1vLLM 0.14.0 대비 엔진에 추가된 환경변수

생성 2026-09-30 10:18 · 기준 커밋 v0.14.0 (b17039bcc) · 엔진 = vllm · vllm_cpu · verify · compressspec · config · 엔진 286개 (실험 하니스 194개는 맨 아래 접힘) · vLLM 원래 변수와 HF_·CUDA_·OMP_ 등 외부 변수는 제외
1.1
82
기능 · 실험에서 설정함
1.2
116
기능 · 기본값으로만 씀
1.3
27
로그·디버그 · 실험에서 설정함
1.4
61
로그·디버그 · 설정 안 함
1.5분류 기준 (펼치기)
정리 계획 문서 → 카테고리별: 지금 → 통합 · 이름 변경 · 삭제 ↗
1.6

1.7안 쓰는 환경변수 후보 (48)

코드는 아직 고치지 않았다. 근거 열의 커밋·파일로 확인한 뒤 지운다. A 설정만 남음 38 · C 폐기(옛 이름) 8 · C 폐기(LEGACY) 2
판정 기준 (펼치기)
#종류이름지금 읽는 곳 수위치(설정한 곳·읽는 파일)설정 횟수마지막 변경 커밋근거
1.7.1A 설정만 남음BREAKDOWN_DUMP_DIR0.env1001:d08c25e 2026-07-01 feat: compaction-threshold sweep harness + no-coldstart breakdown plott지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.2A 설정만 남음BREAKDOWN_TAG_COMPACT0jh_script/claude_script/27_g_curve_skipk.sh1001:d08c25e 2026-07-01 feat: compaction-threshold sweep harness + no-coldstart breakdown plott지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.3A 설정만 남음CLIENT_PY0jh_script/claude_script/rtx_pro6000_sweep/script/nebius_arm_run.sh jh_script/claude_script/rtx_pro6000_sweep/script/nebius_arm_run2.sh2지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.4A 설정만 남음COMPACT_NO_STACK0ysshin:.env1001:cd24516 2026-06-06 Chore: Add Benchmark and compaction debug script Add acceptance-sweep +지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.5A 설정만 남음COMPACT_RECOMPACT_ON_FULL0jh_script/claude_script/runners/run_aime2.sh1001:ce27b98 2026-09-19 docs: hand-off note for delta transfer, and bump both submodules지금 엔진(001 feature·ysshin dev)은 읽지 않음. g07 010 옛 트리에서만 읽음: vllm/vllm/cpu_verify/compaction_upload/scheduler_recompact.py
1.7.6A 설정만 남음COMPACT_RECOMPACT_RATIO0jh_script/claude_script/35_fullonly_vs_vanilla.sh scripts/saturate/lib/sweep_engine.sh ysshin:scripts/saturate/lib/sweep_engine.sh3001:af64fb0 2026-09-19 saturate: forward the delta-transfer knobs지금 엔진(001 feature·ysshin dev)은 읽지 않음. g07 010 옛 트리에서만 읽음: vllm/vllm/cpu_verify/compaction_upload/scheduler_recompact.py vllm_cpu/vllm/compaction/runner_diff_upload.py
1.7.7A 설정만 남음COMPACT_SCORE_RECOMPUTE0jh_script/claude_script/e2e_ab.sh jh_script/claude_script/e2e_ab2.sh jh_script/claude_script/e2e_node.sh jh_script/claude_script/e2e_on3.sh9지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.8A 설정만 남음DUMP_DIR0.env jh_script/claude_script/27_g_curve_skipk.sh2001:b310946 2026-09-05 top_p/top_k long-context verify: bf16+sparse logprob relay, stale-row c지금 엔진(001 feature·ysshin dev)은 읽지 않음. g07 010 옛 트리에서만 읽음: vllm/vllm/cpu_verify/cspec_debug.py
1.7.9A 설정만 남음EXP_DECODE_ATTN_INTERVAL0scripts/saturate/lib/sweep_engine.sh ysshin:scripts/saturate/lib/sweep_engine.sh2001:3367962 2026-09-08 saturate: do not track the editor backup지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.10A 설정만 남음HELD_ORDER_DIAG0jh_script/claude_script/runners/run_ab.sh1지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.11A 설정만 남음HI_GATE_MODEL0scripts/longcontext/Qwen3-30B-A3B-Thinking-2507/01_gates_tiny.sh scripts/longcontext/Qwen3-30B-A3B-Thinking-2507/02_gates_real.sh ysshin:scripts/longcontext/Qwen3-30B-A3B-Thinking-2507/01_gates_tiny.sh ysshin:scripts/longcontext/Qwen3-30B-A3B-Thinking-2507/02_gates_real.sh4001:5682a9a 2026-09-06 longcontext: Qwen3-30B-A3B (MoE) harness + cloud sm_120 5-way지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.12A 설정만 남음INDEXMAP_SHORTDRAFT_SHRINK_KEEP_HELD0jh_script/claude_script/runners/probe_util.sh jh_script/claude_script/runners/run_ab.sh jh_script/claude_script/runners/run_aime.sh jh_script/claude_script/runners/run_aime2.sh54지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.13A 설정만 남음KEEP_SPREAD_DIAG0jh_script/claude_script/runners/run_ks.sh1vllm_cpu:17132da6e 2026-09-19 feat(compaction): self-describing keep-map SHM and a correct held led지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.14A 설정만 남음KV_OFFLOAD_COALESCE0jh_script/claude_script/kv_coalesce_verify.sh jh_script/claude_script/kv_determinism_check.sh2지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.15A 설정만 남음MULTIBP_SINGLE0jh_script/claude_script/jh_q5q6.sh jh_script/claude_script/jh_q5x16.sh jh_script/claude_script/jh_q5xN.sh jh_script/claude_script/jh_revert_2tests.sh4001:80a874b 2026-06-20 refactor: extract config/common helpers + always-on CPU-verify profilin지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.16A 설정만 남음POOLGB_256K0jh_script/claude_script/runners/run_rs256.sh jh_script/claude_script/runners/run_rs256b.sh jh_script/claude_script/runners/run_rs3.sh3지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.17A 설정만 남음PRESERVE_DRAFT_ON_ALLACCEPT0.env1001:71771cc 2026-06-11 docs: debug knob registry (P0) + unified-levels and probe-status update지금 엔진(001 feature·ysshin dev)은 읽지 않음. g07 010 옛 트리에서만 읽음: vllm/tests/CPUVerify/test_engine_aftermath.py
1.7.18A 설정만 남음QSTART0jh_script/claude_script/runners/probe_util.sh jh_script/claude_script/runners/run_bn.sh jh_script/claude_script/runners/run_ff.sh jh_script/claude_script/runners/run_fx2_all.sh12001:0a6f35c 2026-09-19 feat(verify): score-server handshake, per-request arms, and the sweep d지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.19A 설정만 남음REQS_OVERRIDE0jh_script/claude_script/runners/probe_util.sh jh_script/claude_script/runners/retry_128k_full.sh jh_script/claude_script/runners/run_bn.sh jh_script/claude_script/runners/run_bprobe.sh43지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.20A 설정만 남음SSPEC_ACCEPT_CHECK0jh_script/claude_script/10_run_sparsespec_short_pod.sh jh_script/claude_script/18_dryrun_l40s.sh2SparseSpec 소스 계측용으로 보임. SparseSpec 은 클라우드 트리에 있어 이 검사로 확인 불가(노드 접속 불가). 우리 엔진·하니스·dev 트리는 읽지 않음
1.7.21A 설정만 남음SSPEC_BOUND_TRACE0jh_script/claude_script/_sspec_bound.sh jh_script/claude_script/_sspec_boundfix.sh jh_script/claude_script/_sspec_full.sh jh_script/claude_script/_sspec_valid.sh5SparseSpec 소스 계측용으로 보임. SparseSpec 은 클라우드 트리에 있어 이 검사로 확인 불가(노드 접속 불가). 우리 엔진·하니스·dev 트리는 읽지 않음
1.7.22A 설정만 남음SSPEC_BOUND_TRACE_PATH0jh_script/claude_script/32_sspec_repro.sh1SparseSpec 소스 계측용으로 보임. SparseSpec 은 클라우드 트리에 있어 이 검사로 확인 불가(노드 접속 불가). 우리 엔진·하니스·dev 트리는 읽지 않음
1.7.23A 설정만 남음SSPEC_ITER_TRACE0jh_script/claude_script/25_spec_cycle_probe.sh jh_script/claude_script/_sspec_trace.sh2001:1d9e75c 2026-09-21 chore: short context tps/batch size plot monitor utility updatedSparseSpec 소스 계측용으로 보임. SparseSpec 은 클라우드 트리에 있어 이 검사로 확인 불가(노드 접속 불가). 우리 엔진·하니스·dev 트리는 읽지 않음
1.7.24A 설정만 남음SSPEC_LAYER_TRACE0jh_script/claude_script/_sspec_plan.sh1SparseSpec 소스 계측용으로 보임. SparseSpec 은 클라우드 트리에 있어 이 검사로 확인 불가(노드 접속 불가). 우리 엔진·하니스·dev 트리는 읽지 않음
1.7.25A 설정만 남음SSPEC_LAYER_TRACE_PATH0jh_script/claude_script/32_sspec_repro.sh1SparseSpec 소스 계측용으로 보임. SparseSpec 은 클라우드 트리에 있어 이 검사로 확인 불가(노드 접속 불가). 우리 엔진·하니스·dev 트리는 읽지 않음
1.7.26A 설정만 남음SSPEC_LOGIT_TRACE0jh_script/claude_script/_sspec_budget.sh jh_script/claude_script/_sspec_dettopk.sh jh_script/claude_script/_sspec_layer.sh jh_script/claude_script/_sspec_logit.sh7SparseSpec 소스 계측용으로 보임. SparseSpec 은 클라우드 트리에 있어 이 검사로 확인 불가(노드 접속 불가). 우리 엔진·하니스·dev 트리는 읽지 않음
1.7.27A 설정만 남음SSPEC_LOGIT_TRACE_PATH0jh_script/claude_script/32_sspec_repro.sh1SparseSpec 소스 계측용으로 보임. SparseSpec 은 클라우드 트리에 있어 이 검사로 확인 불가(노드 접속 불가). 우리 엔진·하니스·dev 트리는 읽지 않음
1.7.28A 설정만 남음SSPEC_PLAN_TRACE0jh_script/claude_script/_sspec_bound.sh jh_script/claude_script/_sspec_boundfix.sh jh_script/claude_script/_sspec_fp.sh jh_script/claude_script/_sspec_full.sh10SparseSpec 소스 계측용으로 보임. SparseSpec 은 클라우드 트리에 있어 이 검사로 확인 불가(노드 접속 불가). 우리 엔진·하니스·dev 트리는 읽지 않음
1.7.29A 설정만 남음SSPEC_PLAN_TRACE_PATH0jh_script/claude_script/32_sspec_repro.sh1SparseSpec 소스 계측용으로 보임. SparseSpec 은 클라우드 트리에 있어 이 검사로 확인 불가(노드 접속 불가). 우리 엔진·하니스·dev 트리는 읽지 않음
1.7.30A 설정만 남음SSPEC_TIE_TRACE0jh_script/claude_script/_sspec_tie.sh1SparseSpec 소스 계측용으로 보임. SparseSpec 은 클라우드 트리에 있어 이 검사로 확인 불가(노드 접속 불가). 우리 엔진·하니스·dev 트리는 읽지 않음
1.7.31A 설정만 남음SSPEC_TIE_TRACE_PATH0jh_script/claude_script/32_sspec_repro.sh1SparseSpec 소스 계측용으로 보임. SparseSpec 은 클라우드 트리에 있어 이 검사로 확인 불가(노드 접속 불가). 우리 엔진·하니스·dev 트리는 읽지 않음
1.7.32A 설정만 남음SSPEC_TOPK_TRACE0jh_script/claude_script/_sspec_seal.sh1SparseSpec 소스 계측용으로 보임. SparseSpec 은 클라우드 트리에 있어 이 검사로 확인 불가(노드 접속 불가). 우리 엔진·하니스·dev 트리는 읽지 않음
1.7.33A 설정만 남음SSPEC_TOPK_TRACE_PATH0jh_script/claude_script/32_sspec_repro.sh1SparseSpec 소스 계측용으로 보임. SparseSpec 은 클라우드 트리에 있어 이 검사로 확인 불가(노드 접속 불가). 우리 엔진·하니스·dev 트리는 읽지 않음
1.7.34A 설정만 남음THRS0jh_script/claude_script/runners/run_n82_queue.sh1지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.35A 설정만 남음TP2DIR0jh_script/claude_script/fi_dir.sh1지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.36A 설정만 남음VERIFY_EOS_OVERRULED_KEEP_RELOAD0jh_script/claude_script/runners/probe_util.sh jh_script/claude_script/runners/run_ab.sh jh_script/claude_script/runners/run_aime.sh jh_script/claude_script/runners/run_aime2.sh56지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.37A 설정만 남음VLLM_HOME0.env1001:8a9d491 2026-02-06 feat: relative path bug fixed지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.38A 설정만 남음VLLM_TEST_HOME0.env1001:d5e50ad 2026-06-01 feat(figure): intro attention-cost + evaluation acceptance figures지금 엔진(001·ysshin dev)·하니스·third-party·g07 010 트리 어디서도 읽지 않음. 러너에만 남은 설정
1.7.39C 폐기(LEGACY)COMPACT_SCORE_EXPORT_LEGACY2vllm_cpu/vllm/v1/attention/backends/cpu_attn.py ysshin:vllm_cpu/vllm/v1/attention/backends/cpu_attn.py0001:2e472cc 2026-06-11 docs(cpu-verify): step op-breakdown + GQA head-reduce result이름에 LEGACY/DEPRECATED
1.7.40C 폐기(LEGACY)LOGPROB_READ_LEGACY_ANCHOR2vllm/vllm/cpu_verify/cpu_verify_engine.py ysshin:vllm/vllm/cpu_verify/cpu_verify_engine.py0vllm:e39c49308 2026-08-21 fix(cpu-verify): read logprob rows the writer actually booked이름에 LEGACY/DEPRECATED
1.7.41C 폐기(옛 이름)COMPRESSSPEC_DRAFT_ENV1compressspec/_tag.py:280001:fe3223c 2026-06-10 feat: compressspec CLI + env bootstrap & packaging새 이름 COMPRESS_SPEC_DRAFT_ENV 의 옛 이름. 읽으면 경고
1.7.42C 폐기(옛 이름)COMPRESSSPEC_NO_AUTOBUILD1compressspec/cli.py:730001:71771cc 2026-06-11 docs: debug knob registry (P0) + unified-levels and probe-status update새 이름 COMPRESS_SPEC_NO_AUTOBUILD 의 옛 이름. 읽으면 경고
1.7.43C 폐기(옛 이름)COMPRESSSPEC_NO_CLEAN1compressspec/cli.py:6290001:71771cc 2026-06-11 docs: debug knob registry (P0) + unified-levels and probe-status update새 이름 COMPRESS_SPEC_NO_CLEAN 의 옛 이름. 읽으면 경고
1.7.44C 폐기(옛 이름)COMPRESSSPEC_SKIP_CHECK1compressspec/_verify.py:930001:e8f85c0 2026-06-11 feat: hybrid sliding-window support새 이름 COMPRESS_SPEC_SKIP_CHECK 의 옛 이름. 읽으면 경고
1.7.45C 폐기(옛 이름)COMPRESSSPEC_SKIP_CHECK1compressspec/_verify.py:1820001:e8f85c0 2026-06-11 feat: hybrid sliding-window support새 이름 COMPRESS_SPEC_SKIP_CHECK 의 옛 이름. 읽으면 경고
1.7.46C 폐기(옛 이름)COMPRESSSPEC_TAG1compressspec/_tag.py:270001:fe3223c 2026-06-10 feat: compressspec CLI + env bootstrap & packaging새 이름 COMPRESS_SPEC_TAG 의 옛 이름. 읽으면 경고
1.7.47C 폐기(옛 이름)COMPRESSSPEC_TARGET1compressspec/_verify.py:1610001:fe3223c 2026-06-10 feat: compressspec CLI + env bootstrap & packaging새 이름 COMPRESS_SPEC_TARGET 의 옛 이름. 읽으면 경고
1.7.48C 폐기(옛 이름)COMPRESSSPEC_VERIFY_ENV1compressspec/_tag.py:290001:fe3223c 2026-06-10 feat: compressspec CLI + env bootstrap & packaging새 이름 COMPRESS_SPEC_VERIFY_ENV 의 옛 이름. 읽으면 경고

