Reference: environment and configuration¶
TandemEngine reads four families of settings. Engine switches are QWEN38_* variables, read through one registry. Server settings are command-line flags (Command-line reference); ops/start.sh builds them from ops/serve.env, and a few QSE_* variables set flags' defaults or serve the scripts. The installer and the service scripts read a few TANDEM_* variables. And the conductor's behaviour lives in its versioned documents, which the dashboard edits. The QWEN38_* knobs the extension reads, and the variables of TandemApp's proxy, collector and plugins, are in TandemApp's docs.
The served profile¶
ops/serve.env is the served configuration. On an installed engine it is ~/TandemLLM/run/ops/serve.env, rewritten by every run of the installer, and local.env beside it is read last for your own values.
The keys ops/serve.env sets, in file order (paths not shown)¶
| Key | Value | Kind |
|---|---|---|
HOME_DIR |
(a path) |
read by ops/start.sh |
REPO |
(a path) |
read by ops/start.sh |
PY |
(a path) |
read by ops/start.sh |
PYTHONPATH |
(a path) |
read by ops/start.sh |
HOST |
0.0.0.0 |
read by ops/start.sh |
PORT |
8000 |
read by ops/start.sh |
SERVED_MODEL |
qwen38-spark-engine |
read by ops/start.sh |
NV |
(a path) |
read by ops/start.sh |
HEAD |
(a path) |
read by ops/start.sh |
CKPT8 |
(a path) |
read by ops/start.sh |
CKPT16 |
(a path) |
read by ops/start.sh |
CORPUS |
(a path) |
read by ops/start.sh |
MAX_LEN |
262144 |
read by ops/start.sh |
DEFAULT_MAX_TOKENS |
32768 |
read by ops/start.sh |
PATTERN_STOP |
(empty) |
read by ops/start.sh |
QSE_MODE |
(empty) |
server setting |
CONDUCTOR |
on |
read by ops/start.sh |
REASONING_FORMAT |
tags |
read by ops/start.sh |
REASONING_EFFORT |
medium |
read by ops/start.sh |
LEN_FIXED |
0 |
read by ops/start.sh |
LEN_LATCH |
1 |
read by ops/start.sh |
BUDGET |
16 |
read by ops/start.sh |
MEM_CEILING_GIB |
70 |
read by ops/start.sh |
MEM_PEAK_GIB |
77 |
read by ops/start.sh |
RESERVE_SLACK_GIB |
2 |
read by ops/start.sh |
RESIDENT_GB |
2.5 |
read by ops/start.sh |
RESIDENT_STASH_GB |
1.0 |
read by ops/start.sh |
RESIDENT_GUEST_MAX |
8192 |
read by ops/start.sh |
RESIDENT_PIN_GB |
4.0 |
read by ops/start.sh |
RESIDENT_PARK_GB |
4 |
read by ops/start.sh |
CACHE_GB |
2 |
read by ops/start.sh |
STORE_ENTRY_GB |
1 |
read by ops/start.sh |
NARROW_ARM |
1 |
read by ops/start.sh |
END_ANCHOR |
on |
read by ops/start.sh |
END_ALIGN |
on |
read by ops/start.sh |
DROP_PAGE_CACHE |
1 |
read by ops/start.sh |
REQUEST_TIMEOUT |
3600 |
read by ops/start.sh |
MAX_QUEUE |
8 |
read by ops/start.sh |
QUEUE_TIMEOUT |
240 |
read by ops/start.sh |
FUSE_PROJ |
0 |
read by ops/start.sh |
QWEN38_DF2_TREE_MODE |
nodes |
engine switch |
QWEN38_TREE_ALIAS_STATE |
0 |
engine switch |
DROP_IDLE |
1 |
read by ops/start.sh |
QWEN38_DRAFT_TEMP |
0.3 |
engine switch |
SAMPLED_TREE |
det |
read by ops/start.sh |
QWEN38_FUSED_GDNPREFILL |
1 |
engine switch |
QWEN38_DECODE_ATTN |
1 |
engine switch |
QWEN38_DRAFT_NVFP4 |
1 |
engine switch |
QWEN38_NVFP4_PREFILL_V2 |
1 |
engine switch |
QWEN38_NVFP4_PREFILL_V2_UNTIL |
1025 |
engine switch |
QWEN38_NVFP4_SKINNY |
1 |
engine switch |
QWEN38_SKINNY_TILES |
(a path) |
engine switch |
QWEN38_FUSED_COMMIT |
1 |
engine switch |
QWEN38_FUSED_GDNVERIFY |
1 |
engine switch |
QWEN38_DRAFT_FC_NVFP4 |
1 |
engine switch |
QWEN38_FUSED_ADDNORM |
1 |
engine switch |
QWEN38_TREE_HOST_DEPTH |
1 |
engine switch |
QWEN38_HOST_WALK |
1 |
engine switch |
QWEN38_VERIFY_GRAPH |
1 |
engine switch |
QWEN38_COMMIT_IN_VERIFY |
1 |
engine switch |
QWEN38_DRAFT_GRAPH |
1 |
engine switch |
QWEN38_GDNV_ONE_WARP |
1 |
engine switch |
QWEN38_TREE_NODES_NARROW |
16 |
engine switch |
QWEN38_TREE_MS |
8:77.9,16:78.9,24:87.0,32:97.0 |
engine switch |
QWEN38_VERIFY_ROWS |
32 |
engine switch |
QWEN38_SKINNY_TILES_WIDE |
(a path) |
engine switch |
QWEN38_TREE_NODES |
24 |
engine switch |
QWEN38_TREE_WIDE_AFTER |
32 |
engine switch |
QWEN38_DEEP |
32 |
engine switch |
QWEN38_DEEP_AFTER |
2 |
engine switch |
QWEN38_LEN_SWITCH |
1 |
engine switch |
QWEN38_LEN_MODE |
wide |
engine switch |
QWEN38_STAIR_TABLES |
(a path) |
engine switch |
QWEN38_STAIR_RHO |
(a path) |
engine switch |
QWEN38_STAIR_SKIP |
1 |
engine switch |
QWEN38_SKINNY_SRUN |
1 |
engine switch |
QSE_USAGE_LEDGER |
(a path) |
server setting |
QSE_CONDUCTOR_CONFIG |
(a path) |
server setting |
QSE_BACKENDS |
(empty) |
server setting |
QSE_PARALLEL |
8 |
server setting |
QSE_SIDE_MAX_LEN |
0 |
server setting |
QSE_PARALLEL_FLOOR_GB |
15 |
server setting |
QSE_SIDE_GRAPHS |
off |
server setting |
QSE_SIDE_SLOT_CONST_GB |
1.5 |
server setting |
QSE_KV_POOL |
on |
server setting |
QSE_KV_HARD_FLOOR_GB |
10 |
server setting |
QSE_KV_RECOMPUTE |
on |
server setting |
QSE_KV_ADMIT_ROWS |
8192 |
server setting |
QSE_BARRIER_IDLE_OWNER |
on |
server setting |
QWEN38_GDNV_CONV_BLOCK |
32 |
engine switch |
QWEN38_GDNV_CONV_WARPS |
1 |
engine switch |
QWEN38_GDNV_WY |
1 |
engine switch |
QWEN38_GDNV_WY_FUSED |
1 |
engine switch |
QWEN38_GDNV_WY_MAXT |
16 |
engine switch |
QWEN38_GDNV_WY_CHAIN_MAXT |
16 |
engine switch |
QWEN38_GDNV_WY_FUSED_MAXT |
16 |
engine switch |
LIVE_ACTIVITY |
on |
read by ops/start.sh |
QSE_DASHBOARD_LOGIN |
off |
server setting |
ROLES_AUTOLOAD |
(empty) |
read by ops/start.sh |
QWEN38_DECIDER_MODELS |
(empty) |
engine switch |
QWEN38_DECIDER_DIR |
