-`i* `ddlZddlZddlmZddlmZddlmZddlm Z ddl m Z ddl m Z ddlmZd d lmZmZd d lmZ ddlmZddlZddlZn?#e$r7ed d ZedZedZYnwxYwdeee fde!efdZ"dddeee fde!ede#de$dzfdZ%deee fdedzde$dzfdZ&deee fdedzde$fdZ'deee fdede$fdZ(de!eee fdedzdedzde)e!eee fe*ffdZ+d d!d"d#d$ed%ed&e$dd#f d'Z,d(e d)e)e)eefd*fde fd+Z-d(e d,e)e)e)eefd*fe!eee ffd-e!ed.e#fd/Z.d0e dedzdedzd-e!ed.e#f d1Z/eGd2d3Z0d4e0fd5Z1d4ej2fd6Z3e4d7krJej5e0j68Z7e08e7e3e79dSdS)9N)ProcessPoolExecutor) dataclass)partial)Path)ClassVar) full_groupby)PlaceholderModule) DummyExecutor_json_load_bytes)sanitize_filename matplotlibpyplotpandasseabornrun_datakeysc|D]B}||dd|ddhD]}||vr ||ccSCdS)N_-)replace)rrkey candidates u/home/jaya/work/projects/VOICE-AGENT/VIET/agent-env/lib/python3.11/site-packages/vllm/benchmarks/sweep/plot_pareto.py_first_presentrsv++s{{344ckk#s6K6KL + +IH$$ ******% + 4T allow_zerorreturnc t||}|dS t|}n2#ttf$r}td|d|d||d}~wwxYw|s|dkrdS|S)Nz"Expected numeric value for one of z , but found z in run_data=r)rfloat TypeError ValueError)rrrvaluenumericexcs r _get_numericr'#s 8T * *E }t,, z " 2 2 2 2 2&. 2 2    'Q,,t Ns&AAAuser_count_varc|r|gng}|dgt||d}||St|dgdS)N request_rateFrmax_concurrent_requests)extendr')rr( candidates user_counts r_infer_user_countr/;sh&4;.!!J~&'''h uEEEJ #<"=% P P PPr gpu_count_varc|r|gng}t||d}|r|St|ddg}t|ddg}t|ddg}d }|r||z}|r||z}|r||z}|S) NFrtensor_parallel_sizetppipeline_parallel_sizeppdata_parallel_sizedpg?)r')rr0direct_candidatesdirect_gpu_counttp_sizepp_sizedp_size world_sizes r_infer_gpu_countr>Is,9@b#H.?ERRR 8&%EFFG8&:D%ABBGJg g g rthroughput_varcpt||g}|"td|dt||S)NzCannot find throughput metric z in run data. Available keys: )r'r#sorted)rr? throughputs r_get_throughputrC`s\h(899J 2^ 2 2%h// 2 2   rall_datac g}d}|D]b}t|d}t||}||dz }*t||}||z } ||z } |i|| | ||dc||fS)Nroutput_throughputr )tokens_per_usertokens_per_gpuuser_count_estimate gpu_count)rCr/r>append) rDr(r0preparedskipped_missing_usersrecordrBr.rJrGrHs r_prepare_recordsrOns H  $V-@AA &v~>>   !Q & ! $V];; $z1#i/  #2"0'1&         * **rg& .>)epsilondfz pd.DataFramex_coly_colrPc|||gddg}g}tj }|D];\}}||} | ||z kr%||t || }<|j|S)NF) ascending) sort_valuesmathinfiterrowsrKmaxloc) rQrRrSrP sorted_dffrontier_indicesbest_yidxrowy_vals r_pareto_frontierrbsu~%HHIhYF&&((((SE  FW$ $ $  # #C ( ( (''F 6" ##rfig_dir fig_group.cdg}|r|d|Dtd|dz}||z S)NPARETOc3*K|]\}}|d|VdS)=N).0kvs r z _get_fig_path..s066DAqZZAZZ666666rrz.png)r,r join)rcrdpartsfilenames r _get_fig_pathrqs[JE7 66I666666 %6!9::H X rfig_group_datalabel_bydry_runc |\}}t||}tdtdt|td||rtddStj|}|ddg}|jr tdtddSt|dd}| d}tj \} } tj |ddd d | d tj|ddd | d|r|D]j\} } g} |D]&}|| vr | |d| |'| r8| | d| dd| dk| d| d| dddd | | |tj| tdt3|dt3|dtddS)Nz[BEGIN FIGURE]zGroup: z Output file: z [END FIGURE]rGrH)subsetz3No data points available after filtering; skipping.z0.5g333333?zAll runs)dataxycoloralphaaxlabelozPareto frontier)rwrxrymarkerr|r}rh )fontsizez Tokens/s/userz Tokens/s/GPUTz--g?) linestyle linewidthr{zPlotted z points; Pareto frontier size: .)rqprintdictpd DataFrame from_recordsdropnaemptyrbrVpltsubplotssns scatterplotlineplotrYrKtextrn set_xlabel set_ylabelgrid tight_layoutsavefigcloselen)rcrrrsrtrdfig_datafig_pathrQfrontierfigr|rr` label_partsrs r _plot_figrs)IxWi00H  %DOO % %&&& $( $ $%%% n  " "8 , ,B ,.