1.8trace·log 경량화 감사

Fable 감사 + 핵심 두 건(1·3행) 코드 직접 확인. 실행 시간은 측정하지 않음(추정). 전체 근거 file:line 은 trace_audit.md ↗
#항목근거조치안
1.8.1기본으로 도는 비용아무 스위치도 켜지 않아도 요청·라운드마다 도는 logger.info 6곳. vllm_cpu gpu_model_runner.py:3984 등은 로그 레벨 확인 전에 f-string 을 만들고 그 안에서 env 를 읽음. 기본 GPU sync 는 없음레벨 확인 뒤로 옮기거나 CS_TRACE≥1 로 묶기
1.8.2hot path env 읽기요청·라운드마다 os.environ.get 35~45회, 같은 _diag() 함수가 6개 파일에 복사됨 (keep_map_reader·cpu_verify_engine·gpu_recompact·scheduler_recompact·index_map·runner_diff_upload)import 시 한 번 읽어 캐시, _diag() 하나로
1.8.3켜면 GPU syncCOMPACT_DIFF_STAT=1(러너 19곳) 이 재압축 step 마다 torch.cuda.synchronize() 2회(gpu_recompact.py:21-23,79). DELTA_COMPACTION_LOG≥2 는 layer 마다 .item(). SCORE_DUMP_DIR·LOG_DELTA_SELECTION=2·KV_DUMP_* 도 step 마다 sync·파일sync 가 필요한 계측은 CS_TRACE=2 이상으로만
1.8.4breakdown 쓰기요청마다 한 줄 json.dumps + 줄 단위 버퍼 + 매 줄 flush(BREAKDOWN_FLUSH_EVERY=1). batch 16 추정 step 당 약 180 write (측정 안 함)배치 쓰기, flush 주기 늘리기
1.8.5중복라운드별 verify 통계 7곳, reload 시간 4곳, 재압축 이벤트 6곳, score dump 4곳, level 방식 5가지, 출력 경로 변수 14개출력 루트 CS_OUT 하나 + 고정 하위 폴더
1.8.6죽은 디버그 코드6월 hybrid-sw 버그(PR4_*, gemma3 DUMP_*, *_DBG 14개), 6월 pr4-item8(*_DBG 8개), 9월 parity 추적(KV_DUMP_*, *_LOG 9개) 등 — 러너가 켜지 않음삭제 후보(도입 커밋 기록 있음)
통일안 (펼치기)

2기능 · 실험에서 설정함 (82)