(a path) |
engine switch |
QWEN38_NARRATOR_MODEL |
(empty) |
engine switch |
QWEN38_NARRATOR_DIR |
(empty) |
engine switch |
QWEN38_SHADOW_MODEL |
(empty) |
engine switch |
QWEN38_SHADOW_DIR |
(empty) |
engine switch |
QWEN38_SHADOW_GB |
4 |
engine switch |
QWEN38_SHADOW_GRID |
512 |
engine switch |
QWEN38_SHADOW_DISK |
auto |
engine switch |
QWEN38_SHADOW_DISK_GB |
20 |
engine switch |
QWEN38_PROBE_SNAPSHOTS |
off |
engine switch |
QWEN38_PROBE_DIR |
(a path) |
engine switch |
QSE_DISK_CACHE |
(a path) |
server setting |
SESSION_SPILL |
on |
read by ops/start.sh |
SESSION_SPILL_GB |
64 |
read by ops/start.sh |
SESSION_SPILL_CODEC |
zstd |
read by ops/start.sh |
DISK_WRITE_BUDGET_GB |
200 |
read by ops/start.sh |
SESSION_SPILL_IDLE |
on |
read by ops/start.sh |
VISION |
on |
read by ops/start.sh |
RESPONSE_CACHE |
off |
read by ops/start.sh |
QSE_STATUS_PORT |
(empty) |
server setting |
QSE_PROFILE_NAME |
(a path) |
server setting |
QSE_PACK_VISION |
(empty) |
server setting |
QSE_SETUP_ENV_FILE |
(a path) |
server setting |
QSE_VISION_ASKED |
(a path) |
server setting |
VISION |
off |
read by ops/start.sh |
QSE_VISION_ASKED |
(a path) |
server setting |
VISION |
off |
read by ops/start.sh |
QSE_MODE |
engine |
server setting |
QSE_MODE |
chat |
server setting |
QSE_CONDUCTOR_CONFIG |
(a path) |
server setting |
QSE_EXT |
(a path) |
server setting |
QSE_MODE |
engine |
server setting |
CONDUCTOR |
raw |
read by ops/start.sh |
QSE_CONDUCTOR_CONFIG |
off |
server setting |
QWEN38_DECIDER_MODELS |
(empty) |
engine switch |
QWEN38_NARRATOR_MODEL |
(empty) |
engine switch |
QWEN38_SHADOW_MODEL |
(empty) |
engine switch |
ROLES_AUTOLOAD |
(empty) |
read by ops/start.sh |
QSE_EXT |
(empty) |
server setting |
QSE_CHATD_ASKED |
(a path) |
server setting |
QSE_CHATD |
off ;; |
server setting |
Engine switches¶
engine/settings.py lists every engine switch with its default and the module that reads it; tests/test_settings.py fails when a module reads one another way. The table shows each switch, its code default, and the value the served profile sets when it sets one.
Every QWEN38_* switch the engine reads, with the value ops/serve.env sets (the extension's, in contracts/engine-ext-v1/knobs.json, are in TandemApp's docs)¶
| Variable | Code default | Served value · read in |
|---|---|---|
QWEN38_ATTN_INV |
0 |
tools/attn_kernels.py |
QWEN38_ATTN_INV_NT |
4 |
tools/attn_kernels.py |
QWEN38_ATTN_INV_SPLIT |
1024 |
tools/attn_kernels.py |
QWEN38_ATTN_STAGES3 |
1 |
tools/attn_kernels.py |
QWEN38_BLOCKING_SYNC |
0 |
server/app.py |
QWEN38_CLIENT_REASONING |
*=reasoning_content,hermes=tags |
server/compat.py |
QWEN38_COMMIT_IN_VERIFY |
0 |
1 · engine/model.py |
QWEN38_CORPUS |
none |
engine/drafters/ngram.py |
QWEN38_DECIDER_COLLATE |
call |
server/decider_api.py |
QWEN38_DECIDER_DEVICE |
auto |
server/decider_api.py |
QWEN38_DECIDER_DIR |
~/.qwen38-spark-engine/deciders/laya |
(a path) · server/decider_api.py |
QWEN38_DECIDER_MAX_BATCH |
64 |
server/decider_api.py |
QWEN38_DECIDER_MAX_QUEUE |
256 |
server/decider_api.py |
QWEN38_DECIDER_MODELS |
none |
server/decider_api.py |
QWEN38_DECIDER_WAIT_MS |
2 |
server/decider_api.py |
QWEN38_DECODE_ATTN |
0 |
1 · engine/model.py |
QWEN38_DEEP |
0 |
32 · server/app.py |
QWEN38_DEEP_AFTER |
2 |
2 · server/app.py |
QWEN38_DF2_DEFER_SYNC |
1 |
engine/drafters/dflash2.py |
QWEN38_DF2_KV_WINDOW |
0 |
engine/drafters/dflash2.py |
QWEN38_DF2_PACK_BATCH |
1 |
engine/drafters/dflash2.py |
QWEN38_DF2_PACK_FOLD |
0 |
engine/drafters/dflash2.py |
QWEN38_DF2_PACK_GQA |
0 |
engine/drafters/dflash2.py |
QWEN38_DF2_PACK_HOST1 |
1 |
engine/drafters/draft_graph.py |
QWEN38_DF2_PACK_KP |
1 |
engine/drafters/dflash2.py |
QWEN38_DF2_PACK_TAIL |
0 |
engine/drafters/dflash2.py |
QWEN38_DF2_PACK_TOPK |
0 |
engine/drafters/dflash2.py |
QWEN38_DF2_SPAN_STEP |
256 |
engine/drafters/draft_graph.py |
QWEN38_DF2_TEMP |
1.0 |
engine/drafters/dflash2.py |
QWEN38_DF2_TOPK_FAST |
1 |
engine/drafters/topk_small.py |
QWEN38_DF2_TREE_MODE |
paths |
nodes · engine/drafters/dflash2.py |
QWEN38_DFLASH2 |
none |
engine/drafters/dflash2.py |
QWEN38_DRAFT_EXL3 |
none |
engine/drafters/dflash2.py |
QWEN38_DRAFT_FC_NVFP4 |
0 |
1 · engine/drafters/dflash2.py |
QWEN38_DRAFT_GRAPH |
0 |
1 · engine/drafters/dflash2.py |
QWEN38_DRAFT_HEAD |
none |
engine/drafters/dflash2.py |
QWEN38_DRAFT_HEAD_EXL3 |
none |
engine/drafters/dflash2.py |
QWEN38_DRAFT_HEAD_NVFP4 |
0 |
engine/drafters/dflash2.py |
QWEN38_DRAFT_NVFP4 |
0 |
1 · engine/drafters/dflash2.py |
QWEN38_DRAFT_TEMP |
1.0 |
0.3 · engine/drafters/dflash2.py |
QWEN38_DSPARK |
none |
engine/drafters/dspark.py |
QWEN38_DSPARK_NO_YARN |
none |
engine/drafters/dspark.py |
QWEN38_END_ALIGN_CALLS |
on |
server/endalign.py |
QWEN38_END_ALIGN_MOVED |
on |
server/endalign.py |
QWEN38_END_ALIGN_REASONING |
on |
server/endalign.py |
QWEN38_END_RESERVE |
on |
server/app.py |
QWEN38_EXL3 |
none |
engine/loader.py |
QWEN38_EXL3_CHUNK_UNTIL |
128 |
tools/exl3_linear.py |
QWEN38_EXL3_FRAG |
1 |
engine/models/qwen3_5/frag.py |
QWEN38_EXL3_FUSE |
1 |
engine/loader.py |
QWEN38_EXL3_MLP_BRIDGE |
1 |
tools/exl3_linear.py |
QWEN38_EXL3_PDL |
0 |
tools/exl3_linear.py |
QWEN38_EXL3_ROWGRID |
1 |
tools/exl3_linear.py |
QWEN38_EXL3_SILU_FUSED |
1 |
tools/exl3_linear.py |
QWEN38_EXL3_STAGES |
3 |
tools/exl3_linear.py |
QWEN38_EXL3_TALL |
none |
engine/models/qwen3_5/tall.py |