>? @ @B x CDDD n$57GHHH##$566HlnnGCO    L      ''))  FAsK < <#::&&#':':C':':;;; )*()IIk** MM/"""MM.!!!GGDDCsG;;;KKIcNNN K3r77KK3x==KKK .r output_dirc @|dz }d|dD}|std||ddt|||\}}|rt d|d |std t |d } t | d krtn t5} t| tt|||| ddddS#1swxYwYdS)Nparetoc6g|]}t|D]}|Sri)r )rjpathrs r zplot_pareto..sG (..  rz**/summary.jsonz/Did not find any parameter sweep results under T)parentsexist_ok)r(r0zSkipped zL runs without a user count (`max_concurrency` or `max_concurrent_requests`).zUNo data points with both throughput and user count available to plot Pareto frontier.ctSN)tuple)items rzplot_pareto..!s r)rr )rsrt) rglobr#mkdirrOrrrr rallmaprr) rr(r0rsrtrcraw_data prepared_datarM fig_groupsexecutors r plot_paretors8#G$$%677H YW:WWXXX MM$M...+;%#,,,(M(    @, @ @ @     '    J  OOq006I6K6K  x LL%#                        s :DDDceZdZUeed<edzed<edzed<eeed<eed<dZe eed<d Z e eed <e d e j fd Ze d e jfdZdS)SweepPlotParetoArgsrNr(r0rsrtr parser_namezYPlot Pareto frontier between tokens/s/user and tokens/s/GPU from parameter sweep results. parser_helpargsct|j}|std||jsgn|jd}|||j|j||jS)Nz!No parameter sweep results under ,rr(r0rsrt) r OUTPUT_DIRexistsr#rssplitr(r0rt)clsrrrss r from_cli_argsz!SweepPlotParetoArgs.from_cli_args@s$/**   "" OMMMNN N!]H22 0C0CC0H0Hs!.,L     rparserc&|dtdd|dtdd|dtdd |d td d |d dd|S)Nrresultsz3The directory containing the sweep results to plot.)typedefaulthelpz--user-count-varmax_concurrencyz_Result key that stores concurrent user count. Falls back to max_concurrent_requests if missing.z--gpu-count-varzResult key that stores GPU count. If not provided, falls back to num_gpus/gpu_count or tensor_parallel_size * pipeline_parallel_size.z --label-byzmax_concurrency,gpu_countzEComma-separated list of fields to annotate on Pareto frontier points.z --dry-run store_truez8If set, prints the figures to plot without drawing them.)actionr) add_argumentstr)rrs r add_cli_argsz SweepPlotParetoArgs.add_cli_argsPs F     %@      @      /      K     r)__name__ __module__ __qualname__r__annotations__rlistboolrrr classmethodargparse NamespacerArgumentParserrrirrrr2s$J:3i MMM!.K#... (#   !3    [  #("9###[###rrrc\t|j|j|j|j|jS)Nr)rrr(r0rsrtrs rrun_mainrws4 ?*(    rcTtt|dSr)rrrrs rmainrs% . .t 4 455555r__main__) description):rrWconcurrent.futuresr dataclassesr functoolsrpathlibrtypingrvllm.utils.collection_utilsrvllm.utils.import_utilsr plotr r utilsr matplotlib.pyplotrrrrrr ImportErrorplaceholder_attrrrobjectrrrr!r'r/r>rCrintrOrbrqrrrrrrrrrrr parse_argsrirrrs" 222222!!!!!!44444455555511111111$$$$$$'######'''  L ) ) : :8 D DC  8 $ $B  I & &CCC' T#v+.d3i 3; s)   T\ 0 Q3; Q$J Q T\ Q Q Q Q3;: . 3;       +4V $%+$J+: +  4S&[! "C '( ++++L $$$$ $ $  $  $$$$( U38_c)* H H%c3h 45tDf