실제로 동작을 바꾸고, 러너 스크립트나 .env 가 값을 준다. 논문 실험 재현에 필요한 것
#이름읽는 곳읽는 횟수기본값러너가 준 값설정한 곳첫 위치
2.1ARMvllm+vllm_cpu6$1 | $2 | $ARM | $TAGjh_scriptvllm/vllm/cpu_verify/channels/temp_buffer.py:575
2.2ATTN_KV_RETAIN_DIVvllm_cpu28 | {rd}ysshin:scriptsvllm_cpu/csrc/cpu/cpu_attn_impl.hpp:1631
2.3ATTN_TASK_BALANCEvllm_cpu2scripts ysshin:scriptsvllm_cpu/csrc/cpu/cpu_attn_impl.hpp:484
2.4BLOCK_SIZEcompressspec+config+scripts+test+verify13"16" | "32"$BS | $ST_BLOCK | 16 | 32.env scripts test ysshin:scriptsverify/profile_cpu_forward.py:33
2.5COMPACT_ACT_RESERVE_GIBvllm1"1.0"scripts ysshin:scriptsvllm/vllm/cpu_verify/compaction_upload/gpu_worker_sizing.py:19
2.6COMPACT_DIFF_UPLOADvllm_cpu2"1"$DU | ${DU:-1} | 0 | 0)jh_script scripts ysshin:scriptsvllm_cpu/vllm/v1/worker/gpu_model_runner.py:5056
2.7COMPACT_DISABLEvllm_cpu1"0"scripts ysshin:scriptsvllm_cpu/vllm/v1/worker/gpu_model_runner.py:4743
2.8COMPACT_GATE_SCOREcompressspec+vllm_cpu2"1"$GATE | 0 | 0, | 0``.env jh_script scripts testvllm_cpu/vllm/v1/worker/gpu_model_runner.py:4751
2.9COMPACT_GATE_THRESHOLDvllm_cpu1"0.8"scriptsvllm_cpu/vllm/v1/worker/gpu_model_runner.py:4892
2.10COMPACT_INDEX_MAP_MAXvllm_cpu1"512"scripts ysshin:scriptsvllm_cpu/vllm/compaction/index_map.py:53
2.11COMPACT_KEEPMAP_SLACKvllm_cpu1"0"0jh_scriptvllm_cpu/vllm/compaction/runner_diff_upload.py:317
2.12COMPACT_MAX_GPUBLK_UTILvllm1"0.9"$CAP | 0 | 0, | 0.9jh_script scripts ysshin:scriptsvllm/vllm/cpu_verify/compaction_upload/scheduler_recompact.py:273
2.13COMPACT_METHODverify+vllm+vllm_cpu6"?" | "block" | "token"$METHOD | blockjh_script testvllm/vllm/v1/kv_offload/cpu.py:78
2.14COMPACT_NATIVEverify+vllm_cpu3"0" | "1"1.env ysshin:.envvllm_cpu/vllm/v1/worker/cpu_model_runner.py:68
2.15COMPACT_SCORE_FORCE_OFFvllm_cpu1"0"1jh_scriptvllm_cpu/vllm/v1/attention/backends/cpu_attn.py:51
2.16COMPACT_THRESHOLDcompressspec+vllm_cpu3"0.5" | "0.95"$THR | 0.5 | 0.5, | 0.6.env jh_script scripts testvllm_cpu/vllm/v1/worker/gpu_model_runner.py:5226
2.17COMPACT_TOKEN_NUM_RECENTvllm_cpu1"32"32testvllm_cpu/vllm/v1/worker/gpu_model_runner.py:5244
2.18COMPACT_TOKEN_NUM_SINKvllm_cpu1"32"32testvllm_cpu/vllm/v1/worker/gpu_model_runner.py:5243
2.19COMPACT_UPLOAD_PENDverify+vllm_cpu5"1"${LC_UPLOAD_PEND:-1}scripts ysshin:scriptsvllm_cpu/vllm/compaction/runner_diff_upload.py:76
2.20COMPRESS_SPEC_CYCLE_ANNOTATEtest+vllm3"0"1):testvllm/vllm/cpu_verify/cpu_verify_engine.py:3705
2.21COMPRESS_SPEC_DRAFT_ENVcompressspec1"COMPRESSSPEC_DRAFT_ENV"vllmjh_scriptcompressspec/_tag.py:28
2.22COMPRESS_SPEC_VERIFY_ENVcompressspec1"COMPRESSSPEC_VERIFY_ENV"verifyjh_scriptcompressspec/_tag.py:29
2.23COMPRESSION_SPEC_DECODING_INTERVALconfig+test+vllm13"0" | "1024" | "2048""1024" | "2048" | "64" | <int>testvllm/vllm/v1/attention/backends/flashinfer.py:600
2.24COMPRESSION_SPEC_MAX_TOTAL_CONTEXTvllm2scripts ysshin:scriptsvllm/vllm/v1/core/sched/scheduler.py:323
2.25COMPRESSION_SPEC_SELECTED_TOKENSconfig+scripts+test+vllm18"128" | "2048" | "512""64" | "8192" | <int> | os.environ.get(scripts test ysshin:scriptsvllm/vllm/config/compression_spec_config.py:81
2.26COMPSPEC_DECODE_ATTN_INTERVALscripts+test+vllm4os.environ.get(scripts test ysshin:scriptsvllm/vllm/v1/attention/backends/flashinfer.py:599
2.27COMPSPEC_PREEMPT_OFFLOADvllm3"0"scripts ysshin:scriptsvllm/vllm/cpu_verify/compaction_upload/manager.py:502
2.28COMPSPEC_RESUME_NEED_GATEvllm1"0"scripts ysshin:scriptsvllm/vllm/cpu_verify/cpu_verify_engine.py:399
2.29CPU_OFFLOAD_POOL_BLOCKSvllm+vllm_cpu4""$_kpool | 35000) | Kscripts test ysshin:scriptsvllm/vllm/cpu_verify/cpu_verify_engine.py:2851
2.30CPU_VERIFY_URLtest+vllm4"http://localhost:8000"http://localhost: | http://localhost:$ST_PORT | http://localhost:18000jh_script scripts ysshin:scriptsvllm/vllm/cpu_verify/_compaction_upload_dispatcher.py:96
2.31DRAFT_DISPATCH_FULLvllm1"0"${DRAFT_DISPATCH_FULL:-0}jh_scriptvllm/vllm/cpu_verify/channels/temp_buffer.py:700
2.32DYN_FORCE_OUTvllm1"0"$LC_DYN_FORCE_OUTscripts ysshin:scriptsvllm/vllm/cpu_verify/dyn_router.py:61
2.33DYN_ROUTERscripts+test+vllm3"0"1 | 1)scripts ysshin:scriptsvllm/vllm/cpu_verify/dyn_router.py:51
2.34DYN_ROUTER_HEADROOM_PCTtest+vllm2"0.10"$LC_DYN_HEADROOM_PCTscripts ysshin:scriptsvllm/vllm/cpu_verify/dyn_router.py:60
2.35EXP_DTYPEcompressspec+config+scripts+test+verify14"bf16" | "fp16""bf16" | $DT | $ST_DTYPE | bf16.env jh_script scripts testverify/profile_cpu_forward.py:34
2.36EXP_SELECTED_TOKENScompressspec+scripts+test+vllm_cpu8"128" | "2048"128``scripts test ysshin:scriptsvllm_cpu/vllm/v1/worker/gpu_model_runner.py:4080
2.37LAYER_SERIAL_CHUNKEDvllm4"0" | False1 | 1)jh_script scripts ysshin:scriptsvllm/vllm/config/scheduler.py:307
2.38LOGPROB_POOL_SIZEconfig+dump_debugger+verify6"480" | "8320""32" | 4160 | 4160) | 8320.env jh_scriptverify/test_logprob_stamp.py:31
2.39LOGPROB_VOCAB_SIZEverify4151936jh_script scripts ysshin:scriptsverify/test_logprob_stamp.py:30
2.40LS_CHUNK_TILEvllm1"8192"2048 | 8192jh_script scripts ysshin:scriptsvllm/vllm/v1/worker/gpu_model_runner.py:6486
2.41LS_PIPELINE_RELOADvllm1"1"1scripts ysshin:scriptsvllm/vllm/v1/attention/backends/flashinfer.py:2041
2.42LS_RELOAD_MEMCPYvllm1"1"1scripts ysshin:scriptsvllm/vllm/v1/attention/backends/flashinfer.py:2093
2.43LS_STREAM_STAGEvllm1m1scripts ysshin:scriptsvllm/vllm/v1/attention/backends/flashinfer.py:91
2.44MAX_NUM_SEQStest+vllm_cpu4"0" | "2"ysshin:scriptsvllm_cpu/vllm/compaction/keep_map_shm.py:88
2.45MODEL_DIRbreakdown_debugger+compressspec+config+dump_debugger+scripts+test28"Qwen/Qwen3-8B" | "meta-llama/Llama-3.1-8B-Instruct" | default_model$R/models/Qwen/Qwen3-4B-Thinki | ${VLLM_TEST_HOME}/models/AmanP | ${VLLM_TEST_HOME}/models/googl | ${VLLM_TEST_HOME}/models/meta-.env jh_script scripts testcompressspec/cli.py:274
2.46MULTIBP_MAX_MODEL_LENconfig1"25300""2048" | $MML | 4096 | 45056jh_scriptconfig/connector_config.py:53
2.47MULTIBP_MAX_NUM_SEQSconfig+dump_debugger2"10""2" | 100 | 16 | 2.env jh_scriptconfig/connector_config.py:52
2.48MULTIBP_QUESTIONconfig15jh_scriptconfig/common_util.py:27
2.49MULTIBP_REPEATconfig1"1"$REP | 1 | 16jh_scriptconfig/common_util.py:29
2.50NUM_GPU_BLOCKSdump_debugger+test+vllm_cpu9"0"$LC_NUM_GPU_BLOCKS | ${LC_NUM_GPU_BLOCKS:-5500} | 5500 | <n>scripts test ysshin:scriptsvllm_cpu/vllm/compaction/runner_diff_upload.py:172
2.51NUMA_AWARE_VERIFY_NODESverify+vllm+vllm_cpu7"4"$nnscripts ysshin:scriptsvllm/vllm/cpu_verify/channels/naive_copy.py:419
2.52NUMA_FIRST_TOUCHvllm+vllm_cpu3"0"scripts ysshin:scriptsvllm/vllm/v1/kv_offload/worker/cpu_gpu.py:1074
2.53NUMA_FT_NODEScompressspec+vllm_cpu2""scripts ysshin:scriptsvllm_cpu/vllm/v1/worker/gpu_model_runner.py:7510
2.54OUT_ROOTvllm_cpu1'/tmp'$OUT | $R/jh_script/claude_script/res | $REPO/dump_debugger/dump/sweepjh_scriptvllm_cpu/vllm/compaction/index_map.py:92
2.55PORTcompressspec+verify3"8000"$ST_PORT | 18000 | 8123 | 8155jh_script scripts ysshin:scriptsverify/cpu_verify_server_score.py:602
2.56RS_TRUNCATEverify1"1"scripts ysshin:scriptsverify/utils.py:583
2.57RULER_BATCHconfig+dump_debugger+test12"10" | "16"$B | 1 | 16 | 16,jh_script scripts test ysshin:scriptsconfig/common_util.py:38
2.58RULER_CTXconfig+test5"16384" | "4096"$CTX | 16384 | 2048 | 4096jh_script scripts test ysshin:scriptsconfig/common_util.py:49
2.59RULER_OFFSETconfig+dump_debugger+test6"0" | '0' | str(_do$((DOC-1)) | str(_it * _bsz)jh_script testconfig/common_util.py:68
2.60RULER_QDIRconfig+dump_debugger4_dq | _pd | os.path.join(REPO, "test", "RulerQuestions"$BAB | $CLOUD_ROOT/prompts/lb_v2_qwen | $LB | $NIAHjh_scriptconfig/common_util.py:67
2.61RULER_SEQconfig+verify4<token_len>testverify/profile_cpu_forward.py:242
2.62RULER_SUFFIXconfig1_ds_babilong | _longbench | _niahjh_scriptconfig/common_util.py:82
2.63RULER_TASKconfig+dump_debugger+verify4"cwe,vt" | "niah_single_1"all | cwe | cwe,vt.env jh_scriptverify/profile_cpu_forward.py:249
2.64SCORE_BLOCK_SIZEconfig+test+verify+vllm_cpu5"128"128.envvllm_cpu/vllm/v1/worker/score_shm_reader.py:28
2.65SCORE_NUM_LAYERSverify+vllm_cpu5"32" | str(_hf.num_hidden_layers32.env scripts ysshin:scriptsvllm_cpu/vllm/v1/worker/score_shm_reader.py:25
2.66SCORE_NUM_Q_HEADSverify+vllm_cpu5"32" | str(_hf.num_attention_heads32.env scripts ysshin:scriptsvllm_cpu/vllm/v1/worker/score_shm_reader.py:27
2.67SCORE_POOL_SIZEconfig+dump_debugger+test+verify+vllm_cpu6"100""8" | 4160 | 4160, | 8320.env jh_scriptvllm_cpu/vllm/v1/worker/score_shm_reader.py:26
2.68SEEDconfig1"0"0jh_scriptconfig/connector_config.py:51
2.69SHARED_CPU_POOL_GBcompressspec+test5"2" | "8""2" | 140 | 20 | 200.env jh_scriptcompressspec/cli.py:219
2.70SMALL_MODEL_DIRconfig+test4$M | $MODEL | Mjh_scriptconfig/connector_config.py:31
2.71SPECULATIVE_RUN_STEPconfig+vllm_cpu2"16""4" | 16.env jh_scriptvllm_cpu/vllm/compaction/index_map.py:59
2.72TASK_NAMEconfig1RULER | babilong | longbench_v2.env jh_scriptconfig/common_util.py:53
2.73VERIFY_DP_SIZEverify+vllm+vllm_cpu7"1"scripts ysshin:scriptsvllm/vllm/cpu_verify/channels/naive_copy.py:417
2.74VERIFY_ENFORCE_EAGERverify2"1"1jh_scriptverify/cpu_verify_server_score.py:111
2.75VERIFY_KV_SHM_RANKvllm_cpu1self.parallel_config.rank0 | 0)scripts ysshin:scriptsvllm_cpu/vllm/v1/worker/gpu_model_runner.py:7388
2.76VERIFY_MAX_MODEL_LENcompressspec+verify3"50000"$MML | 1044480 | 20000 | 30000jh_script scripts ysshin:scriptsverify/cpu_verify_server_score.py:139
2.77VERIFY_MAX_NUM_SEQSdump_debugger+verify4"1" | "32" | "4"${VSEQS:-$B} | 100 | 64.env jh_script testverify/profile_cpu_forward.py:278
2.78VERIFY_MODELbreakdown_debugger+compressspec+dump_debugger+verify7"meta-llama/Llama-3.1-8B-Instruct"$M | $MODEL | $R/models/Qwen/Qwen3-4B-Thinki | $R/models/meta-llama/Llama-3.1.env jh_script scripts ysshin:scriptsverify/profile_cpu_forward.py:32
2.79VERIFY_NUMA_INTERLEAVEverify2"1"0 | 0:scripts ysshin:scriptsverify/cpu_verify_server.py:35
2.80VERIFY_TOKENIZERdump_debugger+verify4MODEL_NAME | model$R/models/meta-llama/Llama-3.1jh_script scripts ysshin:scriptsverify/cpu_verify_server_score.py:46
2.81VLLM_ATTENTION_BACKENDconfig+scripts+test16"FLASHINFER""FLASHINFER" | 'FLASHINFER' | BACKEND | FLASHINFERjh_script scripts test ysshin:scriptsconfig/connector_config.py:71
2.82VLLM_SHM_NO_INTERLEAVEconfig11 | 1.jh_script scripts ysshin:scriptsconfig/common_util.py:188