QWEN38_EXL3_TILES |
none |
tools/exl3_linear.py |
QWEN38_FLINT_ANCHOR_BACKOFF |
0 |
engine/models/flint/serving.py |
QWEN38_FLINT_KV_FP8 |
0 |
engine/models/flint/engine.py |
QWEN38_FLINT_LOOKUP |
1 |
engine/models/flint/serving.py |
QWEN38_FLINT_LOOK_BACKOFF |
0 |
engine/models/flint/serving.py |
QWEN38_FLINT_PDL |
1 |
engine/models/flint/engine.py |
QWEN38_FLINT_SIDE_STREAM |
1 |
engine/models/flint/engine.py |
QWEN38_FN_ATTN_MULTI |
0 |
engine/models/flashnext/engine.py |
QWEN38_FN_DENSE |
0 |
engine/models/flashnext/engine.py |
QWEN38_FN_DRAFT_FREEZE |
0 |
engine/models/flashnext/serving.py |
QWEN38_FN_DRAFT_HOT |
32768 |
engine/models/flashnext/serving.py |
QWEN38_FN_DROP_SHARD_CACHE |
1 |
engine/models/flashnext/load.py |
QWEN38_FN_FOLD_COMMIT |
1 |
engine/models/flashnext/engine.py |
QWEN38_FN_GDN_DEFER |
0 |
engine/models/flashnext/engine.py |
QWEN38_FN_GDN_MULTI |
1 |
engine/models/flashnext/engine.py |
QWEN38_FN_GDN_PRE_MULTI |
0 |
engine/models/flashnext/engine.py |
QWEN38_FN_GEMV_MC |
0 |
tools/fn_kernels.py |
QWEN38_FN_GEMV_NB |
0 |
tools/fn_kernels.py |
QWEN38_FN_GEMV_SPLIT |
1 |
tools/fn_kernels.py |
QWEN38_FN_GROUP_ROWS |
64 |
tools/fn_prefill_kernels.py |
QWEN38_FN_JG_PLE_LATE |
0 |
engine/models/flashnext/engine.py |
QWEN38_FN_JOINT_GRAPH |
0 |
engine/models/flashnext/engine.py |
QWEN38_FN_JOINT_GRAPH_AFTER |
2 |
engine/models/flashnext/engine.py |
QWEN38_FN_JOINT_GRAPH_MAX |
12 |
engine/models/flashnext/engine.py |
QWEN38_FN_JOINT_MOE_GROUP |
1 |
engine/models/flashnext/engine.py |
QWEN38_FN_JOINT_PREFILL |
0 |
engine/models/flashnext/joint_prefill.py |
QWEN38_FN_JOINT_ROWS8 |
24 |
engine/models/flashnext/serving.py |
QWEN38_FN_JOINT_ROWWISE |
1 |
engine/models/flashnext/engine.py |
QWEN38_FN_JOINT_TIMING |
0 |
engine/models/flashnext/engine.py |
QWEN38_FN_KV_FP8 |
0 |
engine/models/flashnext/engine.py |
QWEN38_FN_LOOKUP |
0 |
engine/models/flashnext/serving.py |
QWEN38_FN_LOOK_BACKOFF |
0 |
engine/models/flashnext/serving.py |
QWEN38_FN_MIX |
int8 |
engine/models/flashnext/pack.py |
QWEN38_FN_MIXER_MMA |
1 |
tools/fn_kernels.py |
QWEN38_FN_MIX_FUSED |
0 |
engine/models/flashnext/engine.py |
QWEN38_FN_MOE_AS |
0 |
tools/fn_kernels.py |
QWEN38_FN_MOE_GPAR |
0 |
tools/fn_kernels.py |
QWEN38_FN_MOE_GR |
8 |
tools/fn_kernels.py |
QWEN38_FN_MOE_GY |
0 |
tools/fn_kernels.py |
QWEN38_FN_MOE_ST |
3 |
tools/gpu/gb10.py |
QWEN38_FN_MOE_STAGE |
0 |
tools/fn_kernels.py |
QWEN38_FN_MOE_WARPS |
4 |
tools/gpu/gb10.py |
QWEN38_FN_MTP_CORRECTOR |
0 |
engine/models/flashnext/corrector.py |
QWEN38_FN_MTP_CORRECTOR_FILE |
auto |
engine/models/flashnext/corrector.py |
QWEN38_FN_MTP_COUPLED |
0 |
engine/models/flashnext/coupled.py |
QWEN38_FN_MTP_COUPLED_CAL |
0 |
engine/models/flashnext/coupled.py |
QWEN38_FN_MTP_COUPLED_FILTER |
1 |
engine/models/flashnext/coupled.py |
QWEN38_FN_MTP_COUPLED_HOT |
0 |
engine/models/flashnext/coupled.py |
QWEN38_FN_MTP_COUPLED_HOT_LONE |
0 |
engine/models/flashnext/coupled.py |
QWEN38_FN_MTP_COUPLED_SIB |
0 |
engine/models/flashnext/coupled.py |
QWEN38_FN_MTP_COUPLED_SIB_MARGIN |
1.3 |
engine/models/flashnext/coupled.py |
QWEN38_FN_MTP_COUPLED_TAU |
0.85 |
engine/models/flashnext/coupled.py |
QWEN38_FN_PACK_GRAPH |
1 |
engine/models/flashnext/engine.py |
QWEN38_FN_PACK_GRAPH_AFTER |
2 |
engine/models/flashnext/engine.py |
QWEN38_FN_PDL |
1 |
tools/fn_kernels.py |
QWEN38_FN_PF_ATTN_STAGES |
1 |
engine/models/flashnext/engine.py |
QWEN38_FN_PF_HC_FUSED |
1 |
engine/models/flashnext/engine.py |
QWEN38_FN_PF_INJ_FUSED |
1 |
engine/models/flashnext/engine.py |
QWEN38_FN_PF_NOCOPY |
1 |
engine/models/flashnext/engine.py |
QWEN38_FN_PF_ROT_ROWS |
0 |
tools/fn_prefill_kernels.py |
QWEN38_FN_PF_WINDOW |
8192 |
engine/models/flashnext/layer_major.py |
QWEN38_FN_PLE_BATCH |
1 |
engine/models/flashnext/engine.py |
QWEN38_FN_PLE_LATE |
1 |
engine/models/flashnext/engine.py |
QWEN38_FN_PLE_MULTI |
0 |
engine/models/flashnext/engine.py |
QWEN38_FN_PLE_PREFETCH |
0 |
engine/models/flashnext/engine.py |
QWEN38_FN_PLE_THREADS |
16 |
engine/models/flashnext/ple.py |
QWEN38_FN_POOL_MULTI |
0 |
engine/models/flashnext/engine.py |
QWEN38_FN_PREFILL |
kernels |
engine/models/flashnext/engine.py |
QWEN38_FN_PREFILL_CHUNK |
4096 |
engine/models/flashnext/engine.py |
QWEN38_FN_PREFILL_MM |
fp32 |
tools/fn_prefill_kernels.py |
QWEN38_FN_QSA_EPI_TORCH |
0 |
tools/fn_prefill_kernels.py |
QWEN38_FN_QSA_LIVE |
1 |
engine/models/flashnext/engine.py |
QWEN38_FN_QSA_TORCH_BIG |
0 |
tools/fn_seq_kernels.py |
QWEN38_FN_ROUTE_FUSED |
1 |
engine/models/flashnext/engine.py |
QWEN38_FN_TABLE |
warm |
engine/models/flashnext/load.py |
QWEN38_FN_TABLE_CACHE_GB |
2 |
engine/models/flashnext/load.py |
QWEN38_FN_TREE_PLAN |
0 |
engine/models/flashnext/engine.py |
QWEN38_FN_UNION_EVERY |
0 |
engine/models/flashnext/engine.py |
QWEN38_FN_WARM_TABLE |
1 |
engine/models/flashnext/load.py |
QWEN38_FP8_HEAD |
none |
engine/loader.py |
QWEN38_FP8_TILES |
none |
tools/fp8_linear.py |
QWEN38_FRESH_SEED |
auto |
server/app.py |
QWEN38_FUSED_ADDNORM |
0 |
1 · engine/model.py |
QWEN38_FUSED_ATTN |
0 |
engine/model.py |
QWEN38_FUSED_ATTN_PREP |
1 |
engine/model.py |
QWEN38_FUSED_COMMIT |
0 |
1 · engine/model.py |
QWEN38_FUSED_GDN |
1 |
engine/model.py |
QWEN38_FUSED_GDNBLOCK |
1 |
engine/model.py |
QWEN38_FUSED_GDNPRE |
1 |
engine/model.py |
QWEN38_FUSED_GDNPREFILL |
0 |
1 · engine/model.py |
