File size: 3,819 Bytes
f0b90cf
58b9de9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f0b90cf
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b46b972
f0b90cf
 
 
 
b46b972
f0b90cf
 
 
 
 
 
 
 
 
 
 
 
b46b972
f0b90cf
 
 
 
b46b972
f0b90cf
b46b972
f0b90cf
58b9de9
 
 
f0b90cf
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
import argparse
import logging
import pprint

from huggingface_hub import snapshot_download

import src.backend.run_eval_suite as run_eval_suite
import src.backend.manage_requests as manage_requests
import src.backend.sort_queue as sort_queue
import src.envs as envs

logging.basicConfig(level=logging.ERROR)
pp = pprint.PrettyPrinter(width=80)

PENDING_STATUS = "PENDING"
RUNNING_STATUS = "RUNNING"
FINISHED_STATUS = "FINISHED"
FAILED_STATUS = "FAILED"

snapshot_download(repo_id=envs.RESULTS_REPO, revision="main",
                local_dir=envs.EVAL_RESULTS_PATH_BACKEND, repo_type="dataset", max_workers=60)
snapshot_download(repo_id=envs.QUEUE_REPO, revision="main",
                local_dir=envs.EVAL_REQUESTS_PATH_BACKEND, repo_type="dataset", max_workers=60)


def run_auto_eval(args):
    if not args.reproduce:
        current_pending_status = [PENDING_STATUS]

        manage_requests.check_completed_evals(
            api=envs.API,
            checked_status=RUNNING_STATUS,
            completed_status=FINISHED_STATUS,
            failed_status=FAILED_STATUS,
            hf_repo=envs.QUEUE_REPO,
            local_dir=envs.EVAL_REQUESTS_PATH_BACKEND,
            hf_repo_results=envs.RESULTS_REPO,
            local_dir_results=envs.EVAL_RESULTS_PATH_BACKEND
        )
        logging.info("Checked completed evals")
        eval_requests = manage_requests.get_eval_requests(job_status=current_pending_status,
                                                        hf_repo=envs.QUEUE_REPO,
                                                        local_dir=envs.EVAL_REQUESTS_PATH_BACKEND)
        logging.info("Got eval requests")
        eval_requests = sort_queue.sort_models_by_priority(api=envs.API, models=eval_requests)
        logging.info("Sorted eval requests")

        print(f"Found {len(eval_requests)} {','.join(current_pending_status)} eval requests")

        if len(eval_requests) == 0:
            print("No eval requests found. Exiting.")
            return

        eval_request = eval_requests[0]
        pp.pprint(eval_request)

        manage_requests.set_eval_request(
            api=envs.API,
            eval_request=eval_request,
            new_status=RUNNING_STATUS,
            hf_repo=envs.QUEUE_REPO,
            local_dir=envs.EVAL_REQUESTS_PATH_BACKEND
        )
        logging.info("Set eval request to running, now running eval")

        run_eval_suite.run_evaluation(
            eval_request=eval_request,
            local_dir=envs.EVAL_RESULTS_PATH_BACKEND,
            results_repo=envs.RESULTS_REPO,
            batch_size=1,
            device=envs.DEVICE,
            no_cache=True,
        )
        logging.info("Eval finished, now setting status to finished")
    else:
        eval_request = manage_requests.EvalRequest(
            model=args.model,
            status=PENDING_STATUS,
            precision=args.precision
        )
        pp.pprint(eval_request)
        logging.info("Running reproducibility eval")

        run_eval_suite.run_evaluation(
            eval_request=eval_request,
            local_dir=envs.EVAL_RESULTS_PATH_BACKEND,
            results_repo=envs.RESULTS_REPO,
            batch_size=1,
            device=envs.DEVICE,
        )
        logging.info("Reproducibility eval finished")


def main():
    parser = argparse.ArgumentParser(description="Run auto evaluation with optional reproducibility feature")

    # Optional arguments
    parser.add_argument("--reproduce", type=bool, default=False, help="Reproduce the evaluation results")
    parser.add_argument("--model", type=str, default=None, help="Your Model ID")
    parser.add_argument("--precision", type=str, default="float16", help="Precision of your model")

    args = parser.parse_args()

    run_auto_eval(args)


if __name__ == "__main__":
    main()