Upload/Download Artifacts related to Benchmark Tests

[vcanicTT]

All necessary files that contain model configurations and results we preserve as artefacts. We will mechanism for uploading and downloading those files, and parsing them with aim to extract the performance results.

[vcanicTT]

@vmilosevic The files we want to upload are .json files. The files should start with "forge-benchmark-e2e", for example it can be "forge-benchmark-e2e-mnist.json". Also, JSON files are preserved in the root directory.

[vcanicTT]

@vmilosevic I am not sure if this should be part of this issue, these are fields we want to add.

• model, string, name of the model
• config, string, the configuration of the model, very often small, large, medium, etc.
• date, string or datetime, the date and time the model was run.
• hash, string, hash code of git commit on which the model was run.
• machine_name, string, name of the machine on which the model was run
• samples_per_sec, float, speed of the model given in samples per second
• total_samples, integer, total number of the samples for the model
• total_time, float, total execution time
• training, bool, it is training or inference
• batch_size, integer, it represents batch size, number of samples in one batch, it can be total number of samples, but generally that's not true
• loop_count, integer, number of times we run the test, and average perf result
• output, string, name of the output file where we preserve the results and configurations, it is an artifact
• arch, string, the architecture on which we run the model
• chips, integer, number of the chips used for running the model
• dataformat, string, initial data format used in the model
• math_fidelity, string, initial math fidelity used in the model
• device, string, type of the device on which we run the model
• galaxy, bool, indicator that says we are running on the galaxy device
• perf_analysis, bool, model was run with perf analysis mode
• load_tti, string, path the model, from this location we load the model
• save_tti, string, path the model, to this location we preserve the model
• task, string, type of the problem we want to solve, e.g, image classification, image generation, etc.
• evaluation_score, float, score we get on the model for given data and task