QWEN38_FUSED_GDNTREE |
1 |
engine/model.py |
QWEN38_FUSED_GDNVERIFY |
0 |
1 · engine/model.py |
QWEN38_FUSED_HEAD |
1 |
engine/model.py |
QWEN38_FUSED_NORM |
1 |
engine/model.py |
QWEN38_FUSE_PROJ |
0 |
tools/nvfp4_linear_v2.py |
QWEN38_GDNV_BV |
16 |
tools/gdn_verify_kernels.py |
QWEN38_GDNV_CONV_BLOCK |
256 |
32 · tools/gdn_verify_kernels.py |
QWEN38_GDNV_CONV_WARPS |
4 |
1 · tools/gdn_verify_kernels.py |
QWEN38_GDNV_ONE_WARP |
0 |
1 · tools/gdn_verify_kernels.py |
QWEN38_GDNV_TREE_PF |
0 |
tools/gdn_verify_kernels.py |
QWEN38_GDNV_WARPS |
4 |
tools/gdn_verify_kernels.py |
QWEN38_GDNV_WY |
0 |
1 · tools/gdn_verify_kernels.py |
QWEN38_GDNV_WY_CHAIN_MAXT |
32 |
16 · tools/gdn_verify_kernels.py |
QWEN38_GDNV_WY_FUSED |
0 |
1 · tools/gdn_verify_kernels.py |
QWEN38_GDNV_WY_FUSED_MAXT |
32 |
16 · tools/gdn_verify_kernels.py |
QWEN38_GDNV_WY_KC |
32 |
tools/gdn_verify_kernels.py |
QWEN38_GDNV_WY_MAXT |
32 |
16 · tools/gdn_verify_kernels.py |
QWEN38_GDN_AB |
1 |
engine/model.py |
QWEN38_GDN_CHUNK |
64 |
engine/model.py |
QWEN38_GDN_CHUNKED |
1 |
engine/model.py |
QWEN38_GDN_CHUNKED_BV |
16 |
tools/gdn_chunked_kernels.py |
QWEN38_GDN_CHUNKED_C |
64 |
tools/gdn_chunked_kernels.py |
QWEN38_GDN_CHUNKED_WARPS |
4 |
tools/gdn_chunked_kernels.py |
QWEN38_GDN_CHUNKED_WARPS_INTRA |
8 |
tools/gdn_chunked_kernels.py |
QWEN38_GDN_MM |
fp32 |
engine/gdn.py |
QWEN38_GDN_MM_FROM |
64 |
engine/gdn.py |
QWEN38_GDN_PREFILL_BV |
32 |
tools/gdn_prefill_kernels.py |
QWEN38_GDN_PREFILL_NOREP |
0 |
engine/model.py |
QWEN38_GDN_PREFILL_PREC |
bf16x3 |
tools/gdn_prefill_kernels.py |
QWEN38_GDN_PREFILL_PREC_A |
tf32 |
tools/gdn_prefill_kernels.py |
QWEN38_GDN_PREFILL_PREC_SCAN |
none |
tools/gdn_prefill_kernels.py |
QWEN38_GDN_PREFILL_SERIES |
0 |
tools/gdn_prefill_kernels.py |
QWEN38_GDN_PREFILL_STAGES |
1 |
tools/gdn_prefill_kernels.py |
QWEN38_GDN_PREFILL_WARPS_INTRA |
8 |
tools/gdn_prefill_kernels.py |
QWEN38_GDN_PREFILL_WARPS_SCAN |
8 |
tools/gdn_prefill_kernels.py |
QWEN38_GDN_PREP |
1 |
engine/model.py |
QWEN38_GDN_PREP_BV |
16 |
engine/model.py |
QWEN38_GDN_PREP_WARPS |
4 |
engine/model.py |
QWEN38_GDN_SEQ |
0 |
tools/gdn_seq_kernels.py |
QWEN38_GDN_SEQ_BV |
16 |
tools/gdn_seq_kernels.py |
QWEN38_GDN_SEQ_BYNODE |
0 |
tools/gdn_seq_kernels.py |
QWEN38_GDN_SEQ_WARPS |
1 |
tools/gdn_seq_kernels.py |
QWEN38_GQA_FROM |
64 |
engine/model.py |
QWEN38_GRAPH_MAX_CTX |
32768 |
engine/verify_graph.py |
QWEN38_GRAPH_SHARE_FROM |
2 |
engine/verify_graph.py |
QWEN38_GRAPH_SHARE_OUT |
0 |
engine/verify_graph.py |
QWEN38_HEAD_GEMM |
none |
tools/head_gemv.py |
QWEN38_HOST_ASYNC |
0 |
engine/model.py |
QWEN38_HOST_WALK |
0 |
1 · engine/drafters/dflash2.py |
QWEN38_JOINT_ATTN_GROUP |
1 |
engine/models/qwen3_5/attn_group.py |
QWEN38_JOINT_CHAINS |
1 |
engine/models/qwen3_5/adapter.py |
QWEN38_JOINT_GDN |
1 |
engine/model.py |
QWEN38_JOINT_GRAPHS |
1 |
engine/models/qwen3_5/joint_graphs.py |
QWEN38_JOINT_GRAPHS_AFTER |
4 |
engine/models/qwen3_5/joint_graphs.py |
QWEN38_JOINT_GRAPHS_MAX |
24 |
engine/models/qwen3_5/joint_graphs.py |
QWEN38_JOINT_MAX_ROWS |
none |
engine/models/qwen3_5/adapter.py |
QWEN38_JOINT_ROWS_AT_LANES |
4:8 |
engine/models/qwen3_5/joint_graphs.py |
QWEN38_JOINT_ROWS_A_SLOT |
0 |
engine/models/qwen3_5/joint_graphs.py |
QWEN38_JOINT_ROWWISE |
1 |
engine/model.py |
QWEN38_JOIN_VIEWS |
1 |
engine/model.py |
QWEN38_KV_FP8 |
0 |
engine/model.py |
QWEN38_LATCH_PRICE |
0 |
engine/lenrouter.py |
QWEN38_LAUNCH_FIRST |
0 |
server/app.py |
QWEN38_LEN_MODE |
none |
wide · engine/lenrouter.py |
QWEN38_LEN_SWITCH |
0 |
1 · engine/lenrouter.py |
QWEN38_LFM2_KV_FP8 |
0 |
engine/models/lfm2_moe/engine.py |
QWEN38_LOWER_RIGHT_FROM |
64 |
engine/model.py |
QWEN38_MEMORY_PROXIES |
none |
server/app.py |
QWEN38_MEMORY_WRITER_TEXT |
none |
server/app.py |
QWEN38_MM_END |
on |
server/app.py |
QWEN38_MODEL |
none |
engine/config.py |
QWEN38_MTP_BACKLOG_CAP |
2048 |
engine/methods/mtp_drafter.py |
QWEN38_NARRATOR_DEVICE |
auto |
server/conductor/inproc.py |
QWEN38_NARRATOR_DIR |
none |
server/conductor/inproc.py |
QWEN38_NH_KV_FP8 |
0 |
engine/models/nemotron_h/engine.py |
QWEN38_NVFP4 |
none |
engine/loader.py |
QWEN38_NVFP4_BLOCK_TILES |
0 |
tools/nvfp4_linear.py |
QWEN38_NVFP4_DEQUANT_FROM |
512 |
tools/nvfp4_linear.py |
QWEN38_NVFP4_PREFILL_V2 |
0 |
1 · tools/nvfp4_linear.py |
QWEN38_NVFP4_PREFILL_V2_UNTIL |
1024 |
1025 · tools/nvfp4_linear.py |
QWEN38_NVFP4_SKINNY |
0 |
1 · tools/nvfp4_skinny.py |
QWEN38_NVFP4_SPLITK |
0 |
tools/nvfp4_linear.py |
QWEN38_NVFP4_V2 |
1 |
tools/nvfp4_linear_v2.py |
QWEN38_NVFP4_V2_BN |
0 |
tools/nvfp4_linear_v2.py |
QWEN38_NVFP4_V2_DOTS |
1 |
tools/nvfp4_linear_v2.py |
QWEN38_NVFP4_V2_MAX |
32 |
tools/nvfp4_linear_v2.py |
QWEN38_NVFP4_V2_MIN |
1 |
tools/nvfp4_linear_v2.py |
QWEN38_NVFP4_V2_PREFETCH |
0 |
tools/nvfp4_linear_v2.py |
QWEN38_NVFP4_V2_W |
0 |
tools/nvfp4_linear_v2.py |
QWEN38_OWN_BURST_RATE |
peek |
server/app.py |
QWEN38_OWN_ETA |
on |
server/app.py |
QWEN38_OWN_RATE |
solo |
server/app.py |
QWEN38_OWN_SIDE |
on |
server/app.py |
QWEN38_OWN_UNITS |
prefill |
server/app.py |
QWEN38_POOL_SNAPSHOT_TTL_S |
5.0 |
engine/decision/device.py |
QWEN38_PREFILL_CAUSAL |
1 |
engine/model.py |
QWEN38_PREFILL_WINDOW |
8192 |
engine/cache.py |
QWEN38_PREFILL_WINDOW_FREE |
32768 |
engine/cache.py |
QWEN38_PROBE_DAY_MB |
100 |
server/probe.py |
QWEN38_PROBE_DIR |
~/.qwen38-spark-engine/probe |
