Use Lazy Loaders

performance/bluebench_profiler.py

@@ -93,14 +93,7 @@ def profiler_do_the_profiling(self, dataset_query: str, split: str, **kwargs):
             benchmark_recipe=benchmark_recipe, split=split, **kwargs
         )
 
-        model = self.profiler_instantiate_model()
-
-        predictions = self.profiler_infer_predictions(model=model, dataset=dataset)
-
-        evaluation_result = self.profiler_evaluate_predictions(
-            predictions=predictions, dataset=dataset
-        )
-        logger.critical(f"length of evaluation_result: {len(evaluation_result)}")
+        logger.critical(f"length of bluebench generated dataset: {len(dataset)}")
 
 
 dataset_query = "benchmarks.bluebench[loader_limit=30,max_samples_per_subset=30]"
@@ -154,44 +147,29 @@ def main():
         pst.strip_dirs()
         pst.sort_stats("name")  # sort by function name
         pst.print_stats(
-            "profile_benchmark_blue_bench|profiler_instantiate_benchmark_recipe|profiler_generate_benchmark_dataset|profiler_instantiate_model|profiler_infer_predictions|profiler_evaluate_predictions|load_data|load_iterables"
+            "profile_benchmark_blue_bench|profiler_instantiate_benchmark_recipe|profiler_generate_benchmark_dataset|load_data|load_iterables"
         )
         s = f.getvalue()
         assert s.split("\n")[7].split()[3] == "cumtime"
         overall_tot_time = find_cummtime_of(
             "profile_benchmark_blue_bench", "bluebench_profiler.py", s
         )
         load_time = find_cummtime_of("load_data", "loaders.py", s)
-        just_load_no_initial_ms_time = find_cummtime_of(
-            "load_iterables", "loaders.py", s
-        )
+
         instantiate_benchmark_time = find_cummtime_of(
             "profiler_instantiate_benchmark_recipe", "bluebench_profiler.py", s
         )
         generate_benchmark_dataset_time = find_cummtime_of(
             "profiler_generate_benchmark_dataset", "bluebench_profiler.py", s
         )
-        instantiate_model_time = find_cummtime_of(
-            "profiler_instantiate_model", "bluebench_profiler.py", s
-        )
-        inference_time = find_cummtime_of(
-            "profiler_infer_predictions", "bluebench_profiler.py", s
-        )
-        evaluation_time = find_cummtime_of(
-            "profiler_evaluate_predictions", "bluebench_profiler.py", s
-        )
 
         # Data to be written
         dictionary = {
             "dataset_query": dataset_query,
             "total_time": overall_tot_time,
             "load_time": load_time,
-            "load_time_no_initial_ms": just_load_no_initial_ms_time,
             "instantiate_benchmark_time": instantiate_benchmark_time,
             "generate_benchmark_dataset_time": generate_benchmark_dataset_time,
-            "instantiate_model_time": instantiate_model_time,
-            "inference_time": inference_time,
-            "evaluation_time": evaluation_time,
             "used_eager_mode": settings.use_eager_execution,
             "performance.prof file": temp_prof_file_path,
         }

performance/compare_benchmark_performance_results.py

@@ -1,6 +1,5 @@
 import argparse
 import json
-import os
 import sys
 
 # Argument parser to get file paths from the command line
@@ -23,24 +22,11 @@
 print(f'dataset_query = "{main_perf["dataset_query"]}"')
 print(f"used_eager_mode in main = {main_perf['used_eager_mode']}")
 print(f"used_eager_mode in PR = {pr_perf['used_eager_mode']}")
-print(f"use Mocked inference = {os.environ['UNITXT_MOCK_INFERENCE_MODE']}")
 
