.recordMetrics(t,T),e.finalizeSession(t,T),s.recalculateAggregate()}i.addNode(t,{id:u(),type:"output",label:"Generation Complete",sessionId:t,stepIndex:p,state:{tokenCount:p},timestamp:Date.now()})}},a)}async function et(t,n,a){const e=G(),c=R(),s=$(),o=E(),i=q(),l=J(),p=Date.now();let r=0,h=p,f="",d="",b="",y=!1;const x=[],g=[],M=z(),D=e.sessionById(t),C=D?M.isThinkingModel(D.model):!1,v=new V({patternDetection:C});l.initGraph(t);const k=l.getGraph(t);return(!k||k.nodes.length===0)&&l.addNode(t,{id:u(),type:"start",label:"Chat Start",sessionId:t,stepIndex:0,state:{},timestamp:p}),e.updateSessionStatus(t,"streaming"),await W(n,m=>{const T=Date.now(),w=T-h,A=m.message.thinking;if(A){y=!0,d+=A,o.setThinking(t,!0,d),x.push(w),h=T;return}y&&d&&!m.done&&(o.setThinking(t,!1,d),d.trim()&&o.addStep(t,{id:u(),index:0,type:"thought",content:d.trim(),startTokenIndex:0,endTokenIndex:r,timestamp:T,durationMs:T-p}),b=d,d="");const N=m.message.content;f+=N;let P,L;m.logprobs&&m.logprobs.length>0?(L=m.logprobs[0].logprob,P=Math.exp(L)):P=j(w,x);const O={id:u(),index:r++,text:N,timestamp:T,confidence:P,logprob:L,cumulativeText:f,interTokenLatencyMs:w};if(c.pushToken(t,O),e.appendOutput(t,N),x.push(w),h=T,!y){const S=v.processToken(O,f);o.setThinking(t,v.isThinking,v.thinkingContent),S&&o.addStep(t,S)}if(m.message.tool_calls)for(const S of m.message.tool_calls){g.push(S);const _={id:u(),sessionId:t,index:i.getToolCalls(t).length,functionName:S.function.name,arguments:S.function.arguments,status:"pending",startedAt:T,tokenIndexStart:r};i.addToolCall(t,_),l.addNode(t,{id:u(),type:"tool_call",label:`Tool: ${S.function.name}`,sessionId:t,stepIndex:r,state:{tool:S.function.name,args:S.function.arguments},timestamp:T})}if(m.done&&(y&&d.trim()&&(b=d,o.addStep(t,{id:u(),index:0,type:"thought",content:d.trim(),startTokenIndex:0,endTokenIndex:r,timestamp:Date.now(),durationMs:Date.now()-p})),v.finalize(),o.setThinking(t,!1,y?b:""),m.done_reason&&e.setDoneReason(t,m.done_reason),g.length===0)){const S=e.sessionById(t);if(S){const _=F(t,S.model,p,m,x);s.recordMetrics(t,_),e.finalizeSession(t,_),s.recalculateAggregate()}l.addNode(t,{id:u(),type:"output",label:"Chat Complete",sessionId:t,stepIndex:r,state:{tokenCount:r},timestamp:Date.now()})}},a),{toolCalls:g}}function j(t,n){if(n.length<2)return .5;const a=[...n].sort((o,i)=>o-i),e=a[Math.floor(a.length/2)];if(e<=0)return .5;const c=t/e,s=Math.max(0,Math.min(1,1-(c-.5)/2));return Math.round(s*100)/100}export{tt as a,W as b,et as e,I as r};