Skip to content
Mock data. Run values, controls, targets and reproductions on this preview are invented for layout. Claim-card text, upstream figures, the reference price and the RFC list come from the repository.
Inference ProofsMenu

Workload, native division

fp-maverick-w1024-b24

FlashProve's own model and shape: the system's published steady-state shape. A workload fixes what is proved: the model by content hash, the precision, the prompt set, the shape, the decoding rule and the phase.

Definition

RFC-0003 §v0.1 native workloads

Version
1
Division
native
System
flashprove
Model
Llama 4 Maverick, public 4-bit conversion
Precision
4-bit
Shape
1,024 prompt + 1 output tokens, greedy
Batch
24 requests per batch, 5 batches per session, 120 prompts per job
Phase
prefill
Model digest
not pinned yet: filled from the public model repository in the implementing issue; the schema rejects a workload without it (RFC-0003)
Prompt corpus digest
not pinned yet: filled from the public model repository in the implementing issue; the schema rejects a workload without it (RFC-0003)
Active parameters per token
not pinned yet: filled from the public model repository in the implementing issue; the schema rejects a workload without it (RFC-0003)

Configurations on this workload

Claim summary first, then rows.

relation:
The output token is the top choice of the integer model whose weights were committed under the given commitment ID, on the given 1,024-token window.
security.notion:
knowledge_soundness
coverage:
prefill, request
residual:
proves the integer model, which differs from the 16-bit model's top token at some positions

flashprove / exact-match-lean on fp-maverick-w1024-b24

profile prover-h100tier reproduced
Measured

setup_seconds

120.7 s

scope setupgrade measuredtier reproducedmock

mean 120.7, sd 0.7506, min 119.9, max 121.4, n 3, n_failed 0

mean 120.7 srepetition 1: 121.4 srepetition 2: 119.9 srepetition 3: 120.7 ssource: mock harness record

commit_seconds

2,411 s

scope commitgrade measuredtier reproducedmock

mean 2,411, sd 12.35, min 2,398, max 2,423, n 3, n_failed 0

mean 2,411 srepetition 1: 2410.5 srepetition 2: 2398.2 srepetition 3: 2422.9 ssource: mock harness record

prove_session_seconds

1,186 s

scope prove.sessiongrade measuredtier reproducedmock

mean 1,186, sd 6.304, min 1,179, max 1,192, n 3, n_failed 0

mean 1,186 srepetition 1: 1185.3 srepetition 2: 1192 srepetition 3: 1179.4 ssource: mock harness record

prove_steady_seconds

887.9 s

scope prove.steadygrade measuredtier reproducedmock

mean 887.9, sd 6.678, min 881.6, max 894.9, n 3, n_failed 0

mean 887.9 srepetition 1: 887.2 srepetition 2: 894.9 srepetition 3: 881.6 ssource: mock harness record

whole_wall_seconds

3,596 s

scope whole_wallgrade measuredtier reproducedmock

mean 3,596, sd 6.056, min 3,590, max 3,602, n 3, n_failed 0

mean 3,596 srepetition 1: 3595.8 srepetition 2: 3590.2 srepetition 3: 3602.3 ssource: mock harness record

tokens_covered_total

122,900 token

grade measuredtier reproducedmock

mean 122,900, sd 0, min 122,900, max 122,900, n 3, n_failed 0

source: mock harness record

tokens_covered_steady

98,300 token

grade measuredtier reproducedmock

mean 98,300, sd 0, min 98,300, max 98,300, n 3, n_failed 0

source: mock harness record

requests_count

120 count

grade measuredtier reproducedmock

mean 120, sd 0, min 120, max 120, n 3, n_failed 0

source: mock harness record

proofs_count

120 count

grade measuredtier reproducedmock

mean 120, sd 0, min 120, max 120, n 3, n_failed 0

source: mock harness record

proof_bytes_total

2.254e9 B

grade measuredtier reproducedmock

mean 2.254e9, sd 602,800, min 2.253e9, max 2.254e9, n 3, n_failed 0

mean 2.254e9 Brepetition 1: 2253600000 Brepetition 2: 2254100000 Brepetition 3: 2252900000 Bsource: mock harness record