3기능 · 기본값으로만 씀 (116)

동작을 바꾸지만 어떤 러너도 값을 주지 않는다. 기본값이 곧 운영값
#이름읽는 곳읽는 횟수기본값러너가 준 값설정한 곳첫 위치
3.1ACCEPT_MODEjudge_debugger+verify2"standard"verify/utils.py:1306
3.2ATTN_BP_STREAM_CAPvllm1"64"vllm/vllm/distributed/kv_transfer/kv_connector/v1/attn_byproduct_handler/handler.py:291
3.3CACTUS_DELTAjudge_debugger+verify2"1.0"verify/utils.py:1309
3.4COMPACT_BLOCK_KEEP_RATIOvllm_cpu1"0.1"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:4075
3.5COMPACT_BLOCK_POLICYvllm_cpu1"mass"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:4024
3.6COMPACT_BLOCK_SIZEvllm1"32"vllm/vllm/cpu_verify/keep_map_reader.py:32
3.7COMPACT_FINISHED_RESENDvllm1"4"vllm/vllm/cpu_verify/cpu_verify_engine.py:4342
3.8COMPACT_FULL_COPY_GBvllm1"8"vllm/vllm/config/compression_spec_config.py:64
3.9COMPACT_GATE_AGGvllm_cpu1"min"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:4884
3.10COMPACT_HOLD_FULLvllm_cpu1""vllm_cpu/vllm/compaction/index_map.py:96
3.11COMPACT_KEEPMAP_SLOT_MULTvllm_cpu1"2"vllm_cpu/vllm/compaction/keep_map_shm.py:106
3.12COMPACT_SCORE_EXPORT_LEGACYvllm_cpu1"0"vllm_cpu/vllm/v1/attention/backends/cpu_attn.py:47
3.13COMPACT_SCORE_FP16vllm_cpu1"1"vllm_cpu/vllm/v1/attention/backends/cpu_attn.py:65
3.14COMPACT_SCORE_NO_REDUCEvllm_cpu1"0"vllm_cpu/vllm/v1/attention/backends/cpu_attn.py:59
3.15COMPACT_TEMP_BUFFER_GBvllm1"8"vllm/vllm/config/compression_spec_config.py:62
3.16COMPACT_TOKEN_GLOBAL_RANKvllm_cpu2"0"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:3737
3.17COMPACT_TOKEN_HEAD_AGGverify+vllm_cpu6"gqa_max" | 'gqa_max'vllm_cpu/vllm/v1/attention/backends/cpu_attn.py:60
3.18COMPACT_TOKEN_POOL_Kvllm_cpu2"7"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:3726
3.19COMPACT_TOKEN_POOL_TYPEvllm_cpu2"max"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:3727
3.20COMPACT_TOKEN_SCORE_ROWSverify+vllm_cpu4"draft_first_last"vllm_cpu/vllm/v1/attention/backends/cpu_attn.py:69
3.21COMPACTION_IDX_EVERYvllm_cpu1"1"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:4204
3.22COMPACTION_IDX_LAYERSvllm_cpu1"all"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:4207
3.23COMPRESS_SPEC_CPU_POOL_HOLDvllm2"1"vllm/vllm/cpu_verify/cpu_verify_engine.py:1182
3.24COMPRESS_SPEC_NO_AUTOBUILDcompressspec1"COMPRESSSPEC_NO_AUTOBUILD"compressspec/cli.py:73
3.25COMPRESS_SPEC_NO_CLEANcompressspec1"COMPRESSSPEC_NO_CLEAN"compressspec/cli.py:629
3.26COMPRESS_SPEC_NO_FREE_SWdump_debugger+vllm_cpu4"0"vllm_cpu/vllm/v1/core/single_type_kv_cache_manager.py:608
3.27COMPRESS_SPEC_PIN_FULLKVdump_debugger+vllm_cpu4"0"vllm_cpu/vllm/v1/core/kv_cache_manager.py:344
3.28COMPRESS_SPEC_SYNC_VERIFYvllm1"0"vllm/vllm/v1/core/sched/scheduler.py:1730
3.29COMPRESS_SPEC_TARGETcompressspec1"COMPRESSSPEC_TARGET"compressspec/_verify.py:161
3.30COMPSPEC_BONUS_TOKENverify1"0"verify/cpu_verify_server.py:290
3.31COMPSPEC_DRAFT_STALL_Svllm1"60"vllm/vllm/cpu_verify/cpu_verify_engine.py:514
3.32COMPSPEC_OFFLOAD_COVERvllm1"0"vllm/vllm/cpu_verify/cpu_verify_engine.py:2773
3.33COMPSPEC_RELOAD_STALL_Svllm1"120"vllm/vllm/cpu_verify/cpu_verify_engine.py:519
3.34COMPSPEC_RESUME_NEED_MARGINvllm1"0"vllm/vllm/cpu_verify/cpu_verify_engine.py:401
3.35COMPSPEC_RESUME_NEED_TIMEOUT_Svllm1"600"vllm/vllm/cpu_verify/cpu_verify_engine.py:403
3.36COMPSPEC_RUNAHEADvllm1""vllm/vllm/cpu_verify/cpu_verify_engine.py:424
3.37COMPSPEC_SKIP_WATCH_SECSvllm1"120"vllm/vllm/cpu_verify/cpu_verify_engine.py:546
3.38CP_BATCH_HEAD_REDUCEverify+vllm_cpu3"1"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:3865
3.39CPU_FWD_READ_HASHvllm_cpu1"0"vllm_cpu/vllm/v1/attention/backends/cpu_attn.py:442
3.40CPU_VERIFY_TIMEOUT_Svllm2"500"vllm/vllm/cpu_verify/_compaction_upload_dispatcher.py:99
3.41CS_DRAFT_HOLDvllm1"1"vllm/vllm/cpu_verify/cpu_verify_engine.py:2261
3.42CS_DRAFT_HOLD_MAXvllm1"64"vllm/vllm/cpu_verify/cpu_verify_engine.py:2264
3.43CS_LANDING_PRUNEvllm2"1"vllm/vllm/cpu_verify/cpu_verify_engine.py:2082
3.44DENSE_NO_COMPACTvllm_cpu1vllm_cpu/vllm/v1/worker/gpu_model_runner.py:4756
3.45DIFF_NATIVEvllm_cpu2"1"vllm_cpu/vllm/compaction/diff_plan.py:97
3.46DIFF_PAD_SHORT_FATALvllm_cpu1"0"vllm_cpu/vllm/compaction/diff_plan.py:148
3.47DIFF_PARTIAL_RELOADvllm1"1"vllm/vllm/cpu_verify/channels/temp_buffer.py:504
3.48DYN_ROUTER_HOLD_STEPSvllm1"500"vllm/vllm/cpu_verify/dyn_router.py:69
3.49FORCE_ATTN_ISAvllm+vllm_cpu2""vllm/vllm/v1/kv_offload/worker/cpu_gpu.py:248
3.50FSM_RUN_DIRbreakdown_debugger+vllm2""vllm/vllm/cpu_verify/_fsm_log.py:16
3.51GPU_CORESconfig1"8"config/common_util.py:313
3.52HEADINFER_DECODEvllm1Falsevllm/vllm/v1/attention/backends/headinfer_decode.py:39
3.53HEADINFER_DECODE_REALvllm1Falsevllm/vllm/v1/attention/backends/headinfer_decode.py:41
3.54HI_NTBvllm1"32"vllm/vllm/v1/attention/backends/headinfer_decode.py:148
3.55INIT_FILE_PATHvllm2vllm/vllm/cpu_verify/utils.py:664
3.56KEEPMAP_PAD_LAYERSvllm_cpu1"1"vllm_cpu/vllm/compaction/keep_map_shm.py:47
3.57KEEPMAP_STRICT_SLOTSvllm1"1"vllm/vllm/cpu_verify/keep_map_reader.py:30
3.58LAYER_SERIAL_BARRIERvllm1"stream"vllm/vllm/cpu_verify/cpu_verify_engine.py:4281
3.59LAYER_SERIAL_COVERAGE_BLOCKSvllm1vllm/vllm/v1/core/kv_cache_utils.py:871
3.60LAYER_SERIAL_Kvllm9"0"vllm/vllm/cpu_verify/channels/naive_copy.py:371
3.61LAYER_SERIAL_TAIL_BLOCKSvllm3"16"vllm/vllm/cpu_verify/cpu_verify_engine.py:1365
3.62LOGPROB_READ_LEGACY_ANCHORvllm1"0"vllm/vllm/cpu_verify/cpu_verify_engine.py:3478
3.63LOGPROB_READ_POS_OFFSETvllm1"0"vllm/vllm/cpu_verify/cpu_verify_engine.py:3481
3.64LOGPROB_SHM_NAMEverify3f"vllm_logprobs_shm{shm_suffix(verify/test_logprob_stamp.py:29
3.65LOGPROB_SPARSEverify+vllm5"0"vllm/vllm/cpu_verify/logprob_handler.py:25
3.66LOGPROB_SPARSE_CAPverify+vllm3"128"vllm/vllm/cpu_verify/logprob_handler.py:27
3.67LS_CHUNKED_VERIFYvllm1"0"vllm/vllm/v1/attention/backends/flashinfer.py:1878
3.68LS_STREAM_FREEvllm1mvllm/vllm/v1/attention/backends/flashinfer.py:93
3.69LS_STREAM_RELOADvllm1mvllm/vllm/v1/attention/backends/flashinfer.py:92
3.70LSE_SHM_NAMEvllm_cpu1f"vllm_rank_0_attn_lse{shm_suffix(vllm_cpu/vllm/v1/worker/score_shm_reader.py:24
3.71MAX_MODEL_LENtest+verify7"50000" | "8192"verify/profile_cpu_forward.py:36
3.72MAXPOS_OVERRIDEtest+verify7Noneverify/cpu_verify_server_score.py:142
3.73MINIFORGE_DIRcompressspec1str(Path.home(compressspec/cli.py:49
3.74MULTIBP_COMPRESSconfig1"streamingllm"config/connector_config.py:55
3.75OFFLOAD_PREFIX_CACHEvllm1"0"vllm/vllm/cpu_verify/channels/naive_copy.py:408
3.76PR4_DISABLE_COMPRESS_CALLvllm1"0"vllm/vllm/v1/attention/backends/flashinfer.py:2874
3.77READY_TIMEOUTcompressspec1"600"compressspec/cli.py:446
3.78RECOMPACT_SLOT_ALIGNvllm_cpu1"1"vllm_cpu/vllm/compaction/index_map.py:58
3.79REJECT_DISTRIB_PTverify1"0"verify/utils.py:1301
3.80REJECT_DISTRIB_PT_MAXjudge_debugger+verify2"2000"verify/utils.py:1302
3.81REJECT_DISTRIB_SOURCEverify1"server"verify/utils.py:1298
3.82REJECT_DISTRIB_TOPKverify1"8"verify/utils.py:1296
3.83RELAX_RESIDUALjudge_debugger+verify2"v1"verify/utils.py:1308
3.84RELAX_TAUjudge_debugger+verify2"0.5"verify/utils.py:1307
3.85RESERVE_SKIPPED_SW_BLOCKSvllm_cpu1"0"vllm_cpu/vllm/v1/core/single_type_kv_cache_manager.py:198
3.86REUSE_ENGINE_RIDverify1"0"verify/cpu_verify_server.py:505
3.87RS_ALLOW_UNSUPPORTED_SAMPLINGvllm1"0"vllm/vllm/cpu_verify/channels/temp_buffer.py:83
3.88RS_FAST_TRUNCverify1"1"verify/utils.py:620
3.89RS_FAST_TRUNC_MAX_Kverify1"4096"verify/utils.py:623
3.90RS_KTH_CHUNKSverify1"32"verify/utils.py:702
3.91RS_KTH_MIN_VOCABverify1"16384"verify/utils.py:703
3.92RS_Q_FLOORverify1"1e-6"verify/utils.py:874
3.93RS_ROW_MISMATCHverify2"fatal" | "reject"verify/utils.py:875
3.94RS_STALE_MAX_RATEverify1"0.02"verify/utils.py:608
3.95RS_STALE_MIN_POSverify1"200"verify/utils.py:609
3.96RS_TIE_EPSverify1"1e-3"verify/utils.py:599
3.97RS_TOPP_EPSverify1"1e-5"verify/utils.py:589
3.98RULER_DYNAMICconfig1config/common_util.py:54
3.99SCORE_SHM_NAMEvllm_cpu1f"vllm_rank_0_attn_score{shm_suffix(vllm_cpu/vllm/v1/worker/score_shm_reader.py:23
3.100SKIP_DUMMY_BATCHvllm_cpu1"0"vllm_cpu/vllm/v1/engine/core.py:1381
3.101SNAPKV_FUSE_RELOADvllm1mvllm/vllm/v1/attention/backends/flashinfer.py:96
3.102SNAPKV_KEY_CHUNKvllm1vllm/vllm/v1/attention/backends/flashinfer.py:94
3.103SNAPKV_WIN_CHUNKvllm1"8"vllm/vllm/v1/attention/backends/flashinfer.py:2477
3.104STREAMINGLLM_SINK_TOKENSconfig1"64"config/connector_config.py:56
3.105TAS_DELTAverify1"0.3"verify/utils.py:1311
3.106TAS_EPSverify1"0.09"verify/utils.py:1310
3.107UPLOAD_SKIP_FATALvllm2"0"vllm/vllm/cpu_verify/compaction_upload/manager.py:175
3.108VERIFY_BATCH_WINDOW_MSverify1"0"verify/cpu_verify_server.py:139
3.109VERIFY_DISABLE_HYBRID_KVCMverify1"0"verify/cpu_verify_server.py:232
3.110VERIFY_DISPATCH_WORKERSvllm1""vllm/vllm/cpu_verify/cpu_verify_engine.py:459
3.111VERIFY_KV_PREFAULTvllm_cpu1"0"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:2984
3.112VERIFY_KV_PREFAULT_ARM_IDLEvllm_cpu1"2.0"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:3001
3.113VERIFY_NUMA_LOCALverify2verify/cpu_verify_server.py:42
3.114VERIFY_NUM_SPEC_BLOCKSverify+vllm_cpu3"0"vllm_cpu/vllm/attention/layer.py:458
3.115VERIFY_PROMPT_LOGPROBSverify1"-1"verify/cpu_verify_server.py:118
3.116VLLM_SHM_SUFFIXvllm+vllm_cpu2vllm/vllm/shm_ns.py:13