(a path) · server/probe.py |
QWEN38_PROBE_EVERY |
1024 |
server/probe.py |
QWEN38_PROBE_SNAPSHOTS |
off |
off · server/probe.py |
QWEN38_PROMPT_END |
1 |
engine/cache.py |
QWEN38_Q35M_ANCHOR_BACKOFF |
0 |
engine/models/qwen3_5_moe/serving.py |
QWEN38_Q35M_DFLASH2_HOT |
0 |
engine/models/qwen3_5_moe/dflash2.py |
QWEN38_Q35M_DFLASH2_NVFP4 |
0 |
engine/models/qwen3_5_moe/dflash2.py |
QWEN38_Q35M_KV_FP8 |
0 |
engine/models/qwen3_5_moe/engine.py |
QWEN38_Q35M_LOOKUP |
1 |
engine/models/qwen3_5_moe/serving.py |
QWEN38_Q35M_SIBLINGS |
0 |
engine/models/qwen3_5_moe/serving.py |
QWEN38_RANKK |
1 |
engine/model.py |
QWEN38_REGENERATE |
sample |
server/app.py |
QWEN38_REGEN_TEMPERATURE |
0.7 |
server/app.py |
QWEN38_REGEN_TOP_K |
20 |
server/app.py |
QWEN38_REGEN_TOP_P |
0.95 |
server/app.py |
QWEN38_REREAD_SAY_TOKENS |
8192 |
server/app.py |
QWEN38_ROUND_LOG |
none |
server/roundlog.py |
QWEN38_ROUND_LOG_HIDDEN |
0 |
server/roundlog.py |
QWEN38_ROUND_LOG_LATTICE |
0 |
server/roundlog.py |
QWEN38_ROUND_PLAN |
off |
engine/models/qwen3_5/roundplan.py |
QWEN38_ROUND_PLAN_BUCKETS |
4,8,16 |
engine/models/qwen3_5/roundplan.py |
QWEN38_ROUND_PLAN_PRICES |
none |
engine/models/qwen3_5/roundplan.py |
QWEN38_ROUND_PLAN_SAVE_EVERY |
512 |
engine/models/qwen3_5/roundplan.py |
QWEN38_ROUND_PLAN_STATE |
none |
engine/models/qwen3_5/roundplan.py |
QWEN38_SAMPLED_CUT |
0 |
engine/router.py |
QWEN38_SAMPLE_BATCH |
1 |
engine/sample.py |
QWEN38_SAMPLE_KEYED_KERNEL |
0 |
engine/sample.py |
QWEN38_SAMPLE_NUCLEUS |
1 |
engine/sample.py |
QWEN38_SAMPLE_SPARSE_KEYED |
1 |
engine/sample.py |
QWEN38_SAMPLE_TOPP_K |
1024 |
engine/sample.py |
QWEN38_SCALE_ON |
weight |
tools/fp8_linear.py |
QWEN38_SHADOW_DIR |
none |
server/conductor/shadow.py |
QWEN38_SHADOW_DISK |
none |
auto · server/conductor/shadow.py |
QWEN38_SHADOW_DISK_GB |
20 |
20 · server/conductor/shadow.py |
QWEN38_SHADOW_GB |
4 |
4 · server/conductor/shadow.py |
QWEN38_SHADOW_GRID |
512 |
512 · server/conductor/shadow.py |
QWEN38_SHADOW_MODEL |
none |
server/conductor/shadow.py |
QWEN38_SHADOW_POOL_SLACK_GIB |
2 |
server/conductor/shadow.py |
QWEN38_SIDE_CARRY |
on |
server/app.py |
QWEN38_SIDE_CONTINUE |
view |
server/app.py |
QWEN38_SIDE_CONV |
on |
server/app.py |
QWEN38_SIDE_DISK |
read |
server/app.py |
QWEN38_SIDE_GUESTS |
on |
server/app.py |
QWEN38_SIDE_GUEST_END |
promote |
server/app.py |
QWEN38_SIDE_GUEST_IDLE_S |
300 |
server/app.py |
QWEN38_SIDE_HOME_WAIT_S |
10 |
server/app.py |
QWEN38_SIDE_IMAGES |
on |
server/app.py |
QWEN38_SIDE_LRU |
on |
server/app.py |
QWEN38_SIDE_MIN_ROOM |
16384 |
server/app.py |
QWEN38_SIDE_NEW_FIRST |
on |
server/app.py |
QWEN38_SIDE_NEW_WAIT_S |
20 |
server/app.py |
QWEN38_SIDE_RESIDENT_GB |
0.5 |
server/app.py |
QWEN38_SIDE_SLOT0_WAIT_S |
3 |
server/app.py |
QWEN38_SIDE_STASH_GB |
1.0 |
server/app.py |
QWEN38_SKINNY_LDW |
0 |
tools/nvfp4_skinny.py |
QWEN38_SKINNY_PDL |
0 |
tools/nvfp4_skinny.py |
QWEN38_SKINNY_SRUN |
0 |
1 · tools/nvfp4_skinny.py |
QWEN38_SKINNY_SSTUB |
0 |
tools/nvfp4_skinny.py |
QWEN38_SKINNY_TILES |
none |
(a path) · tools/nvfp4_skinny.py |
QWEN38_SKINNY_TILES_B |
none |
tools/nvfp4_skinny.py |
QWEN38_SKINNY_TILES_C |
none |
tools/nvfp4_skinny.py |
QWEN38_SKINNY_TILES_WIDE |
none |
(a path) · tools/nvfp4_skinny.py |
QWEN38_SKINNY_TILES_WIDE_B |
none |
tools/nvfp4_skinny.py |
QWEN38_SKINNY_XSTUB |
0 |
tools/nvfp4_skinny.py |
QWEN38_SLAB_TRITON |
1 |
engine/model.py |
QWEN38_SLOT0_KEEP_S |
1800 |
server/app.py |
QWEN38_SPILL_PARK |
on |
engine/cache.py |
QWEN38_STAIR_MAX_NODES |
none |
engine/lenrouter.py |
QWEN38_STAIR_RHO |
none |
(a path) · engine/lenrouter.py |
QWEN38_STAIR_SKIP |
0 |
1 · engine/lenrouter.py |
QWEN38_STAIR_TABLES |
none |
(a path) · engine/lenrouter.py |
QWEN38_STAIR_TEMP_SAMPLED |
none |
engine/lenrouter.py |
QWEN38_SUFFIX_STORE |
~/.qwen38-spark-engine/suffix |
server/app.py |
QWEN38_TREE_ALIAS_STATE |
0 |
0 · engine/model.py |
QWEN38_TREE_CHAIN_DELEGATE |
1 |
engine/model.py |
QWEN38_TREE_HOST_DEPTH |
0 |
1 · engine/model.py |
QWEN38_TREE_MS |
none |
8:77.9,16:78.9,24:87.0,32:97.0 · engine/router.py |
QWEN38_TREE_NATIVE |
1 |
engine/native_tree.py |
QWEN38_TREE_NODES |
0 |
24 · engine/router.py |
QWEN38_TREE_NODES_NARROW |
0 |
16 · engine/router.py |
QWEN38_TREE_WIDE_AFTER |
0 |
32 · engine/lenrouter.py |
QWEN38_TWO_STREAM |
0 |
engine/model.py |
QWEN38_UT_INVERSE |
1 |
engine/gdn.py |
QWEN38_VERIFY_GRAPH |
0 |
1 · engine/model.py |
QWEN38_VERIFY_ROWS |
16 |
32 · engine/model.py |
QWEN38_VERIFY_TILES |
1 |
tools/nvfp4_verify_tiles.py |
QWEN38_WARM_PREFILL_TOKENS |
2600 |
server/app.py |
QWEN38_WINDOW_MLP |
0 |
engine/model.py |
QWEN38_WINDOW_NORM_ROT |
1 |
engine/model.py |
QWEN38_WINDOW_ROWWISE |
1 |
engine/model.py |
QWEN38_WY_BLOCK |
8 |
tools/gdn_wy_kernels.py |
QWEN38_WY_BV |
16 |
tools/gdn_wy_kernels.py |
QWEN38_WY_DBG |
0 |
tools/gdn_wy_kernels.py |
QWEN38_WY_PREC |
ieee |
tools/gdn_wy_kernels.py |
QWEN38_WY_WARPS |
4 |
tools/gdn_wy_kernels.py |
QWEN38_WY_WARPS_PREP |
4 |
tools/gdn_wy_kernels.py |
Server variables¶
QSE_* variables: the server's settings from the environment or serve.env¶
| Variable | Read in |
|---|---|
QSE_ADMIN_TOKEN |
ops/make-secrets.sh, ops/serve.env, ops/start.sh and 5 more |
QSE_ADVERTISE |
ops/serve.env, server/pairing/mdns.py, server/remote/overlay.py |
QSE_ALLOW_FROM |
ops/serve.env, server/api/access.py, server/app.py and 4 more |
