Add H100 benchmark results

[mgoin]

Kernel Benchmarks

• Device: H100
• Problem: dense n8192 k8192 channelwise
• Activation Type: float16
• Metric: TFLOPS
• Details: https://github.com/inclusionAI/humming/tree/main/benchmarks/results/h100/dense_n8192_k8192_g0

| shape_m | torch-w16a16 | marlin-w4a16 | humming-w4a16 | marlin-w8a16 | humming-w8a16 | humming-w1a16 | humming-w2a16 | humming-w3a16 | humming-w5a16 | humming-w6a16 | humming-w7a16 |                                                                                                     
|---------:|--------------:|--------------:|---------------:|--------------:|---------------:|---------------:|---------------:|---------------:|---------------:|---------------:|---------------:|                                                                                                    
|       1 |         2.30 |         4.98 |          4.96 |         3.46 |          3.26 |          6.27 |          5.99 |          5.59 |          4.28 |          3.82 |          3.48 |
|       2 |         4.63 |        10.02 |          9.92 |         6.87 |          6.48 |         12.66 |         12.11 |         11.13 |          8.48 |          7.64 |          6.93 |
|       4 |         9.26 |        19.85 |         19.65 |        13.75 |         13.04 |         25.05 |         24.22 |         22.46 |         16.95 |         15.40 |         13.93 |
|       8 |        18.44 |        39.62 |         39.21 |        27.45 |         25.91 |         50.01 |         47.90 |         44.84 |         33.78 |         30.76 |         27.59 |
|      16 |        36.80 |        78.46 |         77.11 |        55.01 |         50.96 |         94.55 |         93.02 |         85.25 |         66.54 |         60.59 |         54.57 |
|      32 |        73.32 |       122.18 |        138.57 |        94.58 |         97.25 |        154.23 |        150.90 |        138.55 |        125.67 |        116.86 |        104.33 |
|      64 |       146.37 |       199.43 |        233.59 |       165.51 |        187.37 |        254.94 |        253.53 |        239.38 |        214.90 |        206.16 |        195.58 |
|     128 |       288.74 |       304.20 |        342.23 |       262.33 |        308.94 |        351.01 |        338.15 |        319.05 |        306.43 |        302.34 |        291.30 |
|     256 |       541.23 |       350.24 |        382.56 |       280.53 |        297.00 |        410.76 |        402.22 |        375.45 |        342.58 |        331.39 |        310.53 |
|     512 |       693.81 |       397.18 |        452.94 |       327.69 |        322.54 |        479.21 |        479.62 |        445.02 |        413.71 |        388.13 |        351.09 |
|    1024 |       698.58 |       412.62 |        500.23 |       336.15 |        469.13 |        507.39 |        513.39 |        483.57 |        458.68 |        458.53 |        428.26 |
|    2048 |       676.56 |       412.39 |        486.49 |       329.47 |        434.81 |        520.88 |        517.50 |        475.64 |        443.53 |        437.51 |        411.85 |
|    4096 |       669.48 |       408.80 |        487.60 |       326.02 |        452.61 |        525.28 |        515.19 |        470.66 |        447.35 |        446.75 |        408.97 |
|    8192 |       652.29 |       407.69 |        488.75 |       326.04 |        458.46 |        523.64 |        511.39 |        472.38 |        443.93 |        447.29 |        413.32 |
|   16384 |       648.47 |       406.32 |        487.72 |       324.09 |        458.38 |        522.12 |        510.96 |        469.75 |        443.62 |        440.59 |        411.49 |

• Device: H100
• Problem: dense n8192 k8192 channelwise
• Activation Type: float8e4m3
• Metric: TFLOPS
• Details: https://github.com/inclusionAI/humming/tree/main/benchmarks/results/h100/dense_n8192_k8192_g0

