Hotfix: Device errors

openvalidators/forward.py

@@ -42,15 +42,24 @@ def get_random_uids(self, k: int, exclude: List[int] = None) -> torch.LongTensor
         If `k` is larger than the number of available `uids`, set `k` to the number of available `uids`.
     """
     candidate_uids = []
+    avail_uids = []
 
     for uid in range(self.metagraph.n.item()):
         uid_is_available = check_uid_availability(self.metagraph, uid, self.config.neuron.vpermit_tao_limit)
         uid_is_not_excluded = exclude is None or uid not in exclude
 
-        if uid_is_available and uid_is_not_excluded:
-            candidate_uids.append(uid)
+        if uid_is_available:
+            avail_uids.append(uid)
+            if uid_is_not_excluded:
+                candidate_uids.append(uid)
+
+    # Check if candidate_uids contain enough for querying, if not grab all avaliable uids
+    if len(candidate_uids) > k:
+        available_uids = torch.tensor(candidate_uids, dtype=torch.int64).to(self.device)
+    else:
+        available_uids = torch.tensor(avail_uids, dtype=torch.int64).to(self.device)
+
 
-    available_uids = torch.tensor(candidate_uids, dtype=torch.int64).to(self.device)
     uids = torch.tensor(random.sample(available_uids.tolist(), k), dtype=torch.int64)
     return uids
 

openvalidators/gating.py

@@ -137,7 +137,7 @@ def __init__(
             config = GatingModel.config()
         if model_name is not None:
             config.gating.model_name = model_name
-        config.gating.num_uids = num_uids if num_uids is not None else metagraph.n
+        config.gating.num_uids = num_uids if num_uids is not None else config.gating.num_uids
         self.config = config
         self.num_uids = config.gating.num_uids
         self.device = torch.device(self.config.neuron.device)
@@ -228,7 +228,7 @@ def __init__(
             config = SentenceEmbedGatingModel.config()
         if model_name is not None:
             config.gating.model_name = model_name
-        config.gating.num_uids = num_uids if num_uids is not None else metagraph.n
+        config.gating.num_uids = num_uids if num_uids is not None else config.gating.num_uids
         self.config = config
         self.num_uids = config.gating.num_uids
         self.device = torch.device(self.config.neuron.device)

openvalidators/reward/reward.py

@@ -51,7 +51,7 @@ def normalize_rewards( self, rewards: torch.FloatTensor ) -> torch.FloatTensor:
             - This function uses Welford's online algorithm to update the mean and variance.
             - It standardizes the reward values using the updated mean and variance.
             - It then scales the standardized values to the 0-1 range using the error function (erf) as a CDF.
-        """
+        """        
         # Get the number of rewards (successful responses).
         new_count = rewards.numel()
 
@@ -88,6 +88,7 @@ def apply( self, prompt: str, responses: List[ bt.DendriteCall ], name: str) ->
         """ Applies the reward model across each call. Unsuccessful responses are zeroed.
         """
         # Get indices of correctly responding calls.
+
         successful_completions_indices: List[int] = [ idx for idx, resp in enumerate(responses) if resp.is_success ]
 
         # Get all completions from responding calls.