Merge branch 'dev' into enable_flash_attention

.readthedocs.yaml

@@ -8,15 +8,18 @@ version: 2
 build:
   os: ubuntu-22.04
   tools:
-    python: "mambaforge-4.10"
+    python: "3.10"
+
+python:
+  install:
+    - method: pip
+      path: .
+      extra_requirements:
+        - docs
 
 # Build documentation in the docs/ directory with Sphinx
 sphinx:
   configuration: docs/conf.py
-
 # Optionally build your docs in additional formats such as PDF
 # formats:
 #  - pdf
-
-conda:
-  environment: env_cpu.yml

EventStream/data/pytorch_dataset.py

@@ -153,10 +153,10 @@ def __init__(self, config: PytorchDatasetConfig, split: str):
 
             self.has_task = True
 
-            if len(list(task_dir.glob("{split}*.parquet"))) > 0:
+            if len(list(task_dir.glob(f"{split}*.parquet"))) > 0:
                 print(
                     f"Re-loading task data for {self.config.task_df_name} from {task_dir}:\n"
-                    f"{', '.join([str(fp) for fp in task_dir.glob('{split}*.parquet')])}"
+                    f"{', '.join([str(fp) for fp in task_dir.glob(f'{split}*.parquet')])}"
                 )
                 self.cached_data = pl.scan_parquet(task_dir / f"{split}*.parquet")
                 with open(task_info_fp) as f:

EventStream/evaluation/general_generative_evaluation.py

@@ -97,7 +97,7 @@ class GenerateConfig:
 
     do_overwrite: bool = False
 
-    optimization_config: OptimizationConfig = OptimizationConfig()
+    optimization_config: OptimizationConfig = dataclasses.field(default_factory=lambda: OptimizationConfig())
 
     task_df_name: str | None = None
 

EventStream/transformer/config.py

@@ -924,7 +924,7 @@ def to_dict(self) -> dict[str, Any]:
     @classmethod
     def from_dict(cls, *args, **kwargs) -> "StructuredTransformerConfig":
         raw_from_dict = super().from_dict(*args, **kwargs)
-        if raw_from_dict.measurmeent_configs:
+        if raw_from_dict.measurement_configs:
             new_meas_configs = {}
             for k, v in raw_from_dict.measurement_configs.items():
                 new_meas_configs[k] = MeasurementConfig.from_dict(v)

EventStream/transformer/lightning_modules/fine_tuning.py

@@ -311,7 +311,7 @@ class FinetuneConfig:
             },
         }
     )
-    optimization_config: OptimizationConfig = OptimizationConfig()
+    optimization_config: OptimizationConfig = dataclasses.field(default_factory=lambda: OptimizationConfig())
     data_config: dict[str, Any] | None = dataclasses.field(
         default_factory=lambda: {
             **{k: None for k in PytorchDatasetConfig().to_dict().keys()},

EventStream/transformer/lightning_modules/generative_modeling.py

@@ -511,12 +511,16 @@ class PretrainConfig:
             },
         }
     )
-    optimization_config: OptimizationConfig = OptimizationConfig()
-    data_config: PytorchDatasetConfig = PytorchDatasetConfig()
-    pretraining_metrics_config: MetricsConfig = MetricsConfig(
-        include_metrics={Split.TRAIN: {MetricCategories.LOSS_PARTS: True}},
+    optimization_config: OptimizationConfig = dataclasses.field(default_factory=lambda: OptimizationConfig())
+    data_config: PytorchDatasetConfig = dataclasses.field(default_factory=lambda: PytorchDatasetConfig())
+    pretraining_metrics_config: MetricsConfig = dataclasses.field(
+        default_factory=lambda: MetricsConfig(
+            include_metrics={Split.TRAIN: {MetricCategories.LOSS_PARTS: True}},
+        )
+    )
+    final_validation_metrics_config: MetricsConfig = dataclasses.field(
+        default_factory=lambda: MetricsConfig(do_skip_all_metrics=False)
     )
-    final_validation_metrics_config: MetricsConfig = MetricsConfig(do_skip_all_metrics=False)
 
     trainer_config: dict[str, Any] = dataclasses.field(
         default_factory=lambda: {

README.md

@@ -29,14 +29,7 @@ GitHub issue.
 ## Installation
 
 Installation of the required dependencies can be done via pip with `pip install -e .` in the root directory of
-the repository.
-
-### Other Installation Instructions
-
-Installation (this mode is deprecated) can also be done via conda with the `env.yml` file: `conda env create -n ${ENV_NAME} -f env.yml`
-
-It can also be attempted via poetry, though that approach has issues at present in differentiating between cpu
-and gpu machines.
+the repository. To be able to run tests, use `pip install -e .[tests]`. To be able to build docs, use `pip install -e .[docs]`.
 
 ## Overview