4로그·디버그 · 실험에서 설정함 (27)

출력·계측용인데 러너가 켠다(breakdown 등)
#이름읽는 곳읽는 횟수기본값러너가 준 값설정한 곳첫 위치
4.1ATTN_BP_DEBUG_FILLconfig+test+vllm3"" | "1"0.envvllm/vllm/distributed/kv_transfer/kv_connector/v1/offloading_attn_bp_connector.py:336
4.2ATTN_BP_DUMP_SCORE_CHECKvllm1""0.envvllm/vllm/v1/worker/gpu_model_runner.py:6836
4.3ATTN_KV_DUMP_DIRtest+vllm4""${DUMP_DIR}/kvcache.envvllm/vllm/cpu_verify/debug_utils.py:98
4.4BENCH_BATCHconfig31.env jh_scriptconfig/common_util.py:38
4.5BENCH_CTXconfig+test4262144 | 32768 | _saved_ctx | _wctx.env jh_script testconfig/common_util.py:49
4.6BREAKDOWN_DIRbreakdown_debugger+test+verify+vllm+vllm_cpu13""$BD | $R/breakdown_debugger/dump/bre | $ROOT/breakdown_debugger/dump/ | ${BREAKDOWN_DUMP_DIR}/breakdow.env jh_scriptvllm/vllm/breakdown_logger.py:149
4.7BREAKDOWN_KERNEL_INLINEbreakdown_debugger+vllm_cpu3"0" | "1"0 | 1jh_scriptvllm_cpu/vllm/model_executor/models/_cpu_kernel_timer.py:18
4.8BREAKDOWN_KERNEL_PROFILE_MODEbreakdown_debugger+config+test+verify+vllm_cpu6"0" | "1"0 | 1.env jh_script testvllm_cpu/vllm/model_executor/models/_cpu_kernel_timer.py:17
4.9COMPACT_ACK_LOGvllm4"0"${COMPACT_ACK_LOG:-0}jh_scriptvllm/vllm/cpu_verify/cpu_verify_engine.py:2068
4.10COMPACT_DIFF_STATverify+vllm_cpu3"0"0 | 1jh_scriptvllm_cpu/vllm/v1/core/sched/scheduler.py:1321
4.11COMPRESS_SPEC_SKIP_CHECKcompressspec2"COMPRESSSPEC_SKIP_CHECK"1jh_scriptcompressspec/_verify.py:93
4.12COMPRESSION_SPEC_DEBUGtest+vllm14"true" | truetestvllm/vllm/v1/core/single_type_kv_cache_manager.py:81
4.13CPU_KV_HASH_DBGvllm_cpu1"0"1jh_scriptvllm_cpu/vllm/v1/worker/gpu_model_runner.py:5106
4.14DEBUG_METADATA_DUMP_DIRdump_debugger+verify+vllm_cpu3""$OUT/accept_meta | $R512/accept_meta | $ROOT/results/gptoss20b_e2e | $ROOT/results/gptoss_e2e.env jh_scriptvllm_cpu/vllm/v1/worker/gpu_model_runner.py:4195
4.15DELTA_COMPACTION_LOGvllm+vllm_cpu10"0"1.jh_script scripts ysshin:scriptsvllm/vllm/cpu_verify/compaction_trace.py:16
4.16DRAFT_DISPATCH_LOGvllm1"0"${DRAFT_DISPATCH_LOG:-0}jh_scriptvllm/vllm/cpu_verify/channels/temp_buffer.py:701
4.17DUMP_EXTEND_BEFORE_VERIFYvllm1"0"1jh_scriptvllm/vllm/v1/core/sched/scheduler.py:2303
4.18EXAMPLE_DIAG_QIDcompressspec+test2"1" | "DIAG_QID", "5"1testcompressspec/cli.py:254
4.19GPU_LOGcompressspec1str(log_dir / "gpu_client.log"testcompressspec/cli.py:450
4.20HELD_TRACEvllm+vllm_cpu2"0"0jh_scriptvllm/vllm/cpu_verify/cpu_verify_engine.py:1857
4.21LAST_VERIFY_LOGvllm2"0"${LAST_VERIFY_LOG:-0}jh_scriptvllm/vllm/cpu_verify/_tracker_registry.py:141
4.22LOG_DELTA_SELECTIONvllm+vllm_cpu9"0"${LOG_DELTA_SELECTION:-1}jh_scriptvllm/vllm/cpu_verify/channels/temp_buffer.py:325
4.23LOG_DIRcompressspec1str(ROOT / ".realmode-run"testcompressspec/cli.py:447
4.24LOGPROB_DUMP_DIRdump_debugger+verify2${DUMP_DIR}/logprobs.envverify/utils.py:85
4.25RECOMPACT_TRACEvllm1"0"1jh_scriptvllm/vllm/breakdown_logger.py:296
4.26SCORE_DUMP_DIRdump_debugger+verify+vllm7""${DUMP_DIR}/score.envvllm/vllm/cpu_verify/cpu_verify_engine.py:2480
4.27TOKEN_DUMP_JSONconfig1jh_scriptconfig/common_util.py:129

5로그·디버그 · 설정 안 함 (61)

