Merge branch 'main' into dependabot/github_actions/actions/upload-art…

…ifact-4
allenai · May 30, 2024 · 7ee6e06 · 7ee6e06
2 parents daa9c3d + 6aaa8ff
commit 7ee6e06
Show file tree

Hide file tree

Showing 7 changed files with 20 additions and 10 deletions.
diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml
@@ -20,7 +20,6 @@ env:
   WANDB_API_KEY: ${{ secrets.WANDB_API_KEY }}
   BEAKER_TOKEN: ${{ secrets.BEAKER_TOKEN }}
   BEAKER_WORKSPACE: ai2/tango-testing
-  BEAKER_DEFAULT_CLUSTER: ai2/allennlp-cirrascale
   BEAKER_IMAGE: petew/tango-testing
   GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 
@@ -288,7 +287,7 @@ jobs:
                 image:
                   beaker: ${{ env.BEAKER_IMAGE }}
                 context:
-                  cluster: ${{ env.BEAKER_DEFAULT_CLUSTER }}
+                  preemptible: true 
                 resources:
                   gpuCount: 2
                 envVars:

diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -10,6 +10,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Fixed
 
 - Fixed a bunch of dependencies
+- Upgraded to new version of wandb
 
 ## [v1.3.2](https://github.com/allenai/tango/releases/tag/v1.3.2) - 2023-10-27
 

diff --git a/docs/source/first_steps.md b/docs/source/first_steps.md
@@ -247,7 +247,7 @@ Computing...: 100%|##########| 100/100 [00:05<00:00, 18.99it/s]
 ✓ The output for "add_numbers" is in workspace/runs/live-tarpon/add_numbers
 ```
 
-The last line in the output tells us where we can find the result of our "add_numbers" step. `live-parpon` is
+The last line in the output tells us where we can find the result of our "add_numbers" step. `live-tarpon` is
 the name of the run. Run names are randomly generated and may be different on your machine. `add_numbers` is the
 name of the step in your config. The whole path is a symlink to a directory, which contains (among other things)
 a file `data.json`:

diff --git a/pyproject.toml b/pyproject.toml
@@ -96,7 +96,7 @@ flax = [
   "tensorflow-cpu>=2.9.1"
 ]
 wandb = [
-  "wandb>=0.12,<0.14.3",
+  "wandb>=0.16",
   "retry"
 ]
 beaker = [

diff --git a/tango/integrations/transformers/__init__.py b/tango/integrations/transformers/__init__.py
@@ -93,6 +93,7 @@
             transformers::constant
             transformers::constant_with_warmup
             transformers::cosine
+            transformers::cosine_with_min_lr
             transformers::cosine_with_restarts
             transformers::inverse_sqrt
             transformers::linear

diff --git a/tango/integrations/wandb/step_cache.py b/tango/integrations/wandb/step_cache.py
@@ -72,7 +72,7 @@ def _step_artifact_name(self, step: Union[Step, StepInfo]) -> str:
 
     def _step_result_remote(  # type: ignore
         self, step: Union[Step, StepInfo]
-    ) -> Optional[wandb.apis.public.Artifact]:
+    ) -> Optional[wandb.Artifact]:
         artifact_kind = (step.metadata or {}).get("artifact_kind", ArtifactKind.STEP_RESULT.value)
         try:
             return self.wandb_client.artifact(
@@ -88,9 +88,7 @@ def _step_result_remote(  # type: ignore
     def create_step_result_artifact(self, step: Step, objects_dir: Optional[PathOrStr] = None):
         self._upload_step_remote(step, objects_dir)
 
-    def get_step_result_artifact(
-        self, step: Union[Step, StepInfo]
-    ) -> Optional[wandb.apis.public.Artifact]:
+    def get_step_result_artifact(self, step: Union[Step, StepInfo]) -> Optional[wandb.Artifact]:
         artifact_kind = (step.metadata or {}).get("artifact_kind", ArtifactKind.STEP_RESULT.value)
         try:
             return self.wandb_client.artifact(
@@ -144,7 +142,7 @@ def use_step_result_artifact(self, step: Union[Step, StepInfo]) -> None:
 
     def _download_step_remote(self, step_result, target_dir: PathOrStr):
         try:
-            step_result.download(root=target_dir, recursive=True)
+            step_result.download(root=target_dir)
         except (WandbError, ValueError):
             raise RemoteNotFoundError()
 

diff --git a/tango/integrations/wandb/util.py b/tango/integrations/wandb/util.py
@@ -1,4 +1,5 @@
 import os
+import re
 import warnings
 from enum import Enum
 
@@ -13,7 +14,17 @@ def is_missing_artifact_error(err: WandbError):
     Check if a specific W&B error is caused by a 404 on the artifact we're looking for.
     """
     # This is brittle, but at least we have a test for it.
-    return "does not contain artifact" in err.message
+
+    # This is a workaround for a bug in the wandb API
+    if err.message == "'NoneType' object has no attribute 'get'":
+        return True
+
+    if re.search(r"^artifact '.*' not found in '.*'$", err.message):
+        return True
+
+    return ("does not contain artifact" in err.message) or (
+        "Unable to fetch artifact with name" in err.message
+    )
 
 
 def check_environment():