bump olmo core version and add g cloud compute to reqs (#393)

* bump olmo core version and add g cloud compute to reqs

* set some logs to debug

* add torchrun default

---------

Co-authored-by: hgherzog <henryh@allenai.org>
Co-authored-by: Hgherzog <77407865+Hgherzog@users.noreply.github.com>

Author: pjreddie

olmoearth_pretrain/data/dataset.py

@@ -823,7 +823,7 @@ def _fill_missing_modality(
     ) -> OlmoEarthSample:
         """Fill an array of shape of modality with the missing value."""
         expected_shape = sample.get_expected_shape(modality)
-        logger.info(f"Filling {modality} with shape {expected_shape}")
+        logger.debug(f"Filling {modality} with shape {expected_shape}")
         return np.full(
             expected_shape,
             fill_value=MISSING_VALUE,

olmoearth_pretrain/internal/experiment.py

@@ -275,7 +275,8 @@ def launch(config: OlmoEarthExperimentConfig) -> None:
     logger.info("Launching the experiment")
     logger.info(config)
     # Set follow=False if you don't want to stream the logs to the terminal
-    config.launch.launch(follow=False)
+    # Default to enabling torchrun so we can run multi gpu scripts on single gpu
+    config.launch.launch(follow=False, torchrun=True)
 
 
 def prep(config: OlmoEarthExperimentConfig) -> None:

olmoearth_pretrain/train/train_module/contrastive_latentmim.py

@@ -297,7 +297,7 @@ def model_forward(
             if extra_metrics is not None:
                 self.log_extra_metrics(extra_metrics)
             with torch.no_grad():
-                logger.info("Target Encoder forward pass...")
+                logger.debug("Target Encoder forward pass...")
                 output_dict = self.model.target_encoder.forward(
                     batch.unmask(),
                     patch_size=patch_size,

olmoearth_pretrain/train/train_module/train_module.py

@@ -339,7 +339,9 @@ def state_dict(self) -> dict[str, Any]:
         """Get the state dict."""
         return self._get_state_dict(self.state_dict_save_opts)
 
-    def state_dict_to_load(self, metadata: Metadata) -> dict[str, Any]:
+    def state_dict_to_load(
+        self, metadata: Metadata, optim: bool | None = None
+    ) -> dict[str, Any]:
         """Get the state dict to load."""
         load_opts = self.state_dict_load_opts
         return self._get_state_dict(load_opts)

requirements-beaker.txt

@@ -1,5 +1,6 @@
 beaker-py==1.34.1
 docker>=5.0,<8.0
+google-cloud-compute
 packaging
 pydantic>=1.8.2,<3.0
 PyYAML

requirements.txt

@@ -1,4 +1,4 @@
-ai2-olmo-core @ git+https://github.com/allenai/OLMo-core.git@abc12e50ba756c21e575452cfc6f150dafa9509e # Pin here until >2.1.0 is released.
+ai2-olmo-core==2.3.0
 albumentations
 cartopy
 class-registry