 ratio1 = (
-    (pr_perf["generate_benchmark_dataset_time"] - pr_perf["load_time_no_initial_ms"])
-    / (
-        main_perf["generate_benchmark_dataset_time"]
-        - main_perf["load_time_no_initial_ms"]
-    )
-    if (
-        main_perf["generate_benchmark_dataset_time"]
-        - main_perf["load_time_no_initial_ms"]
-    )
-    > 0
-    else 1
-)
-ratio2 = (
-    pr_perf["evaluation_time"] / main_perf["evaluation_time"]
-    if main_perf["evaluation_time"] > 0
+    (pr_perf["generate_benchmark_dataset_time"] - pr_perf["load_time"])
+    / (main_perf["generate_benchmark_dataset_time"] - main_perf["load_time"])
+    if (main_perf["generate_benchmark_dataset_time"] - main_perf["load_time"]) > 0
     else 1
 )
 # Markdown table formatting
@@ -49,26 +35,19 @@
 line2 = "--------------------|-------------|-------------|---------------\n"
 line3 = f" Total time         | {main_perf['total_time']:>11} | {pr_perf['total_time']:>11} | {pr_perf['total_time'] / main_perf['total_time']:.2f}\n"
 ratio_line4 = (
-    pr_perf["load_time_no_initial_ms"] / main_perf["load_time_no_initial_ms"]
-    if main_perf["load_time_no_initial_ms"] > 0
-    else 1
+    pr_perf["load_time"] / main_perf["load_time"] if main_perf["load_time"] > 0 else 1
 )
-line4 = f" Load time          | {main_perf['load_time_no_initial_ms']:>11} | {pr_perf['load_time_no_initial_ms']:>11} | {ratio_line4:.2f}\n"
+line4 = f" Load time          | {main_perf['load_time']:>11} | {pr_perf['load_time']:>11} | {ratio_line4:.2f}\n"
 line5 = f" DS Gen. inc. Load  | {main_perf['generate_benchmark_dataset_time']:>11} | {pr_perf['generate_benchmark_dataset_time']:>11} | {pr_perf['generate_benchmark_dataset_time'] / main_perf['generate_benchmark_dataset_time']:.2f}\n"
-line6 = f" DS Gen. exc. Load  | {round(main_perf['generate_benchmark_dataset_time'] - main_perf['load_time_no_initial_ms'], 3):>11} | {round(pr_perf['generate_benchmark_dataset_time'] - pr_perf['load_time_no_initial_ms'], 3):>11} | {ratio1:.2f}\n"
-line7 = f" Inference time     | {main_perf['inference_time']:>11} | {pr_perf['inference_time']:>11} | {pr_perf['inference_time'] / main_perf['inference_time']:.2f}\n"
-line8 = f" Evaluate  time     | {main_perf['evaluation_time']:>11} | {pr_perf['evaluation_time']:>11} | {ratio2:.2f}\n"
-line9 = f" Benchmark Instant. | {main_perf['instantiate_benchmark_time']:>11} | {pr_perf['instantiate_benchmark_time']:>11} | {pr_perf['instantiate_benchmark_time'] / main_perf['instantiate_benchmark_time']:.2f}\n"
-line10 = f" Model Instantiation| {main_perf['instantiate_model_time']:>11} | {pr_perf['instantiate_model_time']:>11} | {pr_perf['instantiate_model_time'] / main_perf['instantiate_model_time']:.2f}\n"
+line6 = f" DS Gen. exc. Load  | {round(main_perf['generate_benchmark_dataset_time'] - main_perf['load_time'], 3):>11} | {round(pr_perf['generate_benchmark_dataset_time'] - pr_perf['load_time'], 3):>11} | {ratio1:.2f}\n"
+line7 = f" Benchmark Instant. | {main_perf['instantiate_benchmark_time']:>11} | {pr_perf['instantiate_benchmark_time']:>11} | {pr_perf['instantiate_benchmark_time'] / main_perf['instantiate_benchmark_time']:.2f}\n"
 
 print("### Performance Comparison Results, time expressed in seconds:\n")
-print(line1 + line2 + line3 + line4 + line5 + line6 + line7 + line8 + line9 + line10)
+print(line1 + line2 + line3 + line4 + line5 + line6 + line7)
 print("\n\n")
 # Performance degradation check (5% threshold)
-if ratio1 > 1.05 or ratio2 > 1.05:
-    print(
-        "\n**Warning**: Performance degradation in Dataset Generation and/or Evaluation exceeds 5%!"
-    )
+if ratio1 > 1.05:
+    print("\n**Warning**: Performance degradation in Dataset Generation exceeds 5%!")
     print(
         "Explore branch performance via 'python performance/bluebench_profiler.py --output_file=<path to json file>',"
         "followed by 'snakeviz <the performance.prof file specified in the output json file>'."

prepare/cards/universal_ner.py

@@ -48,7 +48,7 @@
         loader=LoadHF(
             path="universalner/universal_ner",
             name=sub_task,
-            requirements_list=["conllu"],
+            requirements=["conllu"],
         ),
         preprocess_steps=[
             # The dataset is sorted by classes

pyproject.toml

@@ -190,7 +190,7 @@ keep-runtime-typing = true
 "src/unitxt/metric.py" = ["F811", "F401"]
 "src/unitxt/dataset.py" = ["F811", "F401"]
 "src/unitxt/blocks.py" = ["F811", "F401"]
-"tests/library/test_loaders.py" = ["N802", "N803"]
+"tests/library/test_loaders.py" = ["N802", "N803", "RUF015"]
 "tests/library/test_dataclass.py" = ["F811", "E731"]
 "src/unitxt/validate.py" = ["B024"]
 "src/unitxt/standard.py" = ["C901"]

src/unitxt/catalog/cards/universal_ner/ceb/gja.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "ceb_gja",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/da/ddt.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "da_ddt",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/de/pud.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "de_pud",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/en/ewt.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "en_ewt",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/en/pud.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "en_pud",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/hr/set.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "hr_set",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/pt/bosque.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "pt_bosque",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/pt/pud.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "pt_pud",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/ru/pud.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "ru_pud",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/sk/snk.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "sk_snk",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/sr/set.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "sr_set",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/sv/pud.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "sv_pud",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/sv/talbanken.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "sv_talbanken",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/tl/trg.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "tl_trg",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/tl/ugnayan.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "tl_ugnayan",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/zh/gsd.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "zh_gsd",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/zh/gsdsimp.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "zh_gsdsimp",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/catalog/cards/universal_ner/zh/pud.json

@@ -4,7 +4,7 @@
         "__type__": "load_hf",
         "path": "universalner/universal_ner",
         "name": "zh_pud",
-        "requirements_list": [
+        "requirements": [
             "conllu"
         ]
     },

src/unitxt/fusion.py

@@ -34,11 +34,7 @@ def prepare_subsets(self):
             for i in range(len(self.subsets)):
                 self.named_subsets[i] = self.subsets[i]
         else:
-            for name, origin in self.subsets.items():
-                try:
-                    self.named_subsets[name] = origin
-                except Exception as e:
-                    raise RuntimeError(f"Exception in subset: {name}") from e
+            self.named_subsets = self.subsets
 
     def splits(self) -> List[str]:
         self.prepare_subsets()