List your company

ResearchFormerly WizWand

MT-Bench

Benchmarks

TaskDatasetSOTA resultTrendResults

Multi-round conversation

MT-Bench

274.32Tokens Per Second

40

Multi-turn Dialogue Generation

MT-bench

5.62Mean Accepted Length

39

Multi-turn Conversation

MT-Bench

77.1Win Rate

36

LLM Inference Acceleration

MT-bench

5.582Mean Accepted Length (tau)

36

LLM-as-a-judge evaluation

MT-Bench

0.689Pearson's r

36

Instruction Following

MT-Bench (test)

8.33Overall Score

35

Multi-turn Dialogue

MT-Bench

8.9GPT-4 Score

34

LLM Judge Agreement

MT-bench First Turn

0.97Agreement Rate

34

Long-form Dialogue

MT-Bench+

90.5Quality Score

32

Judge Agreement

MT-bench Second Turn

95Agreement

32

Multi-turn Dialogue

MT-Bench

6.93MAT Score

30

Speculative Sampling

MT-Bench

4.13Average Acceptance Length

28

Conversational Ability

MT-Bench

7.58MT-Bench Score

28

Code generation

MT-Bench (test)

3.934Speedup Ratio

26

Multi-turn instruction following

MT-Bench High-Variance (Top 20%)

7.54Reward Score

26

Multi-turn Conversation Evaluation

MT-bench

3.82Speedup

25

self-affirmation

MT-Bench-101

0.58Success Rate

25

Multi-turn dialogue

MT-Bench

5.78Acceptance Length (τ)

20

Human-Preference Alignment

MT-Bench

6.83SAG

20

Multi-turn Dialogue

MT-Bench

270.96Tokens/s

20

Output Diversity

MT-Bench

48.36Lexical Diversity Score

20

Response Quality Evaluation

MT-Bench

8.71Average Response Quality

19

MT-Bench 1.0 (test)

8MT-Bench Score

19

Open-ended dialogue

MT-Bench 80 questions

5.18Speed-up Ratio

18

MT-Bench

5.62Fluency

18