QSE_ANSWER_RESERVE |
server/app.py |
QSE_ASSETS |
engine/models/qwen3_5/community.py, ops/profiles/_flashnext.env, ops/profiles/_flint.env and 16 more |
QSE_BACKENDS |
install.sh, ops/engines.sh, ops/serve.env and 5 more |
QSE_BARRIER_IDLE_OWNER |
ops/serve.env, server/slots.py |
QSE_CAPTURE_GUARD_DEBUG |
engine/capture_lock.py |
QSE_CHATD |
install.sh, ops/engines.sh, ops/serve.env and 5 more |
QSE_CHATD_ASKED |
install.sh, ops/serve.env, ops/start.sh and 1 more |
QSE_CHATD_ENV |
install.sh |
QSE_CHAT_DATA |
install.sh, server/setup/doctor.py |
QSE_CHAT_FORWARD_KEY |
ops/make-secrets.sh |
QSE_CHAT_URL |
server/api/__init__.py, server/api/chat_url.py, server/mode.py |
QSE_COMMUNITY_DIR |
server/setup/community.py |
QSE_COMMUNITY_ENV_DIR |
ops/serve.env, ops/use-profile.sh |
QSE_COMMUNITY_URL |
ops/serve.env, server/setup/community.py |
QSE_COMPACTION |
server/app.py |
QSE_COMPACTION_IDLE_S |
server/app.py |
QSE_COMPACTION_MIN_TOKENS |
server/app.py |
QSE_COMPACTION_PREPARE |
server/app.py |
QSE_CONDUCTOR_CONFIG |
ops/engines.sh, ops/serve.env, ops/start.sh and 4 more |
QSE_CONDUCTOR_ROOT |
ops/roles.sh, server/conductor/residency.py |
QSE_CORPUS_BUILD |
ops/use-profile.sh |
QSE_CORRECTOR_HF |
engine/models/flashnext/corrector.py, ops/fetch_weights.py, ops/profiles/_flashnext.env and 2 more |
QSE_CORRECTOR_PATH |
engine/models/flashnext/corrector.py, ops/fetch_weights.py, ops/profiles/flashnext-3.05.env and 1 more |
QSE_CORRECTOR_REVISION |
engine/models/flashnext/corrector.py, ops/fetch_weights.py, ops/profiles/flashnext-3.05.env and 1 more |
QSE_CORRECTOR_SHA256 |
engine/models/flashnext/corrector.py, ops/fetch_weights.py, ops/profiles/flashnext-3.05.env and 1 more |
QSE_DASHBOARD_LOGIN |
ops/serve.env, ops/start.sh, server/api/auth.py and 1 more |
QSE_DISK_ADOPT_FROM |
ops/serve.env, server/diskcache.py |
QSE_DISK_CACHE |
engine/models/nemotron_h/community.py, engine/models/qwen3_5/community.py, ops/engines.sh and 11 more |
QSE_ENGINE_ALLOC_CONF |
engine/cache.py, ops/profiles/_flashnext.env, ops/start.sh |
QSE_ENGINE_FLAGS |
engine/models/nemotron_h/community.py, engine/models/qwen3_5/community.py, ops/engines.sh and 10 more |
QSE_ENGINE_PORT |
ops/engine_status.py |
QSE_EXL3_BPW |
ops/profiles/_exl3.env, ops/profiles/exl3-2.5.env, ops/profiles/exl3-3.0.env and 4 more |
QSE_EXT |
ops/serve.env, server/__init__.py, server/app.py and 6 more |
QSE_EXT_ARGS |
server/app.py, server/ext.py |
QSE_FAKE_CAPTURE_HOLD_S |
server/fake_engine.py |
QSE_FAKE_CAPTURE_MARK |
server/fake_engine.py |
QSE_FAKE_CARD_DIR |
server/fake_engine.py |
QSE_FAKE_PROMPTS |
server/fake_engine.py |
QSE_FETCH_ASSETS |
ops/use-profile.sh, server/setup/apply.py |
QSE_FLINT_GROUPED |
engine/models/flint/engine.py |
QSE_FLINT_VARIANT |
ops/profiles/_flint.env, ops/profiles/flint-bf16.env, ops/profiles/flint-exl3-4.0.env and 1 more |
QSE_FN_BPW |
ops/profiles/_flashnext.env, ops/profiles/flashnext-2.05.env, ops/profiles/flashnext-3.05.env and 3 more |
QSE_FN_TABLE |
ops/profiles/_flashnext.env, ops/profiles/flashnext-2.05.env, ops/profiles/flashnext-3.05.env and 3 more |
QSE_FN_VISION_FILE |
ops/profiles/_flashnext.env, ops/profiles/flashnext-4.05.env |
QSE_GRAPH_POOL_GIB |
server/app.py |
QSE_GRAPH_SET_MIN_IDLE_S |
server/app.py |
QSE_GUARD_MARGIN_GB |
server/app.py |
QSE_HANDLER_TIMEOUT_S |
server/app.py |
QSE_HEALTH_TTL_MS |
ops/profiles/_flashnext.env, server/app.py, server/health_cache.py |
QSE_IMAGE_BACKEND |
server/app.py |
QSE_KV_ADMIT_ROWS |
ops/serve.env, server/app.py |
QSE_KV_HARD_FLOOR_GB |
ops/profiles/_flashnext.env, ops/serve.env, server/app.py |
QSE_KV_MARGIN_ROWS |
server/app.py |
QSE_KV_POOL |
engine/kvpool.py, ops/engines.sh, ops/profiles/_flashnext.env and 7 more |
QSE_KV_POOL_MIN_ROWS |
engine/kvpool.py |
QSE_KV_RECOMPUTE |
ops/serve.env, server/app.py |
QSE_KV_TEST_REFUSE_AFTER |
server/round_runner.py |
QSE_KV_TEST_REFUSE_SLOTS |
server/round_runner.py |
QSE_LFM_VARIANT |
ops/profiles/_lfm25.env, ops/profiles/lfm25-bf16.env, ops/profiles/lfm25-exl3-2.5.env and 5 more |
QSE_LIVE_DECODE_WARMUP_TOKENS |
ops/serve.env, server/api/live.py |
QSE_LIVE_HZ |
ops/serve.env, ops/start.sh, server/api/live.py and 1 more |
QSE_LIVE_MOSTLY_CACHED_PCT |
ops/serve.env, server/api/live.py |
QSE_LIVE_MOSTLY_CACHED_TOKENS |
ops/serve.env, server/api/live.py |
QSE_LIVE_NOW_WINDOW_S |
ops/serve.env, server/api/live.py |
QSE_LIVE_PREFILL_WARMUP_CHUNKS |
ops/serve.env, server/api/live.py |
QSE_MANAGED_ENV |
ops/serve.env, server/remote/overlay.py |
QSE_MAX_BODY_MB |
server/api/http_body.py, server/app.py |
QSE_MEMINFO |
ops/engines.sh |
QSE_MEM_NEED_GIB |
ops/profiles/flashnext-2.05.env, ops/profiles/flashnext-3.05.env, ops/profiles/flashnext-4.05.env and 27 more |
QSE_MEM_WAIT_S |
server/app.py |
QSE_METRICS_TOKEN |
ops/make-secrets.sh, ops/start.sh, server/api/auth.py |
QSE_MODE |
install.sh, ops/engines.sh, ops/serve.env and 5 more |
QSE_NH_DENSE_ROWS |
engine/models/nemotron_h/engine.py |
QSE_NH_DIR |
engine/models/nemotron_h/community.py, ops/profiles/_nemotron.env, ops/profiles/nemotron-super-vl-exl3-2.0.env and 2 more |
QSE_NH_GGUF |
engine/models/nemotron_h/load.py, ops/profiles/nemotron-bolt-iq2.env, server/diskcache.py |
QSE_NH_HEAD |
engine/models/nemotron_h/community.py, ops/profiles/_nemotron.env, ops/profiles/nemotron-bolt-exl3-2.0.env and 10 more |
QSE_NH_JOINT_ROWS8 |
