Oracle Cloud Infrastructure Documentation

Cohere Command R+ 08-2024

Review performance benchmarks for the cohere.command-r-plus-08-2024 (Cohere Command R+ 08-2024) model hosted on one Large Cohere V2_2 unit of a dedicated AI cluster in OCI Generative AI.

  • See details for the model and review the following sections:
    • Available regions for this model.
    • Dedicated AI clusters for hosting this model.
  • Review the metrics.

Random Length

This scenario mimics text generation use cases where the size of the prompt and response are unknown ahead of time. Because of the unknown prompt and response lengths, we've used a stochastic approach where both the prompt and response length follow a normal distribution. The prompt length follows a normal distribution with a mean of 480 tokens and a standard deviation of 240 tokens. The response length follows a normal distribution with a mean of 300 tokens and a standard deviation of 150 tokens.

The cohere.command-r-plus-08-2024 model hosted on one Large Cohere V2_2 unit of a dedicated AI cluster for all regions except for the Saudi Arabia Central (Riyadh) region.
Concurrency Time to First Token (TTFT)(second) Token-level Inference Speed (tokens/second) Token-level Throughput (tokens/second) Request-level Latency (second) Request-level Throughput (Request per second) (RPS) Total Throughput (tokens/second)
1 0.58 59.40 51.39 3.97 0.25 177.94
2 0.54 56.82 99.37 4.07 0.49 343.94
4 0.55 53.54 187.30 4.29 0.92 629.78
8 0.55 45.14 321.74 4.98 1.57 1,086.90
16 0.58 37.45 527.84 6.06 2.52 1,803.26
32 0.71 31.06 835.95 7.05 4.17 2,930.36
64 1.12 21.23 1,013.92 9.91 5.36 3,690.25
128 2.46 12.27 1,013.50 15.17 6.28 4,184.62
256 28.89 11.97 711.27 41.84 4.49 2,995.92
The cohere.command-r-plus-08-2024 model hosted on one Large Cohere V2_2 unit of a dedicated AI cluster for the Saudi Arabia Central (Riyadh) region.
Concurrency Time to First Token (TTFT)(second) Token-level Inference Speed (tokens/second) Token-level Throughput (tokens/second) Request-level Latency (second) Request-level Throughput (Request per second) (RPS) Total Throughput (tokens/second)
1 0.67 68.26 57.04 4.56 0.22 159.08
2 0.68 66.98 113.61 4.65 0.43 320.69
4 0.69 63.73 217.25 4.94 0.80 576.20
8 0.69 59.32 403.93 5.15 1.53 1,175.95
16 0.70 54.51 737.83 5.63 2.75 2,092.64
32 0.74 45.65 1,213.09 6.57 4.55 3,306.18
64 0.94 35.81 1,791.62 8.42 6.71 4,945.78
128 1.81 24.20 2,030.12 11.87 8.22 5,947.53
256 9.35 23.53 2,064.19 19.82 8.42 6,063.79

Chat

This scenario covers chat and dialog use cases where the prompt and responses are short. The prompt and response length are each fixed to 100 tokens.

The cohere.command-r-plus-08-2024 model hosted on one Large Cohere V2_2 unit of a dedicated AI cluster for all regions except for the Saudi Arabia Central (Riyadh) region.
Concurrency Time to First Token (TTFT)(second) Token-level Inference Speed (tokens/second) Token-level Throughput (tokens/second) Request-level Latency (second) Request-level Throughput (Request per second) (RPS) Total Throughput (tokens/second)
1 0.51 59.64 44.86 1.99 0.50 93.29
2 0.50 57.81 87.16 2.02 0.98 182.21
4 0.51 54.31 164.73 2.12 1.86 344.89
8 0.52 46.73 290.36 2.43 3.22 601.40
16 0.57 40.77 496.24 2.76 5.52 1,031.61
32 0.97 34.74 786.28 3.53 8.77 1,634.87
64 1.03 25.08 1,074.45 4.57 12.13 2,249.96
128 1.93 17.46 1,402.86 7.15 15.58 2,908.96
256 6.39 14.68 1,321.52 12.48 14.82 2,754.62
The cohere.command-r-plus-08-2024 model hosted on one Large Cohere V2_2 unit of a dedicated AI cluster for the Saudi Arabia Central (Riyadh) region.
Concurrency Time to First Token (TTFT)(second) Token-level Inference Speed (tokens/second) Token-level Throughput (tokens/second) Request-level Latency (second) Request-level Throughput (Request per second) (RPS) Total Throughput (tokens/second)
1 0.64 69.01 47.06 1.97 0.50 96.12
2 0.63 67.99 92.76 1.99 1.00 189.46
4 0.64 66.71 182.12 2.03 1.95 370.65
8 0.66 64.71 345.04 2.09 3.70 704.19
16 0.70 61.79 658.79 2.19 7.06 1,343.09
32 0.76 57.74 1,196.06 2.37 12.79 2,437.58
64 1.01 50.01 1,861.32 2.86 19.97 3,798.59
128 1.53 37.88 2,266.40 4.00 24.28 4,611.58
256 3.56 35.73 2,753.63 6.15 29.50 5,616.33

Generation Heavy

This scenario is for generation and model response heavy use cases. For example, a long job description generated from a short bullet list of items. For this case, the prompt length is fixed to 100 tokens and the response length is fixed to 1,000 tokens.