출력·덤프·검증용. 필요할 때 손으로 켜는 것
#이름읽는 곳읽는 횟수기본값러너가 준 값설정한 곳첫 위치
5.1ALLOC_NEW_COMPUTED_DBGvllm_cpu1"0"vllm_cpu/vllm/v1/core/single_type_kv_cache_manager.py:244
5.2BREAKDOWN_FLUSH_EVERYvllm1"1"vllm/vllm/breakdown_logger.py:88
5.3BREAKDOWN_GPU_EVENT_DEPTHvllm1"4"vllm/vllm/breakdown_logger.py:91
5.4BREAKDOWN_LIGHT_LATvllm1"0"vllm/vllm/breakdown_logger.py:105
5.5BREAKDOWN_PER_REQvllm1"0"vllm/vllm/breakdown_logger.py:89
5.6BREAKDOWN_RUN_DIRbreakdown_debugger+vllm2""vllm/vllm/breakdown_logger.py:147
5.7BREAKDOWN_SKIP_WARMUPbreakdown_debugger+vllm2"0" | "1"vllm/vllm/breakdown_logger.py:159
5.8COMPACT_DUMP_INDICESvllm_cpu1"0"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:4298
5.9COMPACT_DUMP_KVvllm_cpu1"0"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:4464
5.10COMPACT_HOLD_DUMPvllm_cpu1vllm_cpu/vllm/compaction/index_map.py:91
5.11COMPACTION_IDX_DUMPvllm_cpu1"0"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:4193
5.12COMPRESS_SPEC_DEBUGverify+vllm4"0"vllm/vllm/cpu_verify/cspec_debug.py:33
5.13COMPRESS_SPEC_TAGcompressspec1"COMPRESSSPEC_TAG"compressspec/_tag.py:27
5.14COMPRESS_SPEC_TRACEvllm1""vllm/vllm/cpu_verify/_trace.py:58
5.15COMPRESS_SPEC_TRACE_REQvllm1""vllm/vllm/cpu_verify/_trace.py:60
5.16CPU_ATTN_BUILD_DBGvllm_cpu2"0"vllm_cpu/vllm/v1/attention/backends/cpu_attn.py:195
5.17CPU_LOGcompressspec1str(log_dir / "cpu_server.log"compressspec/cli.py:449
5.18CPU_PROFILE_CHUNKverify1"0"verify/profile_cpu_forward.py:47
5.19CPU_PROFILE_CHUNK_ITERSverify1"12"verify/profile_cpu_forward.py:48
5.20CPU_PROFILE_COMPACT_RATEverify3"" | "0"verify/profile_cpu_forward.py:65
5.21CPU_PROFILE_CTX_TRUNCverify1"0"verify/profile_cpu_forward.py:292
5.22CPU_PROFILE_MAX_TOKENSverify1"32"verify/profile_cpu_forward.py:35
5.23CPU_PROFILE_PROMPT_LOGPROBSverify1verify/profile_cpu_forward.py:52
5.24CPU_VERIFY_DUMP_DIRvllm2vllm/vllm/cpu_verify/utils.py:650
5.25CS_PARITY_ASSERTvllm_cpu1"0"vllm_cpu/vllm/compaction/runner_diff_upload.py:109
5.26CS_STOP_LOGvllm1"0"vllm/vllm/v1/core/sched/utils.py:93
5.27DRAFT_TOKEN_LOGverify1"0"verify/cpu_verify_server_score.py:455
5.28DUMMY_BATCH_DBGvllm_cpu1"0"vllm_cpu/vllm/v1/engine/core.py:1373
5.29DUMP_FORWARD_OUTvllm1"0"vllm/vllm/model_executor/models/gemma3.py:373
5.30DUMP_LAST_HIDDENvllm1"0"vllm/vllm/model_executor/models/gemma3.py:532
5.31DUMP_LOGITS_OUTvllm1"0"vllm/vllm/model_executor/models/gemma3.py:550
5.32DUMP_SAMPLED_IDSvllm1"0"vllm/vllm/v1/worker/gpu_model_runner.py:4213
5.33DUMP_TRACKER_DISPATCHvllm1"0"vllm/vllm/cpu_verify/channels/temp_buffer.py:907
5.34FAKE_COMPACTION_VERIFYvllm_cpu1"0.0"vllm_cpu/vllm/v1/worker/gpu_model_runner.py:5525
5.35FAKE_NEEDS_KV_RELOADvllm1"0"vllm/vllm/cpu_verify/_fake_verify.py:247
5.36FREE_BLOCKS_DBGvllm_cpu1"0"vllm_cpu/vllm/v1/core/block_pool.py:514
5.37FWD_TENSOR_STATSvllm_cpu2"0"vllm_cpu/vllm/v1/attention/backends/cpu_attn.py:484
5.38INPUT_PROCESSOR_DBGvllm_cpu1"0"vllm_cpu/vllm/v1/engine/input_processor.py:612
5.39KEEPMAP_SLOT_LOGvllm1"1"vllm/vllm/cpu_verify/keep_map_reader.py:157
5.40KV_DST_PTR_DBGvllm1"0"vllm/vllm/v1/kv_offload/worker/cpu_gpu.py:262
5.41KV_DUMP_DIRvllm+vllm_cpu4"/tmp/qwg_kvdump"vllm/vllm/cpu_verify/channels/temp_buffer.py:572
5.42KV_DUMP_NCTvllm5vllm/vllm/cpu_verify/channels/temp_buffer.py:566
5.43KV_DUMP_SEQvllm_cpu2vllm_cpu/vllm/v1/worker/gpu_model_runner.py:5363
5.44KV_TRANSFER_DBGvllm1"0"vllm/vllm/distributed/kv_transfer/kv_connector/v1/offloading_naivecopy_connector.py:335
5.45LOGPROB_READ_DEBUGvllm1"0"vllm/vllm/cpu_verify/cpu_verify_engine.py:3517
5.46LS_PEAK_PROBEvllm1"0"vllm/vllm/v1/core/sched/scheduler.py:1914
5.47NAIVE_STORE_DBGvllm1"0"vllm/vllm/distributed/kv_transfer/kv_connector/v1/offloading_naivecopy_connector.py:471
5.48NUMA_VERIFY_DEBUGvllm1vllm/vllm/v1/kv_offload/backends/temp_cpu.py:80
5.49PAYLOAD_BUILD_DBGvllm1"0"vllm/vllm/cpu_verify/cpu_verify_engine.py:2899
5.50PR4_8_DBG_EXTENDEDvllm1"0"vllm/vllm/cpu_verify/channels/naive_copy.py:327
5.51PR4_SW_GUARD_DEBUGvllm1"0"vllm/vllm/v1/attention/backends/flashinfer.py:1841
5.52PRESERVE_LOGvllm2"0"vllm/vllm/cpu_verify/channels/temp_buffer.py:516
5.53RECOMPACT_FP_LOGvllm1"0"vllm/vllm/cpu_verify/compaction_upload/gpu_recompact.py:26
5.54REJECT_DISTRIB_DUMPverify1"0"verify/utils.py:1295
5.55RESERVE_DBGvllm_cpu1"0"vllm_cpu/vllm/v1/core/block_pool.py:386
5.56SAVE_KV_LAYER_DBGvllm1"0"vllm/vllm/distributed/kv_transfer/kv_connector/v1/offloading_naivecopy_connector.py:2
5.57SCORE_LOGconfig1"0"config/connector_config.py:185
5.58TEMPBUF_DBGvllm1"0"vllm/vllm/cpu_verify/channels/temp_buffer.py:666
5.59VERIFY_BATCH_LOGvllm_cpu1"0"vllm_cpu/vllm/v1/attention/backends/cpu_attn.py:231
5.60VERIFY_DEBUGverify2"0"verify/cpu_verify_server_score.py:114
5.61VERIFY_DEBUG_PAYLOADverify1"0"verify/cpu_verify_server.py:409

6실험 하니스 변수 (194)