engine/models/nemotron_h/serving.py |
QSE_NH_LILICORR |
engine/models/nemotron_h/serving.py, ops/profiles/nemotron-bolt-exl3-2.0.env, ops/profiles/nemotron-bolt-exl3-2.5.env and 8 more |
QSE_NH_LILICORR_PACK |
engine/drafters/lilicorr.py |
QSE_NH_LILICORR_PACK_GRAPHS |
engine/drafters/lilicorr.py |
QSE_NH_LOOKUP |
engine/models/nemotron_h/serving.py |
QSE_NH_MODEL |
engine/models/nemotron_h/serving.py, ops/profiles/_nemotron.env, ops/profiles/nemotron-bolt-exl3-2.0.env and 11 more |
QSE_NH_VARIANT |
engine/models/nemotron_h/community.py, ops/profiles/_nemotron.env, ops/profiles/nemotron-bolt-exl3-2.0.env and 11 more |
QSE_NO_ROLLBACK |
ops/use-profile.sh, server/setup/apply.py |
QSE_OPS_DIR |
ops/start.sh, server/setup/apply.py, server/setup/device.py |
QSE_ORN_DFLASH2 |
engine/models/qwen3_5_moe/serving.py, ops/profiles/_ornith.env, ops/profiles/ornith-1.5-122b-exl3-4.0.env |
QSE_ORN_DRAFTER |
ops/profiles/_ornith.env |
QSE_ORN_HEAD |
ops/profiles/_ornith.env, ops/profiles/ornith-1.5-122b-exl3-4.0.env |
QSE_ORN_VARIANT |
ops/profiles/_ornith.env, ops/profiles/ornith-1.5-122b-exl3-4.0.env |
QSE_PACK_VISION |
engine/models/nemotron_h/community.py, engine/models/qwen3_5/community.py, ops/serve.env and 2 more |
QSE_PAGE_CACHE_DROP |
engine/models/nemotron_h/community.py, ops/profiles/_flashnext.env, ops/profiles/_flint.env and 6 more |
QSE_PAGE_CACHE_KEEP |
ops/profiles/_flashnext.env, ops/serve.env, server/remote/overlay.py |
QSE_PARALLEL |
ops/engines.sh, ops/profiles/_flashnext.env, ops/profiles/_flint.env and 6 more |
QSE_PARALLEL_AUTO_N |
server/app.py, server/slots.py |
QSE_PARALLEL_BATCH |
server/app.py |
QSE_PARALLEL_BUSY_WAIT_MS |
server/slots.py |
QSE_PARALLEL_CUT |
server/app.py |
QSE_PARALLEL_DRAFT |
server/app.py |
QSE_PARALLEL_DRAFT_ROWS |
server/app.py |
QSE_PARALLEL_FLOOR_GB |
engine/kvpool.py, ops/profiles/_flashnext.env, ops/serve.env and 1 more |
QSE_PARALLEL_LAYER_YIELD |
server/app.py |
QSE_PARALLEL_MAX_ROWS |
server/app.py |
QSE_PARALLEL_ROW_CAP |
server/app.py |
QSE_PARALLEL_SWITCH_MS |
server/app.py |
QSE_PARALLEL_TIMING |
server/app.py |
QSE_PARALLEL_WINDOW_ROWS |
server/app.py |
QSE_PEAK_STRICT |
engine/cache.py, server/app.py |
QSE_PEER |
ops/engines.sh, ops/start.sh, ops/stop.sh |
QSE_PEER_DISK_CACHE |
ops/engines.sh |
QSE_PEER_ENGINE_FLAGS |
ops/engines.sh |
QSE_PEER_FLOOR_GB |
ops/engines.sh |
QSE_PEER_HOST |
ops/engines.sh |
QSE_PEER_NEED_GB |
ops/engines.sh, ops/start.sh |
QSE_PEER_PORT |
ops/engines.sh, ops/start.sh |
QSE_PREFILL_BURST |
server/app.py, server/prefill_burst.py |
QSE_PREFILL_BURST_ASKING |
server/app.py |
QSE_PREFILL_BURST_WAIT_S |
server/app.py |
QSE_PRICE_TABLE |
server/app.py |
QSE_PROFILE_NAME |
ops/serve.env, ops/start.sh, server/remote/cli.py and 1 more |
QSE_PROFILE_OVERRIDE |
ops/serve.env, ops/use-profile.sh, server/setup/apply.py and 1 more |
QSE_RECLAIM_GRAPHS |
server/app.py |
QSE_REQ_PEAK |
server/app.py |
QSE_ROLES_DYNAMIC |
ops/serve.env, server/rolemem.py |
QSE_ROLE_MAX_HOLD_S |
ops/serve.env, server/rolemem.py |
QSE_ROLE_RELOAD_MARGIN_GB |
ops/serve.env, server/rolemem.py |
QSE_ROLE_RELOAD_S |
ops/serve.env, server/rolemem.py |
QSE_ROLE_THRASH_S |
ops/serve.env, server/rolemem.py |
QSE_ROLE_TICK_FAKE |
server/app.py |
QSE_ROLE_TICK_S |
server/rolemem.py |
QSE_ROLE_UNLOAD_LAYA |
ops/serve.env, server/app.py, server/rolemem.py |
QSE_ROUND_RUNNER |
server/round_runner.py |
QSE_ROUND_WALK |
server/app.py |
QSE_RUNNER_BUSY_WORK |
ops/profiles/_flashnext.env, server/api/live.py, server/health_cache.py and 1 more |
QSE_RUNNER_COMMIT_MULTI |
server/round_runner.py |
QSE_RUNNER_COUNTERS |
engine/models/flashnext/serving.py, server/round_runner.py |
QSE_RUNNER_GC |
engine/cache.py, ops/profiles/_flashnext.env, server/app.py and 1 more |
QSE_RUNNER_GC_FULL_ROUNDS |
ops/profiles/_flashnext.env, server/round_runner.py |
QSE_RUNNER_HOST_BLOCKS |
engine/models/flashnext/serving.py, server/round_runner.py |
QSE_RUNNER_JOINT1 |
engine/model.py, engine/models/qwen3_5/joint_graphs.py, server/round_runner.py |
QSE_RUNNER_JOINT_PRICE |
server/round_runner.py |
QSE_RUNNER_LANE_MARGIN |
server/round_runner.py |
QSE_RUNNER_LANE_PLAIN |
engine/models/nemotron_h/serving.py, server/round_runner.py |
QSE_RUNNER_LANE_PROBE |
server/round_runner.py |
QSE_RUNNER_MTP_PACK |
server/round_runner.py |
QSE_RUNNER_PIPE |
engine/models/flashnext/engine.py, server/round_runner.py |
QSE_RUNNER_PRICED_CUT |
server/round_runner.py |
QSE_RUNNER_PRICED_EXPLORE |
server/round_runner.py |
QSE_RUNNER_ROWCAP |
server/app.py |
QSE_RUNNER_ROWCOPY |
engine/rowcopy.py, server/round_runner.py |
QSE_RUNNER_SUPERVISED |
server/round_runner.py |
QSE_RUNNER_VERDICT_WAIT_MS |
server/round_runner.py |
QSE_SECRETS |
install.sh, ops/make-secrets.sh, ops/start.sh and 3 more |
QSE_SESSION_S |
server/api/auth.py |
QSE_SETUP_ENV |
ops/serve.env, server/setup/apply.py |
QSE_SETUP_ENV_FILE |
ops/serve.env, server/remote/overlay.py, server/setup/doctor.py |
QSE_SETUP_MODE |
ops/start.sh |
QSE_SIDE_GRAPHS |
ops/serve.env, server/app.py |
QSE_SIDE_MAX_LEN |
ops/profiles/_flashnext.env, ops/serve.env, server/app.py and 2 more |
QSE_SIDE_RESIDENT_GB |
server/app.py |
QSE_SIDE_SLOT_CONST_GB |
ops/serve.env, server/app.py |
QSE_SSH_PORT |
ops/serve.env, server/pairing/hostkey.py, server/remote/overlay.py |
QSE_STACK_SAMPLE |
server/app.py, server/stack_sampler.py |
QSE_STATE_DIR |
