For Qwen3-8B using the largest fitting batch, calibrated TP=2 reached 2.16× on AIME25 and 2.07× on LB[16k-18k], above the table’s TP=1 runs. The timing model achieved 5–11% MAPE when refit at either TP degree; reusing TP=1 coefficients at TP=2 raised MAPE to 35–50%, left AIME speedup benign at 2.21×, but reduced LB speedup to 1.51×.

Reported and observed measurements

Assessments (0)

No immutable Assessment has been published for this Claim yet.

Runs (0)

No execution has been linked to this Claim yet.

Each row is a measurement with its reported value, latest available observation, and cumulative evidence state.
Metric and conditions	Reported	Latest observation	Evidence judgment
model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 1	30 requests	—	Not assessed
model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 1	1,353 output token/s	—	Not assessed
model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 1	2,553 output token/s	—	Not assessed
model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 1	1.89 × AutoRegressive	—	Not assessed
model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 2	64 requests	—	Not assessed
model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 2	2,494 output token/s	—	Not assessed
model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 2	5,392 output token/s	—	Not assessed
model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 2	2.16 × AutoRegressive	—	Not assessed
model: Qwen3-8B · dataset: AIME25 · calibration: TP=1 coefficients reused at TP=2 · tensorParallelism: 2 stale	64 requests	—	Not assessed
model: Qwen3-8B · dataset: AIME25 · calibration: TP=1 coefficients reused at TP=2 · tensorParallelism: 2 stale	2,494 output token/s	—	Not assessed
model: Qwen3-8B · dataset: AIME25 · calibration: TP=1 coefficients reused at TP=2 · tensorParallelism: 2 stale	5,509 output token/s	—	Not assessed
model: Qwen3-8B · dataset: AIME25 · calibration: TP=1 coefficients reused at TP=2 · tensorParallelism: 2 stale	2.21 × AutoRegressive	—	Not assessed

Metric and conditions

Reported

Latest observation

Evidence judgment

model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 1

30 requests

—

Not assessed

model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 1

1,353 output token/s

—

Not assessed

model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 1

2,553 output token/s

—

Not assessed

model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 1

1.89 × AutoRegressive

—

Not assessed

model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 2

64 requests

—

Not assessed

model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 2

2,494 output token/s

—

Not assessed

model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 2

5,392 output token/s

—

Not assessed

model: Qwen3-8B · dataset: AIME25 · calibration: fitted for this TP degree · tensorParallelism: 2

2.16 × AutoRegressive

—

Not assessed

model: Qwen3-8B · dataset: AIME25 · calibration: TP=1 coefficients reused at TP=2 · tensorParallelism: 2 stale

64 requests

—

Not assessed

model: Qwen3-8B · dataset: AIME25 · calibration: TP=1 coefficients reused at TP=2 · tensorParallelism: 2 stale

2,494 output token/s

—

Not assessed

model: Qwen3-8B · dataset: AIME25 · calibration: TP=1 coefficients reused at TP=2 · tensorParallelism: 2 stale

5,509 output token/s

—

Not assessed

model: Qwen3-8B · dataset: AIME25 · calibration: TP=1 coefficients reused at TP=2 · tensorParallelism: 2 stale

2.21 × AutoRegressive

—

Not assessed

Reported and observed measurements

Assessments (0)

Runs (0)

Measurement results