Spaces:
Sleeping
Sleeping
File size: 1,934 Bytes
1ffc326 f712ac1 1ffc326 f712ac1 1ffc326 |
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 |
import json
import os
import logging
from datetime import datetime
from lm_eval import tasks, evaluator, utils
from src.envs import API
from src.backend.manage_requests import EvalRequest
logging.getLogger("openai").setLevel(logging.WARNING)
def run_evaluation(eval_request: EvalRequest, task_names, num_fewshot, batch_size, device, local_dir: str, results_repo: str, no_cache=True, limit=None):
if limit:
print(
"!!!!!! WARNING: NOT A FULL EVALUATION !!!!! Just looking at %d."
"The `LIMIT` environment variable and/or `limit` in `run_evaluation` "
"SHOULD ONLY BE USED FOR TESTING. "
"REAL METRICS SHOULD NOT BE COMPUTED USING LIMIT." % limit
)
task_names = utils.pattern_match(task_names, tasks.ALL_TASKS)
print(f"Selected Tasks: {task_names}")
results = evaluator.simple_evaluate(
model="hf-causal-experimental", # "hf-causal"
model_args=eval_request.get_model_args(),
tasks=task_names,
num_fewshot=num_fewshot,
batch_size=batch_size,
device=device,
no_cache=no_cache,
limit=limit,
write_out=True,
output_base_path="logs"
)
results["config"]["model_dtype"] = eval_request.precision
results["config"]["model_name"] = eval_request.model
results["config"]["model_sha"] = eval_request.revision
dumped = json.dumps(results, indent=2)
print(dumped)
output_path = os.path.join(local_dir, *eval_request.model.split("/"), f"results_{datetime.now()}.json")
os.makedirs(os.path.dirname(output_path), exist_ok=True)
with open(output_path, "w") as f:
f.write(dumped)
print(evaluator.make_table(results))
API.upload_file(
path_or_fileobj=output_path,
path_in_repo=f"{eval_request.model}/results_{datetime.now()}.json",
repo_id=results_repo,
repo_type="dataset",
)
return results
|