ops/make-secrets.sh, ops/start.sh, server/setup/apply.py and 2 more |
QSE_STATUS_FILE |
ops/engine_status.py |
QSE_STATUS_HOST |
ops/engine_status.py, ops/hold.sh, ops/serve.env and 1 more |
QSE_STATUS_KEEP |
ops/hold.sh, ops/stop.sh |
QSE_STATUS_PORT |
install.sh, ops/engine_status.py, ops/engines.sh and 4 more |
QSE_STATUS_TREE |
install.sh, ops/engine_status.py, ops/hold.sh |
QSE_SUFFIX_RAW |
server/app.py |
QSE_TEST |
server/api/ledger.py, server/app.py |
QSE_USAGE_CONTEXT |
server/app.py |
QSE_USAGE_DEFAULT |
server/app.py |
QSE_USAGE_LEDGER |
install.sh, ops/engines.sh, ops/serve.env and 4 more |
QSE_VISION_ASKED |
ops/serve.env, ops/start.sh, server/remote/overlay.py |
QSE_WEIGHTS |
engine/models/flashnext/corrector.py, engine/models/flashnext/coupled.py, engine/models/flashnext/serving.py and 20 more |
QSE_WEIGHTS_ALSO |
ops/profiles/_flashnext.env, ops/use-profile.sh |
QSE_WEIGHTS_DIR |
engine/models/nemotron_h/community.py, engine/models/qwen3_5/community.py, ops/profiles/_exl3.env and 12 more |
QSE_WEIGHTS_FILES |
engine/models/nemotron_h/community.py, engine/models/qwen3_5/community.py, ops/profiles/_exl3.env and 9 more |
QSE_WEIGHTS_GB |
engine/models/nemotron_h/community.py, engine/models/qwen3_5/community.py, ops/profiles/exl3-2.5.env and 35 more |
QSE_WEIGHTS_HF |
engine/models/nemotron_h/community.py, engine/models/qwen3_5/community.py, ops/profiles/_exl3.env and 27 more |
QSE_WEIGHTS_REVISION |
engine/models/nemotron_h/community.py, engine/models/qwen3_5/community.py, ops/profiles/_flashnext.env and 30 more |
QSE_WREN_MODEL |
server/setup/wren.py |
The most used: QSE_CONDUCTOR_CONFIG, QSE_USAGE_LEDGER, QSE_DISK_CACHE and QSE_BACKENDS name the conductor's directory, the ledger, the disk tier's directory and the proxy's config database; ops/start.sh passes them to the service only, so a benchmark server started with serve.env exported never writes them. QSE_DASHBOARD_LOGIN turns the dashboard login on. QSE_ADMIN_TOKEN and QSE_METRICS_TOKEN live in secrets.env, which ops/make-secrets.sh writes.
The wait for memory (Waiting for memory) reads these, with their code defaults: QSE_MEM_WAIT_S (10,800) is the longest wait in seconds of a streamed request, or of one that sends wait_for_memory: true; QSE_PARALLEL_FLOOR_GB (15) is the floor the admission keeps MemAvailable at, the same as --parallel-floor-gb; QSE_KV_HARD_FLOOR_GB (10) is the floor the KV pool's commits keep, the same as --kv-hard-floor-gb; QSE_GUARD_MARGIN_GB (5) is the margin each floor holds under the largest step the engine takes between two checks; QSE_KV_ADMIT_ROWS (8,192) is the part of max_tokens the admission reserves KV pages for, 0 for all of it; QSE_KV_RECOMPUTE (on) lets a lane the pool refuses give way and go on later; QSE_PEAK_STRICT (0) set to 1 makes a lone request that the memory gate refuses wait too.
TANDEM variables¶
TANDEM_* variables the engine's code reads: the installer, the service scripts and the server¶
| Variable | Read in |
|---|---|
TANDEM_ADMIN_TOKEN |
read in server/api/auth.py, server/setup/cli.py |
TANDEM_DIR |
read in install.sh, server/setup/cli.py |
TANDEM_HEALTH_TIMEOUT |
read in install.sh |
TANDEM_HOST |
read in install.sh |
TANDEM_LINK_KEY |
read in ops/engines.sh, ops/serve.env |
TANDEM_LINK_LISTEN |
read in ops/engines.sh, ops/serve.env |
TANDEM_LINK_PEER |
read in ops/engines.sh, ops/serve.env |
TANDEM_MIN_FREE_GIB |
read in install.sh |
TANDEM_MODEL_NAME |
read in install.sh |
TANDEM_NODE_BIN |
read in install.sh |
TANDEM_NODE_PORT |
read in install.sh, ops/start.sh |
TANDEM_PORT |
read in install.sh |
TANDEM_REF |
read in install.sh |
TANDEM_REPO |
read in install.sh |
TANDEM_SKIP_MEMCHECK |
read in install.sh, server/remote/lifecycle.py |
TANDEM_STAGE |
read in install.sh, ops/engines.sh |
TANDEM_STAGE_CACHES |
read in ops/engines.sh, ops/serve.env |
TANDEM_STAGE_CACHE_DIR |
read in ops/engines.sh, ops/serve.env |
TANDEM_STAGE_CACHE_GB |
read in ops/engines.sh, ops/serve.env |
TANDEM_STAGE_SNAPSHOTS_GB |
read in ops/engines.sh, ops/serve.env |
TANDEM_TRUSTED_PROXY |
read in server/api/auth.py, server/app.py |
TANDEM_URL |
read in server/setup/cli.py |
The conductor's documents¶
| document | what it holds | edited on |
|---|---|---|
profiles |
one document per profile: mode, decisions, task modes and caps, thinking, ui, outcomes, library, memory, shadow |
Profiles |
router |
the classes and their policy, the classifier stages and thresholds, the structural rules, side shapes, outcomes, library | Routing |
packages |
every package: source, switch, limits, flags | Packages |
roles |
every role's contract | Profiles, Roles |
settings |
the default profile, transcripts, the disk tier, memory notes and the writer | Memory, Profiles |
deciders |
the decision models, their questions, Laya for scripts | Deciders |
residency |
the floor, the margin, the settle time, the load order, the claims | Profiles, Residency |
monitor |
the session monitor's policy (version 0 is ops/supervisor.yaml) |
Monitor |
harness |
the harness's settings per backend, behind the proxy | Backends |
The conductor's versioned documents.
A document is JSON, checked against the conductor's schema (GET /v1/conductor/meta serves it, and TandemApp's contracts/conductor-v1/ holds a copy). A version stored by an older release loads with defaults for every field it lacks, and the journal says so once.