| shape_m | cutlass-w8a8 | humming-w8a8 | humming-w7a8 | humming-w6a8 | humming-w5a8 | humming-w4a8 | humming-w3a8 | humming-w2a8 | humming-w1a8 |
|---------:|--------------:|--------------:|--------------:|--------------:|--------------:|--------------:|--------------:|--------------:|--------------:|                  
|       1 |         3.75 |         3.53 |         3.75 |         4.20 |         4.60 |         5.50 |         6.24 |         7.09 |         7.66 |  
|       2 |         7.45 |         7.06 |         7.57 |         8.38 |         9.27 |        10.92 |        12.60 |        14.20 |        15.53 |  
|       4 |        14.88 |        14.16 |        15.14 |        16.77 |        18.36 |        22.01 |        24.95 |        28.14 |        30.64 |  
|       8 |        29.75 |        28.22 |        30.26 |        33.53 |        36.70 |        43.54 |        49.84 |        56.24 |        61.15 |  
|      16 |        59.49 |        56.14 |        59.44 |        66.82 |        72.72 |        86.46 |        97.89 |       110.75 |       118.56 |  
|      32 |       118.08 |       112.19 |       120.66 |       133.76 |       146.66 |       164.84 |       178.30 |       204.37 |       219.38 |  
|      64 |       235.36 |       218.49 |       232.15 |       245.44 |       255.95 |       286.38 |       315.88 |       345.87 |       365.91 |  
|     128 |       448.94 |       356.27 |       354.26 |       368.53 |       367.65 |       392.03 |       394.68 |       427.52 |       454.56 |  
|     256 |       839.98 |       345.99 |       383.42 |       433.96 |       441.98 |       521.18 |       509.60 |       583.39 |       625.69 |  
|     512 |      1216.94 |       446.98 |       506.50 |       567.31 |       528.91 |       625.68 |       598.56 |       654.62 |       697.90 |  
|    1024 |      1312.05 |       574.59 |       585.12 |       663.14 |       606.33 |       711.74 |       653.61 |       711.31 |       750.16 |  
|    2048 |      1377.33 |       762.33 |       664.29 |       680.85 |       600.74 |       716.15 |       677.91 |       736.94 |       781.69 |  
|    4096 |      1286.52 |       772.75 |       694.11 |       695.45 |       602.83 |       705.75 |       676.08 |       742.06 |       796.91 |  
|    8192 |      1268.03 |       799.39 |       684.71 |       695.55 |       602.19 |       706.61 |       670.29 |       747.73 |       804.51 |  
|   16384 |      1269.79 |       782.13 |       693.12 |       698.25 |       602.65 |       710.36 |       672.10 |       747.89 |       809.38 |  

[gemini-code-assist]

Summary of Changes

Hello, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request integrates new benchmark results for the NVIDIA H100 80GB HBM3 GPU. It offers detailed performance metrics for dense matrix multiplication, exploring a diverse array of quantization strategies and data types. The addition of these benchmarks provides valuable insights into the efficiency of different numerical precision configurations and underlying libraries on the H100 architecture, which is essential for optimizing deep learning workloads.

Highlights

• New H100 Benchmark Results: Added a comprehensive set of benchmark results for the NVIDIA H100 80GB HBM3 GPU, covering various quantization schemes and data types.
• Quantization Scheme Coverage: Included performance data for weight and activation data types ranging from 1-bit integers to 8-bit floats (float8e4m3, float6e3m2, float7e4m2) and float16.
• Library Comparison: Provided benchmarks for dense matrix multiplication using different backend libraries including CUTLASS, Hummingbird, Marlin, and native PyTorch (torch).
• Problem Size: All benchmarks were conducted for a dense matrix multiplication problem with N=8192 and K=8192, with results across various batch sizes.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for GitHub and other Google products, sign up here.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request introduces several new JSON files containing benchmark results for various configurations under benchmarks/results/h100/dense_n8192_k8192_g0/. These files detail problem specifications, device information, package versions, and performance metrics (time, memory_gbps, compute_tops) across different batch sizes. A consistent issue identified in the review is that all newly added JSON files are missing a final newline character, which is a good practice for consistency and adherence to POSIX standards.

[jinzhen-lin]

Thanks! I've optimized the performance on H100/H200, and now Humming's performance is on par with or even exceeds Torch and CUTLASS.

