+ iHj@Rt^RIHt^RIt^RIt^RIt^RIHtHtH t ]P!4t Rt RRlt RRltRR ltR R lt]P"R R l4tRRlt!RR4tR#)a'Prompt-reading progress from llama-server. llama-server reads a long prompt for minutes before it writes the first token. Asked with ``return_progress``, it streams a progress chunk after each batch it reads. Hermes uses those chunks as signs of life: the stall timers stay quiet while the server works, and the router notices a client hang-up within seconds (it only checks the client socket when it writes). Compression also leaves the time the server spends reading the prompt out of its total time limit (``conversation_compression_deadline``). ) annotationsN)AnyCallableOptionalc V^8dQhRRRR/#)kwargszdict[str, Any]return)formats"@/home/ubuntu/.hermes/hermes-agent/agent/local_prompt_progress.py __annotate__r s,,,N,cVPR4p\V\4'd \V4M/pVPRR4/VCRV/C#)zTCopy of chat-completions *kwargs* that asks llama-server for prompt progress chunks. extra_bodyreturn_progressT)get isinstancedict setdefault)rrmergeds& r with_prompt_progressrsKL)J!+J!=!=T* 2F '. +f +lF ++rc V^8dQhRRRR/#)rchunkrr zOptional[dict]r )r s"r r r s3>rc \VRR4p\V\4'dVPR4MRpVf(\V\4'dVPR4p\V\4'gR#\VRR4pVf(\V\4'dVPR4pT;'g.Fp\V\4'dVPR4M \VRR4p\V\4'dVPR4M \VRR4pV'dR#\FBp\V\4'dVPV4M \WgR4pV'gK@R# K V#)zThe ``prompt_progress`` payload of a progress-only chunk, else None. A progress chunk carries an empty assistant delta. A chunk with text, tool calls, reasoning or a finish reason is a real chunk even if a server attaches progress to it. model_extraNprompt_progresschoices finish_reasondelta)getattrrrr_CONTENT_FIELDS) rextraprogressrchoicefinishrfieldvalues & r rr s1 E=$ /E/9%/F/Fuyy*+DHJud3399./ h % %eY-G:eT22))I&--R-0:640H0HO,gV\^mosNt'1&$'?'? 7#WVU\^bEc $E(25$(?(?EIIe$WU[_E`Eu%   Orc$V^8dQhRRRRRR/#)rr#rkeystrr Optional[int]r )r s"r r r ;s!T rc`\W,4# \\\3dR#i;i)N)intKeyError TypeError ValueError)r#r)s&&r _countr1;s-8=!! i ,s --c V^8dQhRRRR/#)rr#rr r+r )r s"r r r Bs < AcV^8dQhRR/#)rr r<r )r s"r r r Ys++%F+rc$\\RR4#)zIThis thread's prompt-reading hook (carried into provider daemon threads).r;N)r r>r rr current_prompt_reading_hookrCYs 8VT **rc2]tRt^^tRtRRltRRltRtR#)PromptReadingClockaTime llama-server spends reading the prompt. Counted per progress chunk, and only when the server reports more tokens read than before: a server that keeps reporting without reading anything gets no extra time. Each stretch is the smaller of the wall time since the previous chunk and the server's own reading time, so time spent queued or loading the model before the reading starts doesn't count. cV^8dQhRR/#)rstartedfloatr )r s"r r PromptReadingClock.__annotate__gsrc <WnRVn^VnRVnR#)N_mark _processed _server_msseconds)selfrGs&&r __init__PromptReadingClock.__init__gs  rc V^8dQhRRRR/#)rr#rr boolr )r s"r r rImsTdrc \P!4pW P, Vuq0n\VR4\VR4rTVeW@P8:dR#VfTMWPP , R, pYEf VP MTuVnVn\ R\W644pV;PV, un\4pVe V!V4R#)zTCount the reading up to this progress chunk; False when it reports nothing new read.r5time_msFirLT) time monotonicrOr1rPrQr6r7rRrC) rSr#nowwallr5 server_msserverspentr;s && r tickPromptReadingClock.tickmsnn+Sj%h Y9   __ <"*__1LPT0T+4IZdoo`i(CT*+  *,   KrrNN)__name__ __module__ __qualname____firstlineno____doc__rTra__static_attributes__r rr rErE^s rrE)content tool_callsreasoning_content reasoningrefusal)rg __future__r contextlib threadingrZtypingrrrlocalr>r!rrr1r9contextmanagerr@rCrEr rr rtsl # ** ?? X,6 < !!+ r