펼치기: test · scripts · debugger 가 읽는 변수(RULER_ · SAT_ · SWEEP_ 등). 엔진 동작이 아니라 실험 구성용
#이름분류읽는 곳기본값러너가 준 값첫 위치
6.1AIME_DATA_PATHfunctionalscriptsscripts/saturate/lib/aime_data.py:33
6.2AIME_Nfunctionalscripts"960"scripts/saturate/lib/aime_data.py:179
6.3AIME_ONLY_PIDSfunctionalscripts""scripts/saturate/lib/aime_data.py:92
6.4AIME_REPEATfunctionalscripts+test"32" | 32scripts/saturate/lib/aime_data.py:180
6.5AIME_THINKINGfunctionalscripts+test"1"scripts/saturate/lib/draft_saturate_async.py:421
6.6AIME_UNIQ_PREFIXfunctionalscripts+test"0"scripts/saturate/lib/draft_saturate_async.py:424
6.7ASYNC_BIN_Sfunctionaldump_debugger+scripts+test"2.0"scripts/saturate/lib/draft_saturate_async.py:304
6.8ATTN_BP_DUMP_DIRlog/debugtest""test/test_score.py:241
6.9ATTN_BP_DUMP_TSlog/debugtest""test/test_score.py:242
6.10BATCHfunctionalscripts"1"$2 | $B | 1 | 4scripts/longcontext/lib/hi_lb.py:18
6.11BD_CAP_COLDfunctionalbreakdown_debugger"0"breakdown_debugger/breakdown_utils/timeline_html.py:367
6.12BLOCK_DEBUG_DUMP_DIRlog/debugtest""test/test_multi_bp.py:85
6.13BREAKDOWN_CPU_NAMElog/debugbreakdown_debugger""breakdown_debugger/breakdown_utils/latency_plot.py:79
6.14BREAKDOWN_GPU_NAMElog/debugbreakdown_debugger""breakdown_debugger/breakdown_utils/latency_plot.py:78
6.15BREAKDOWN_HW_NAMElog/debugbreakdown_debugger""breakdown_debugger/breakdown_util.py:27
6.16BREAKDOWN_KERNEL_FAMILYlog/debugbreakdown_debugger"llama"breakdown_debugger/breakdown_debugger.py:116
6.17BREAKDOWN_KERNEL_LAYERWISElog/debugbreakdown_debugger"0"breakdown_debugger/breakdown_utils/cpu_inline_kernels.py:174
6.18BREAKDOWN_PROFILE_STACKlog/debugbreakdown_debugger"0"1breakdown_debugger/breakdown_util.py:105
6.19BREAKDOWN_TRACE_WAIT_Slog/debugbreakdown_debugger"20"breakdown_debugger/breakdown_util.py:252
6.20CHAT_MAX_TOKENSfunctionaltest"256"test/chat.py:156
6.21CHAT_VERBOSElog/debugtest"0"test/chat.py:62
6.22CHUNKfunctionalscripts"16384"$CHUNKscripts/longcontext/lib/hi_lb.py:19
6.23COLLAPSE_Sfunctionaldump_debugger"10"15 | 999dump_debugger/sweep_test.py:46
6.24COMPRESS_METHODfunctionalscripts+test"snapkv" | "streamingllm"snapkvscripts/saturate/lib/draft_saturate_async.py:244
6.25COMPRESSSPEC_BENCH_CACHElog/debugtesttest/make_datasets.py:68
6.26CONDA_PREFIXfunctionaldump_debugger""/home2/qwg724/miniconda3/envs/dump_debugger/sweep_test.py:36
6.27CONDA_SHfunctionaldump_debugger$LC_CONDA | /home2/qwg724/miniconda3/etc/pdump_debugger/sweep_test.py:31
6.28CPU_POOL_GIBfunctionalscripts+test"16"$kvspacescripts/saturate/lib/draft_saturate_async.py:199
6.29CSVfunctionalscripts""$(find | $CSV | $OUT/hop_probe.csv | $OUT/microbench.csvscripts/longcontext/lib/hi_lb.py:23
6.30CTXfunctionaldump_debugger+scripts"16384"$((GEN | $1 | $CTX | $RULER_CTXscripts/longcontext/lib/hi_lb.py:17
6.31DEBUG_DUMP_TRACKERlog/debugtest"0"1 | 1) | 1.test/gpu_only_debug.py:96
6.32DEBUG_PATHlog/debugdump_debuggerdump_debugger/dump_debugger.py:175
6.33DIAG_QIDlog/debugtest"1"test/example_hybrid.py:158
6.34DISABLE_HYBRID_KVCMfunctionaltest"0"test/gpu_only_debug.py:272
6.35DISABLE_SLIDING_WINDOWfunctionaltest"0"test/gpu_only_debug.py:271
6.36DRAFT_BLOCK_SIZEfunctionalscripts+testos.environ.get("BLOCK_SIZE", "32"scripts/saturate/lib/draft_saturate_async.py:155
6.37DRAIN_STALL_Sfunctionaldump_debugger"240"1800dump_debugger/ruler_saturate_async_cs.py:165
6.38DRV_CPUSfunctionaldump_debugger"8-15,24-31,40-47,56-63"${DRV_CPUS:-16-19} | ${DRV_CPUS:-24-29,56-57} | 24-29,56-57 | nobinddump_debugger/saturate_run.py:17
6.39DRV_SANITIZERfunctionaldump_debuggerdump_debugger/sweep_test.py:245
6.40DYN_ROUTER_ADMIT_RESERVEfunctionalscripts+test"2048"$LC_DYN_ADMIT_RESERVEscripts/saturate/lib/draft_saturate_async.py:463
6.41ENFORCE_EAGERfunctionaldump_debugger+scripts+test"0"1scripts/saturate/lib/draft_saturate_async.py:172
6.42EXAMPLE_DIAG_BATCHlog/debugtest"DIAG_BATCH", "1"test/example.py:229
6.43EXP_DECODE_INTERVALfunctionalscripts+test"2048"scripts/saturate/lib/draft_saturate_async.py:195
6.44EXP_SCORE_INFLIGHTfunctionalscripts+testspec_steps * 6 + 32scripts/saturate/lib/draft_saturate_async.py:234
6.45EXP_SCORE_POOL_SIZEfunctionalscripts+testscore_pool_sizescripts/saturate/lib/draft_saturate_async.py:237
6.46EXP_SEEDfunctionaldump_debugger+scripts+test"0"$ST_SEED | ${EXP_SEED:-0} | 0 | 0.scripts/saturate/lib/draft_saturate_async.py:399
6.47EXP_SEED_PER_REQfunctionalscripts+test"0"scripts/saturate/lib/draft_saturate_async.py:407
6.48EXP_SPEC_STEPSfunctionalscripts+test"16"scripts/saturate/lib/draft_saturate_async.py:153
6.49EXP_TEMPfunctionaldump_debugger+scripts+test"0.0" | "0.5"$ST_TEMP | ${EXP_TEMP:-0.0} | 0.0 | 0.5scripts/saturate/lib/draft_saturate_async.py:392
6.50EXP_TEMPERATUREfunctionaltest"0.0"test/test_multi_bp.py:139
6.51EXP_TOP_Kfunctionaldump_debugger+scripts+test"-1" | "0" | "20"$ST_TOP_K | 20scripts/saturate/lib/draft_saturate_async.py:398
6.52EXP_TOP_Pfunctionaldump_debugger+scripts+test"0.95" | "1.0"$ST_TOP_P | 0.95scripts/saturate/lib/draft_saturate_async.py:397
6.53FAKE_ALL_ACCEPT_PROBlog/debugtest"0.0"<float>test/gpu_only_hybrid_debug.py:93
6.54FAKE_SELECTED_BLOCK_RATElog/debugtest"0.4"test/gpu_only_debug.py:254
6.55FSM_DEBUG_DIRlog/debugbreakdown_debugger""$ROOT/results/fsm_A_${TS} | $ROOT/results/fsm_cwe_vt_b4 | ${VLLM_TEST_HOME}/breakdownbreakdown_debugger/breakdown_util.py:55
6.56GPU_MEMORY_UTILIZATIONfunctionaltest"0.5"test/chat.py:101
6.57GPU_MEM_UTILfunctionaldump_debugger+test"0.8" | "0.9"${GPU_MEM_UTIL:-0.9} | 0.70 | 0.8 | 0.8.test/test_multi_bp.py:123
6.58HI_MODELfunctionalscriptsos.path.expanduser("~/models/Gradient-1048k"$R/models/ibm-granite/granite- | $R/models/mistralai/Magistral- | $R/models/skt/A.X-3.1scripts/longcontext/lib/hi_lb.py:16
6.59HI_SRCfunctionalscriptsos.path.join(_REPO, "third-party", "headinfer"/home2/qwg724/tmp/headinfer_miscripts/longcontext/lib/hi_lb.py:7
6.60HYBRID_COMPRESS_MODEfunctionaltest"streamingllm"test/chat.py:89
6.61HYBRID_MODEL_DIRfunctionaltest"google/gemma-3-1b-it"test/gpu_only_hybrid_debug.py:110
6.62IGNORE_EOSfunctionaldump_debugger+test"0" | "1"$EOS | ${IGNORE_EOS:-1} | 0 | 1test/test_multi_bp.py:144
6.63ITERSfunctionaldump_debugger+scripts+test"1" | "2"$IT | 1 | 10 | 5scripts/longcontext/lib/hi_lb.py:20
6.64LABELfunctionalscripts"hi"128K | AIME | hiscripts/longcontext/lib/hi_lb.py:23
6.65LB_DIRfunctionalscriptsos.path.join($D/ax_docs | $D/granite_docs | $DOCSscripts/longcontext/lib/hi_lb.py:21
6.66LC_CS_EXTRAfunctionalscripts""dir1[,dir2]:scripts/longcontext/lib/collect_natural.py:77
6.67LC_FONT_DIRfunctionalscripts""scripts/longcontext/lib/plot_long.py:22
6.68LC_LB_DOCSfunctionalscripts/home2/ysshin/exp_cs/neo_1m/lb | /pathscripts/longcontext/lib/extract_lb_docs.py:17
6.69LC_LB_MANIFESTfunctionalscripts""scripts/longcontext/lib/archive_results.py:40
6.70LC_N_E2Efunctionalscripts"256"scripts/longcontext/lib/collect.py:13
6.71LC_TITLEfunctionalscripts"Qwen3-30B-A3B-Thinking (1M window | "Qwen3-4B-Thinking"scripts/longcontext/lib/plot_long.py:151
6.72LONGBENCH_ROOTfunctionaltest_DEFAULT_ROOTtest/ruler/longbench_data.py:26
6.73LONGBENCH_TRUNCfunctionaltest"middle"headtest/ruler/longbench_v2_data.py:124
6.74LONGBENCH_V2_DIFFICULTYfunctionaltesttest/ruler/longbench_v2_data.py:66
6.75LONGBENCH_V2_IDSfunctionalscripts+test""",".join(ids)scripts/longcontext/lib/lb_manifest.py:241
6.76LONGBENCH_V2_LENGTHfunctionaltestlongtest/ruler/longbench_v2_data.py:60
6.77LONGBENCH_VERfunctionaltest"v1"v2test/ruler/longbench_gate.py:60
6.78MAX_CELL_Sfunctionaldump_debugger"360"5400 | 800dump_debugger/sweep_test.py:48
6.79MAX_GEN_TOKENSfunctionaldump_debugger+test"128"128 | 512 | 6 | 64test/test_multi_bp.py:144
6.80MAX_TOKENSfunctionaltest"32" | "512" | "64"<int>test/gpu_only_hybrid_debug.py:135
6.81MODEL_TYPEfunctionaldump_debugger+testos.environ.get("EXP_DTYPE", "bf16"test/test_multi_bp.py:12
6.82MULTIBP_MAX_BATCHED_TOKENSfunctionaldump_debuggerstr(cc.MAX_MODEL_LEN8192dump_debugger/sweep_test.py:81
6.83N_DECODEfunctionalscripts"16"$DEC | 16scripts/longcontext/lib/hi_lb.py:17
6.84N_E2Efunctionalscriptsstr(N256scripts/longcontext/lib/hi_lb.py:97
6.85NEO_DENSIFY_ABfunctionalscripts"0"1:scripts/longcontext/lib/neo_lb.py:117
6.86NEO_DENSIFY_PASTfunctionalscripts"1" if it % 2 else "0"scripts/longcontext/lib/neo_lb.py:118
6.87NEO_KV_BACKENDfunctionalscripts"cpu_pool"cpu_pool | gpuscripts/longcontext/lib/neo_lb.py:82
6.88NEO_MODEL_NAMEfunctionalscripts"llama3_8b"scripts/longcontext/lib/neo_lb.py:21
6.89NEO_OWNED_ABfunctionalscripts"0"1:scripts/longcontext/lib/neo_lb.py:111
6.90NEO_OWNED_CPU_POOLfunctionalscripts"1" if it % 2 else "0"scripts/longcontext/lib/neo_lb.py:112
6.91NEO_THREAD_SWEEPfunctionalscripts""scripts/longcontext/lib/neo_lb.py:101
6.92NSYS_CAPTUREfunctionaltesttest/ruler/longbench_gate.py:409
6.93NUM_CPU_BLOCKSfunctionaltest"64"test/gpu_only_debug.py:53
6.94OUTfunctionaldump_debugger"/home2/qwg724/001_CompressSpec/dump_debugger/run10/babilong | "/home2/qwg724/001_CompressSpec/dump_debugger/run10/longbenc | "/home2/qwg724/001_CompressSpec/test/GSK8K"$(ls | $AG/results/integration | $CLOUD_ROOT/run/${TAG}_$TS | $CS/jh_script/claude_script/redump_debugger/run10/gen_niah.py:19
6.95OUT_DUMPlog/debugdump_debugger""$OUT/$ARM/outputs.jsonl | $R/results/aime_tokens_${PHASEdump_debugger/ruler_saturate_async_cs.py:78
6.96PACPU_SOfunctionalscriptsos.path.join($NEO_SRC/pacpu/build/libpacpu- | /home2/qwg724/tmp/neo_ws/pacpuscripts/longcontext/lib/neo_lb.py:22
6.97PREFIX_CACHINGfunctionaldump_debugger"1"0dump_debugger/ruler_saturate_async_cs.py:72
6.98PROMPT_FILEfunctionaltest<path>test/gpu_only_debug.py:174
6.99PROMPT_FILE_RAWfunctionaltest"1"test/gpu_only_debug.py:179
6.100REALMODE_ACCEPT_TOLERANCEfunctionaltest"0.5"test/test_real_mode_e2e.py:44
6.101REALMODE_ELAPSED_TOLERANCE_SECfunctionaltest"20"test/test_real_mode_e2e.py:46
6.102REALMODE_EXPECT_ACCEPT_PCTfunctionaltest"100.0"test/test_real_mode_e2e.py:43
6.103REALMODE_EXPECT_ELAPSED_SECfunctionaltest"13"test/test_real_mode_e2e.py:45
6.104REALMODE_LOG_DIRlog/debugteststr(PROJECT_ROOT / ".realmode-run"test/test_real_mode_e2e.py:48
6.105REF_TOKENSfunctionaldump_debugger""dump_debugger/ruler_saturate_async_mix.py:91
6.106RELOAD_FORCE_SYNCfunctionaldump_debugger"0"dump_debugger/ruler_saturate_async_cs.py:38
6.107REP_MAXfunctionaldump_debugger"200"dump_debugger/ruler_saturate_async_cs.py:86
6.108REP_MODEfunctionaldump_debugger"shadow"dump_debugger/ruler_saturate_async_cs.py:90
6.109REP_NGRAMfunctionaldump_debugger"8"dump_debugger/ruler_saturate_async_cs.py:85
6.110ROUTER_RESERVEfunctionalscripts+test"32768"$LC_ROUTER_RESERVEscripts/saturate/lib/draft_saturate_async.py:466
6.111RULER_ARMSfunctionaldump_debugger""dump_debugger/ruler_saturate_async_mix.py:104
6.112RULER_CHATfunctionaltest"1"test/ruler/longbench_gate.py:155
6.113RULER_CTX_LABELfunctionaldump_debugger"16K"${LABEL:-$CTX}dump_debugger/ruler_saturate_async_cs.py:8
6.114RULER_DATASETfunctionaltest"ruler"test/ruler/ruler_saturate_async.py:380
6.115RULER_DUMP_LOGPROBSlog/debugtest"2"${LC_DUMP_LOGPROBS:-2}test/ruler/longbench_gate.py:371
6.116RULER_DUMP_LOGPROB_DIGITSlog/debugtest"4"test/ruler/longbench_gate.py:446
6.117RULER_DUMP_OUTlog/debugtest1test/ruler/longbench_gate.py:370
6.118RULER_DUMP_TOKENSlog/debugtest"0"1:test/ruler/ruler_saturate_async.py:442
6.119RULER_GPU_MEM_UTILfunctionalscripts+test"0.8" | "0.9"$GPU_MEM_UTIL | 0.8 | 0.85scripts/longcontext/lib/router.py:117
6.120RULER_HYBRID_KVCMfunctionaltesttest/ruler/ruler_saturate_async.py:159
6.121RULER_IGNORE_EOSfunctionaltest"0" | "1"0 | 1test/ruler/longbench_gate.py:365
6.122RULER_KV_SEQ_LENfunctionaltestctx + 1024test/ruler/ruler_saturate_async.py:193
6.123RULER_LOGPROBS_MODEfunctionaltestprocessed_logprobstest/ruler/longbench_gate.py:206
6.124RULER_MAX_BATCHED_TOKENSfunctionaldump_debugger+scripts+test"2048" | "8192" | max(max_model_len, 253008192scripts/longcontext/lib/router.py:115
6.125RULER_MAX_MODEL_LENfunctionaltestmax(4096, ctx + 1024test/ruler/longbench_gate.py:167
6.126RULER_MAX_NEW_TOKENSfunctionaldump_debugger+test"" | "512" | 512$OUT_TOKENS | none):test/ruler/longbench_gate.py:362
6.127RULER_MODEfunctionaltest"compress"compress | vanillatest/ruler/longbench_gate.py:149
6.128RULER_NO_TRUNCfunctionaltest"0"1 | 1). | 1:test/ruler/longbench_gate.py:351
6.129RULER_OUT_DIRfunctionaldump_debugger+test"." | _REPO$WORKtest/ruler/longbench_gate.py:168
6.130RULER_PREFIX_CACHINGfunctionaltest0test/ruler/longbench_gate.py:199
6.131RULER_Q_FILESfunctionaldump_debugger""dump_debugger/ruler_saturate_async_mix.py:53
6.132RULER_Q_LISTfunctionaldump_debugger""dump_debugger/ruler_saturate_async_cs.py:46
6.133RULER_Q_STARTfunctionaldump_debugger"1"dump_debugger/ruler_saturate_async_cs.py:43
6.134RULER_REQSfunctionaldump_debugger+test"512" | _env_int("RULER_POOL", batch#test/ruler/ruler_saturate_async.py:371
6.135RULER_ROUTERfunctionaltest"0"1 | 1,test/ruler/longbench_gate.py:311
6.136RULER_SAMPLESfunctionaltestNone$SAMPLES | ${FIXED_SAMPLES:-$B} | 2*B | 32test/ruler/longbench_gate.py:165
6.137RULER_SORT_BY_LENfunctionaltest1 | 1:test/ruler/longbench_gate.py:342
6.138RULER_TAGlog/debugdump_debugger+test"" | f"b{B}"$LC_DUMP_OUT | $tagtest/ruler/longbench_gate.py:173
6.139RULER_TASKSfunctionaltest"all"$TASK | all | code_repo | niah_single_1,qa_1test/ruler/longbench_gate.py:163
6.140RULER_WARMUP_TOKENSfunctionaltestfixed_outtest/ruler/ruler_saturate_async.py:416
6.141RUN_NUMAfunctionaldump_debugger"0,1,2,3"dump_debugger/saturate_run.py:18
6.142SAT_BATCHfunctionalscripts$B | batch(=max_num_seqs),scripts/saturate/lib/draft_saturate_async.py:152
6.143SAT_CTXfunctionalscripts"16384"$ST_AIME_MAX_PROMPTscripts/saturate/lib/draft_saturate_async.py:151
6.144SAT_DUMP_TOKENSlog/debugscripts"0"1:scripts/saturate/lib/draft_saturate_async.py:491
6.145SAT_GPU_MEM_UTILfunctionalscripts"0.8"scripts/saturate/lib/draft_saturate_async.py:154
6.146SAT_HYBRID_KVCMfunctionalscripts0 | 1scripts/saturate/lib/draft_saturate_async.py:174
6.147SAT_IGNORE_EOSfunctionalscripts"1"$ST_IGNORE_EOSscripts/saturate/lib/draft_saturate_async.py:402
6.148SAT_KV_SEQ_LENfunctionalscriptsctx + 1024scripts/saturate/lib/draft_saturate_async.py:208
6.149SAT_MAX_BATCHED_TOKENSfunctionalscriptsmax(max_model_len, 25300scripts/saturate/lib/draft_saturate_async.py:165
6.150SAT_MAX_MODEL_LENfunctionalscriptsmax(4096, ctx + 1024scripts/saturate/lib/draft_saturate_async.py:159
6.151SAT_MAX_NEW_TOKENSfunctionalscripts512$ST_OUT_TOKENSscripts/saturate/lib/draft_saturate_async.py:401
6.152SAT_MODEfunctionalscripts"compress"${ST_MODE:-compress}scripts/saturate/lib/draft_saturate_async.py:181
6.153SAT_OUT_DIRfunctionalscripts_REPO$WORKscripts/saturate/lib/draft_saturate_async.py:390
6.154SAT_OUT_NAMEfunctionaldump_debugger"saturate"dump_debugger/saturate_run.py:30
6.155SAT_PREFIX_CACHINGfunctionalscripts"0"1.scripts/saturate/lib/draft_saturate_async.py:168
6.156SAT_REQSfunctionalscripts_env_int("SAT_POOL", batch# | $_reqsscripts/saturate/lib/draft_saturate_async.py:389
6.157SAT_ROUTERfunctionalscripts"0"scripts/saturate/lib/draft_saturate_async.py:454
6.158SAT_TAGlog/debugscripts""$tagscripts/saturate/lib/draft_saturate_async.py:391
6.159SAT_UNIQ_PREFIXfunctionalscripts"0"1scripts/longcontext/lib/hi_lb.py:47
6.160SAT_WARMUP_TOKENSfunctionalscriptsfixed_outscripts/saturate/lib/draft_saturate_async.py:430
6.161SEQUENTIALfunctionaldump_debugger"0"1dump_debugger/ruler_saturate_async_cs.py:22
6.162SERVER_ENVfunctionaldump_debugger"verify_L40S"${SERVER_ENV:-verify}dump_debugger/saturate_run.py:53
6.163SNAPKV_AGGfunctionalscripts+test"mean"scripts/saturate/lib/draft_saturate_async.py:249
6.164SNAPKV_KERNELfunctionalscripts+test"7"scripts/saturate/lib/draft_saturate_async.py:248
6.165SNAPKV_POOLINGfunctionalscripts+test"maxpool"scripts/saturate/lib/draft_saturate_async.py:250
6.166SNAPKV_SINK_BLOCKSfunctionalscripts+test"0"scripts/saturate/lib/draft_saturate_async.py:251
6.167SNAPKV_WINDOWfunctionalscripts+test"32"scripts/saturate/lib/draft_saturate_async.py:247
6.168SPLITfunctionaldump_debugger"test"dump_debugger/run10/gen_gsm8k.py:15
6.169SRV_BINDfunctionaldump_debugger"0-7,16-23,32-39,48-55"${SRV_BIND:-0-15} | ${SRV_BIND:-0-5,10-11,16-23,32 | 0-5,10-11,16-23,32-37,42-43,48 | 24-55dump_debugger/saturate_run.py:16
6.170STALL_Sfunctionaldump_debugger"180"400 | 480 | 500 | 5400dump_debugger/sweep_test.py:47
6.171START_ITERfunctionaldump_debugger"0"0 | 4dump_debugger/sweep_test.py:96
6.172STREAMINGLLM_SINKfunctionaltest"64"test/chat.py:91
6.173SWEEP_BATCHESfunctionaltest"1,2,4,8"test/throughput_sweep.py:24
6.174SWEEP_DRV_CPUSfunctionaldump_debugger""56-63 | 8-15,24-31,40-47,56-63dump_debugger/sweep_test.py:236
6.175SWEEP_DRV_OMPfunctionaldump_debuggerdump_debugger/sweep_test.py:275
6.176SWEEP_DUMP_TOKENSlog/debugdump_debuggerdump_debugger/sweep_test.py:113
6.177SWEEP_MAXTOKfunctionaltest"128"test/throughput_sweep.py:26
6.178SWEEP_OUTfunctionaltestos.path.join(test/throughput_sweep.py:27
6.179SWEEP_OUT_ROOTfunctionaldump_debugger$OUT | $R512dump_debugger/sweep_test.py:42
6.180SWEEP_SEQSfunctionaltest"4096,8192,16384"test/throughput_sweep.py:25
6.181SWEEP_SERVER_ENVfunctionaldump_debugger"verify_L40S"dump_debugger/sweep_test.py:43
6.182SWEEP_SRV_BINDfunctionaldump_debugger""0-55 | 0-7,16-23,32-39,48-55dump_debugger/sweep_test.py:227
6.183SWEEP_SRV_ISAfunctionaldump_debuggerdump_debugger/sweep_test.py:254
6.184SWEEP_SRV_OMPfunctionaldump_debuggerdump_debugger/sweep_test.py:253
6.185SWEEP_SRV_SPINfunctionaldump_debuggerdump_debugger/sweep_test.py:255
6.186SWEEP_TEMPERATUREfunctionaldump_debugger"0.0"dump_debugger/sweep_test.py:92
6.187TASKSfunctionaldump_debugger"qa2,qa3,qa4,qa5"dump_debugger/run10/gen_babilong.py:14
6.188VLLM_CPU_VERIFY_TRACKER_JSON_DIRfunctionaldump_debuggerdump_debugger/logprobs_debugger.py:30
6.189WAIT_READY_Sfunctionaldump_debugger"420"3600dump_debugger/saturate_run.py:57
6.190WARMUPfunctionalscripts+test"0" | "1"$W | 0 | 0으로 | 1scripts/longcontext/lib/hi_lb.py:20
6.191WARMUP_COUNTfunctionaltest"0"test/test_multi_bp.py:48
6.192WARMUP_CTXfunctionaltest""4096test/test_multi_bp.py:56
6.193WARMUP_MAX_TOKENSfunctionalteststr(sampling_params.max_tokens6test/test_multi_bp.py:152
6.194WINDOW_Sfunctionalbreakdown_debugger"0"breakdown_debugger/breakdown_debugger.py:169