|   shape_m |   torch-w16a16 |   marlin-w4a16 |   humming-w4a16 |   marlin-w8a16 |   humming-w8a16 |   humming-w1a16 |   humming-w2a16 |   humming-w3a16 |   humming-w5a16 |   humming-w6a16 |   humming-w7a16 |
|----------:|---------------:|---------------:|----------------:|---------------:|----------------:|----------------:|----------------:|----------------:|----------------:|----------------:|----------------:|
|         1 |           2.68 |           5.81 |            5.99 |           4.29 |            4.3  |            6.93 |            6.91 |            6.05 |            5.24 |            4.99 |            4.51 |
|         2 |           5.32 |          11.63 |           11.85 |           8.55 |            8.52 |           13.84 |           13.63 |           12.1  |           10.57 |            9.99 |            9.01 |
|         4 |          10.63 |          23.14 |           23.9  |          17.22 |           17.17 |           27.27 |           27.23 |           24.17 |           21.12 |           19.97 |           18    |
|         8 |          21.23 |          46.36 |           47.74 |          34.1  |           34.02 |           55.27 |           54.49 |           47.69 |           42.22 |           39.53 |           35.95 |
|        16 |          42.09 |          90.52 |           93.79 |          67.18 |           67.23 |          106.16 |          106.41 |           92.68 |           82.8  |           78.17 |           71.72 |
|        32 |          96.24 |         128.14 |          178.22 |         105.16 |          132.39 |          195.69 |          193.92 |          175.88 |          158.83 |          153.3  |          139.45 |
|        64 |         189.82 |         198.62 |          305.53 |         175.61 |          253.4  |          323.4  |          325.75 |          295.02 |          271.08 |          266.67 |          250.79 |
|       128 |         375.76 |         308.06 |          477.55 |         283.18 |          449.66 |          485.24 |          486.19 |          475.93 |          458.02 |          454.34 |          438.98 |
|       256 |         606.81 |         362.9  |          584.2  |         314.58 |          570.95 |          596.8  |          592.55 |          580.21 |          553.16 |          549.19 |          520.75 |
|       512 |         692.02 |         397.81 |          636.09 |         353.09 |          584.83 |          687.99 |          673.08 |          659.23 |          611.83 |          593.35 |          567.42 |
|      1024 |         684.03 |         413.8  |          683.35 |         355.15 |          649.62 |          745.99 |          720.81 |          691.01 |          645.51 |          653.26 |          607.24 |
|      2048 |         669.18 |         410.24 |          685.21 |         349.41 |          667.51 |          750.17 |          722.8  |          684.92 |          644.65 |          656.42 |          618.83 |
|      4096 |         667.43 |         408.07 |          685.83 |         347.04 |          694.06 |          738.2  |          721.21 |          688.72 |          647.44 |          667.61 |          638.69 |
|      8192 |         647.4  |         406.92 |          699.09 |         345.64 |          718.86 |          759.72 |          734.09 |          697.93 |          690    |          708.55 |          665.89 |
|     16384 |         639.81 |         406.23 |          697.69 |         343.66 |          722.8  |          746.86 |          735.31 |          700.5  |          679.73 |          718.77 |          675.31 |

|   shape_m |   cutlass-w8a8 |   humming-w8a8 |   humming-w7a8 |   humming-w6a8 |   humming-w5a8 |   humming-w4a8 |   humming-w3a8 |   humming-w2a8 |   humming-w1a8 |
|----------:|---------------:|---------------:|---------------:|---------------:|---------------:|---------------:|---------------:|---------------:|---------------:|
|         1 |           4.78 |           4.4  |           4.7  |           5.14 |           5.81 |           6.37 |           6.69 |           7.33 |           7.71 |
|         2 |           9.46 |           8.84 |           9.5  |          10.29 |          11.48 |          12.73 |          13.38 |          14.41 |          15.19 |
|         4 |          19.03 |          17.67 |          18.81 |          20.78 |          22.96 |          25.45 |          26.71 |          29.27 |          30.43 |
|         8 |          38.06 |          35.16 |          37.61 |          41.54 |          46.4  |          50.87 |          52.69 |          58.4  |          61.54 |
|        16 |          75.87 |          70.01 |          75.29 |          82.2  |          92.02 |         100.65 |         103.4  |         112.52 |         118.81 |
|        32 |         147.02 |         138.3  |         147.44 |         162.19 |         179.39 |         200.5  |         200.51 |         211.47 |         228.7  |
|        64 |         298.54 |         268.11 |         287.66 |         312.48 |         346.16 |         375.95 |         366.01 |         385.93 |         402.04 |
|       128 |         555.09 |         503.98 |         534.15 |         571.79 |         612.42 |         645.07 |         616.99 |         648.81 |         653.41 |
|       256 |         970.36 |         869.06 |         807.44 |         862.04 |         867.52 |         933.66 |         838.03 |         896.43 |         895.59 |
|       512 |        1288.56 |        1001.37 |         918.31 |         991.71 |        1034.64 |        1121.48 |        1021.42 |        1069.58 |        1159.11 |
|      1024 |        1367.24 |        1252.57 |        1005.56 |        1078.16 |        1094.95 |        1244.55 |        1068.45 |        1187.27 |        1257.46 |
|      2048 |        1375.15 |        1245.56 |        1011.68 |        1094.72 |        1116.31 |        1292.54 |        1107.61 |        1240.94 |        1329.73 |
|      4096 |        1289.73 |        1269.4  |        1041    |        1127.1  |        1146.87 |        1305.34 |        1132.14 |        1266.85 |        1364.5  |
|      8192 |        1272.02 |        1307.11 |        1078.07 |        1162.22 |        1179.75 |        1342.96 |        1156.27 |        1328.78 |        1412.53 |
|     16384 |        1274.54 |        1302.51 |        1083.65 |        1168.22 |        1186.88 |        1347.01 |        1170.99 |        1318.73 |        1422.67 |