Merge branch 'main' of https://huggingface.co/spaces/open-rl-leaderboard/backend

README.md

@@ -8,4 +8,6 @@ sdk_version: 4.20.0
 app_file: app.py
 pinned: true
 license: apache-2.0
+tags:
+- leaderboard
 ---

src/backend.py

@@ -39,41 +39,43 @@ def _backend_routine():
         evaluated_models.add((report["config"]["model_id"], report["config"]["model_sha"]))
 
     # Find the models that are not associated with any results
-    pending_models = set(compatible_models) - evaluated_models
+    pending_models = list(set(compatible_models) - evaluated_models)
     logger.info(f"Found {len(pending_models)} pending models")
 
+    if len(pending_models) == 0:
+        return None
+
     # Run an evaluation on the models
     with tempfile.TemporaryDirectory() as tmp_dir:
         commits = []
-        for model_id, sha in pending_models:
-            logger.info(f"Running evaluation on {model_id}")
-            report = {"config": {"model_id": model_id, "model_sha": sha}}
-            try:
-                evaluations = evaluate(model_id, revision=sha)
-            except Exception as e:
-                logger.error(f"Error evaluating {model_id}: {e}")
-                evaluations = None
-
-            if evaluations is not None:
-                report["results"] = evaluations
-                report["status"] = "DONE"
-            else:
-                report["status"] = "FAILED"
-
-            # Update the results
-            dumped = json.dumps(report, indent=2)
-            path_in_repo = f"{model_id}/results_{sha}.json"
-            local_path = os.path.join(tmp_dir, path_in_repo)
-            os.makedirs(os.path.dirname(local_path), exist_ok=True)
-            with open(local_path, "w") as f:
-                f.write(dumped)
-
-            commits.append(CommitOperationAdd(path_in_repo=path_in_repo, path_or_fileobj=local_path))
-
-        if len(commits) > 0:
-            API.create_commit(
-                repo_id=RESULTS_REPO, commit_message="Add evaluation results", operations=commits, repo_type="dataset"
-            )
+        model_id, sha = pending_models[0]
+        logger.info(f"Running evaluation on {model_id}")
+        report = {"config": {"model_id": model_id, "model_sha": sha}}
+        try:
+            evaluations = evaluate(model_id, revision=sha)
+        except Exception as e:
+            logger.error(f"Error evaluating {model_id}: {e}")
+            evaluations = None
+
+        if evaluations is not None:
+            report["results"] = evaluations
+            report["status"] = "DONE"
+        else:
+            report["status"] = "FAILED"
+
+        # Update the results
+        dumped = json.dumps(report, indent=2)
+        path_in_repo = f"{model_id}/results_{sha}.json"
+        local_path = os.path.join(tmp_dir, path_in_repo)
+        os.makedirs(os.path.dirname(local_path), exist_ok=True)
+        with open(local_path, "w") as f:
+            f.write(dumped)
+
+        commits.append(CommitOperationAdd(path_in_repo=path_in_repo, path_or_fileobj=local_path))
+
+        API.create_commit(
+            repo_id=RESULTS_REPO, commit_message="Add evaluation results", operations=commits, repo_type="dataset"
+        )
 
 
 def backend_routine():
@@ -81,3 +83,6 @@ def backend_routine():
         _backend_routine()
     except Exception as e:
         logger.error(f"{e.__class__.__name__}: {str(e)}")
+
+if __name__=="__main__":
+    backend_routine()

src/evaluation.py

@@ -347,7 +347,7 @@ def evaluate(model_id, revision):
 
     # Evaluate the agent on the environments
     for env_id in env_ids:
-        envs = gym.vector.SyncVectorEnv([make(env_id) for _ in range(3)])
+        envs = gym.vector.SyncVectorEnv([make(env_id) for _ in range(1)])
         observations, _ = envs.reset()
         episodic_returns = []
         while len(episodic_returns) < NUM_EPISODES:

texts/getting_my_agent_evaluated.md

@@ -93,7 +93,7 @@ class Agent(nn.Module):
 agent = Agent(policy) # instantiate the agent
 
 # A few tests to check if the agent is working
-observations = torch.tensor(env.observation_space.sample()).unsqueeze(0)  # dummy batch of observations
+observations = torch.randn(env.observation_space.shape).unsqueeze(0)  # dummy batch of observations
 actions = agent(observations)
 actions = actions.numpy()[0]
 assert env.action_space.contains(actions)
@@ -109,10 +109,9 @@ from huggingface_hub import metadata_save, HfApi
 
 # Save model along with its card
 metadata_save("model_card.md", {"tags": ["reinforcement-learning", env_id]})
-dummy_input = torch.tensor(env.observation_space.sample()).unsqueeze(0)  # dummy batch of observations
+dummy_input = torch.randn(env.observation_space.shape).unsqueeze(0)  # dummy batch of observations
 agent = torch.jit.trace(agent.eval(), dummy_input)
-agent = torch.jit.freeze(agent)  # required for for the model not to depend on the training library
-agent = torch.jit.optimize_for_inference(agent)
+agent = torch.jit.freeze(agent)  # required for the model not to depend on the training library
 torch.jit.save(agent, "agent.pt")
 
 # Upload model and card to the 🤗 Hub