verify_seconds

3.120 s

scope verifygrade measuredtier reproducedmock

mean 3.120, sd 0.03000, min 3.090, max 3.150, n 3, n_failed 0

mean 3.120 srepetition 1: 3.12 srepetition 2: 3.09 srepetition 3: 3.15 ssource: mock harness record

verify_seconds_total

374.4 s

scope verifygrade measuredtier reproducedmock

mean 374.4, sd 3.600, min 370.8, max 378, n 3, n_failed 0

mean 374.4 srepetition 1: 374.4 srepetition 2: 370.8 srepetition 3: 378 ssource: mock harness record

verify_single_thread_seconds

21.14 s

scope verify.single_threadgrade measuredtier reproducedmock

mean 21.14, sd 0.03000, min 21.11, max 21.17, n 3, n_failed 0

mean 21.14 srepetition 1: 21.14 srepetition 2: 21.11 srepetition 3: 21.17 ssource: mock harness record
Derived (computed by the derivation code)

rate_whole_wall

34.17 token/s

scope whole_wallgrade derivedtier reproducedmock

mean 34.17, sd 0.05753, min 34.11, max 34.23, n 3, n_failed 0

mean 34.17 token/srepetition 1: 34.173202069080595 token/srepetition 2: 34.22650548715949 token/srepetition 3: 34.11153984954057 token/ssource: mock derivation record

rate_prove.session

103.6 token/s

scope prove.sessiongrade derivedtier reproducedmock

mean 103.6, sd 0.5510, min 103.1, max 104.2, n 3, n_failed 0

mean 103.6 token/srepetition 1: 103.66995697291826 token/srepetition 2: 103.08724832214764 token/srepetition 3: 104.18857045955569 token/ssource: mock derivation record

rate_prove.steady

110.7 token/s

scope prove.steadygrade derivedtier reproducedmock

mean 110.7, sd 0.8317, min 109.8, max 111.5, n 3, n_failed 0

mean 110.7 token/srepetition 1: 110.80252479711451 token/srepetition 2: 109.84914515588333 token/srepetition 3: 111.50635208711434 token/ssource: mock derivation record

cost_per_1k_tokens_whole_wall

0.03487 USD/1k_token

scope whole_wallgrade derivedtier reproducedmock

mean 0.03487, sd 5.873e-5, min 0.03482, max 0.03493, n 3, n_failed 0

mean 0.03487 USD/1k_tokenrepetition 1: 0.03487137858072917 USD/1k_tokenrepetition 2: 0.03481707085503472 USD/1k_tokenrepetition 3: 0.034934414333767365 USD/1k_tokensource: mock derivation record

proof_bytes_per_token

18,340 B/token

grade derivedtier reproducedmock

mean 18,340, sd 4.905, min 18,330, max 18,340, n 3, n_failed 0

mean 18,340 B/tokenrepetition 1: 18339.84375 B/tokenrepetition 2: 18343.912760416668 B/tokenrepetition 3: 18334.147135416668 B/tokensource: mock derivation record

verify_seconds_per_token

0.003047 s/token

scope verifygrade derivedtier reproducedmock

mean 0.003047, sd 2.930e-5, min 0.003018, max 0.003076, n 3, n_failed 0

mean 0.003047 s/tokenrepetition 1: 0.0030468749999999997 s/tokenrepetition 2: 0.0030175781250000003 s/tokenrepetition 3: 0.003076171875 s/tokensource: mock derivation record
Runs
  • mock-fp-b24-r1completed
  • mock-fp-b24-r2completed
  • mock-fp-b24-r3completed
  • Superseded, failed and ineligible runs never contribute to statistics (RFC-0009 S7).

Reference row: runs the model without proving, no proof claim (RFC-0015).

ref-serve on fp-maverick-w1024-b24

profile prover-h100tier reported
Measured

whole_wall_seconds

null (not_measurable_on_profile) s

scope whole_wallgrade measuredtier reportedmock
source: mock harness record
Derived (computed by the derivation code)

rate_whole_wall

null (not_measurable_on_profile) token/s

scope whole_wallgrade derivedtier reportedmock
source: mock derivation record
Runs

No runs: the reference system cannot run this workload on the profile.