7엔진 변수 접두어별 개수

접두어개수
COMPACT37
VERIFY19
COMPRESS15
CPU14
RS12
COMPSPEC11
BREAKDOWN9
LOGPROB9
LS8
RULER8
SCORE7
ATTN6
DUMP6
CS5
KV5
LAYER5
MULTIBP5
REJECT5
COMPRESSION4
DYN4
NUMA4
COMPACTION3
DIFF3
DRAFT3
KEEPMAP3
PR43
RECOMPACT3
SNAPKV3
VLLM3
BENCH2
EXP2
FAKE2
GPU2
HEADINFER2
LOG2
MAX2
RELAX2
RESERVE2
TAS2
ACCEPT1
ALLOC1
ARM1
BLOCK1
CACTUS1
CP1
DEBUG1
DELTA1
DENSE1
DUMMY1
EXAMPLE1
FORCE1
FREE1
FSM1
FWD1
HELD1
HI1
INIT1
INPUT1
LAST1
LSE1
MAXPOS1
MINIFORGE1
MODEL1
NAIVE1
NUM1
OFFLOAD1
OUT1
PAYLOAD1
PORT1
PRESERVE1
READY1
REUSE1
SAVE1
SEED1
SHARED1
SKIP1
SMALL1
SPECULATIVE1
STREAMINGLLM1
TASK1
TEMPBUF1
TOKEN1
UPLOAD1