The cohere.command-r-plus-08-2024 model hosted on one Large Cohere V2_2 unit of a dedicated AI cluster for all regions except for the Saudi Arabia Central (Riyadh) region.
Concurrency Time to First Token (TTFT)(second) Token-level Inference Speed (tokens/second) Token-level Throughput (tokens/second) Request-level Latency (second) Request-level Throughput (Request per second) (RPS) Total Throughput (tokens/second)
1 0.53 61.54 58.82 12.20 0.08 66.68
2 0.53 59.46 113.88 14.32 0.14 127.24
4 0.52 56.83 216.43 13.34 0.29 244.89
8 0.52 50.35 383.52 14.16 0.55 436.81
16 0.55 45.71 681.87 16.28 0.94 772.85
32 0.61 42.04 1,203.12 16.03 1.81 1,378.76
64 0.85 32.83 1,870.11 19.12 2.99 2,159.32
128 1.72 24.80 2,728.79 27.59 3.96 3,112.68
256 17.43 24.51 2,532.47 43.73 3.69 2,889.44
The cohere.command-r-plus-08-2024 model hosted on one Large Cohere V2_2 unit of a dedicated AI cluster for the Saudi Arabia Central (Riyadh) region.
Concurrency Time to First Token (TTFT)(second) Token-level Inference Speed (tokens/second) Token-level Throughput (tokens/second) Request-level Latency (second) Request-level Throughput (Request per second) (RPS) Total Throughput (tokens/second)
1 0.65 66.69 62.92 14.14 0.07 69.68
2 0.66 66.40 123.84 14.32 0.14 137.02
4 0.66 64.72 244.79 14.51 0.27 271.17
8 0.67 63.87 486.70 14.90 0.53 538.49
16 0.71 60.49 915.45 15.54 1.02 1,014.47
32 0.77 57.09 1,708.15 16.57 1.89 1,891.81
64 1.02 49.88 2,943.89 18.90 3.30 3,263.78
128 1.56 40.01 4,514.92 23.89 5.05 5,003.86
256 16.89 39.84 4,740.55 39.50 5.26 5,250.54

RAG

The retrieval-augmented generation (RAG) scenario has a very long prompt and a short response such as summarizing use cases. The prompt length is fixed to 2,000 tokens and the response length is fixed to 200 tokens.

The cohere.command-r-plus-08-2024 model hosted on one Large Cohere V2_2 unit of a dedicated AI cluster for all regions except for the Saudi Arabia Central (Riyadh) region.
Concurrency Time to First Token (TTFT)(second) Token-level Inference Speed (tokens/second) Token-level Throughput (tokens/second) Request-level Latency (second) Request-level Throughput (Request per second) (RPS) Total Throughput (tokens/second)
1 0.68 58.51 48.56 4.08 0.24 533.48
2 0.77 55.59 91.08 4.35 0.46 1,001.07
4 0.98 51.26 162.12 4.87 0.81 1,784.54
8 1.52 42.35 251.89 6.25 1.26 2,768.79
16 2.30 32.91 371.92 8.47 1.86 4,093.10
32 3.91 28.74 555.48 11.29 2.78 6,108.24
64 5.98 15.69 627.51 19.00 3.14 6,898.71
128 8.99 7.11 577.71 36.95 2.90 6,359.07
256 31.42 6.27 541.87 63.18 2.72 5,975.51
The cohere.command-r-plus-08-2024 model hosted on one Large Cohere V2_2 unit of a dedicated AI cluster for the Saudi Arabia Central (Riyadh) region.
Concurrency Time to First Token (TTFT)(second) Token-level Inference Speed (tokens/second) Token-level Throughput (tokens/second) Request-level Latency (second) Request-level Throughput (Request per second) (RPS) Total Throughput (tokens/second)
1 0.76 69.52 54.09 3.46 0.29 625.42
2 0.82 67.31 103.91 3.62 0.55 1,200.85
4 0.94 63.19 189.23 3.91 1.01 2,198.58
8 1.18 57.13 331.56 4.48 1.77 3,858.68
16 1.70 48.48 524.40 5.64 2.79 6,090.47
32 2.68 38.31 735.33 7.84 3.90 8,523.79
64 4.09 22.66 894.11 12.58 4.73 10,335.98
128 6.69 10.61 819.11 24.39 4.35 9,499.74
256 21.64 9.16 803.95 42.18 4.25 9,287.02