diff --git a/commons/dataset.py b/commons/dataset.py
index b80f78cb..d746b151 100644
--- a/commons/dataset.py
+++ b/commons/dataset.py
@@ -1,10 +1,12 @@
 from enum import StrEnum
-from typing import Dict, Iterable, List
+from typing import Dict, Iterable, List, Tuple
 import random
 import bittensor as bt
 
 from datasets import load_dataset, interleave_datasets
 
+from template.protocol import Completion, RankingRequest
+
 seed = 42
 
 
@@ -66,42 +68,6 @@ def get_openai_webgpt_comparisons():
     )
 
 
-# NOTE this serves as a start for prompt/completion pairs to be generated because at first there will be no requests coming in
-oasst1 = load_dataset(
-    DatasetName.OPENASSISTANT_OASST1,
-    split="train",
-    streaming=True,
-)
-oasst2 = load_dataset(
-    DatasetName.OPENASSISTANT_OASST2,
-    split="train",
-    streaming=True,
-)
-
-
-def _is_oasst_prompt(row):
-    return not row["parent_id"] and row["role"] == "prompter"
-
-
-# ensure we only grab the 'text' fields from the dataset
-oasst1_prompts = oasst1.filter(lambda row: _is_oasst_prompt(row)).map(
-    lambda row: {"text": row["text"]}
-)
-oasst2_prompts = oasst2.filter(lambda row: _is_oasst_prompt(row)).map(
-    lambda row: {"text": row["text"]}
-)
-ALL_OASST_PROMPTS = "all_oasst_prompts"
-seed_datasets = {
-    ALL_OASST_PROMPTS: iter(
-        interleave_datasets(
-            [oasst1_prompts, oasst2_prompts],
-            probabilities=[0.5, 0.5],
-            seed=seed,
-            stopping_strategy="all_exhausted",
-        )
-    )
-}
-
 eval_datasets = {
     DatasetName.ANTHROPIC_HHRLHF: iter(get_anthropic_hhrlhf()),
     DatasetName.STANFORD_SHP: iter(get_stanford_shp()),
@@ -132,10 +98,69 @@ def get_batch() -> List[Dict]:
         return [next_circular(eval_datasets, key) for _ in range(batch_size)]
 
 
+# # NOTE this serves as a start for prompt/completion pairs to be generated because at first there will be no requests coming in
+# oasst1 = load_dataset(
+#     DatasetName.OPENASSISTANT_OASST1,
+#     split="train",
+#     streaming=True,
+# )
+# oasst2 = load_dataset(
+#     DatasetName.OPENASSISTANT_OASST2,
+#     split="train",
+#     streaming=True,
+# )
+
+
+# def _is_oasst_prompt(row):
+#     return not row["parent_id"] and row["role"] == "prompter"
+
+
+# # ensure we only grab the 'text' fields from the dataset
+# oasst1_prompts = oasst1.filter(lambda row: _is_oasst_prompt(row)).map(
+#     lambda row: {"text": row["text"]}
+# )
+# oasst2_prompts = oasst2.filter(lambda row: _is_oasst_prompt(row)).map(
+#     lambda row: {"text": row["text"]}
+# )
+# ALL_OASST_PROMPTS = "all_oasst_prompts"
+# seed_datasets = {
+#     ALL_OASST_PROMPTS: iter(
+#         interleave_datasets(
+#             [oasst1_prompts, oasst2_prompts],
+#             probabilities=[0.5, 0.5],
+#             seed=seed,
+#             stopping_strategy="all_exhausted",
+#         )
+#     )
+# }
+
+# TODO change name to actual datset name
+seed_dataset_name = "prooompt/test_dataset"
+
+
+def get_seed_dataset():
+    return load_dataset(
+        seed_dataset_name,
+        split="train",
+        streaming=True,
+    )
+
+
+seed_dataset = iter(get_seed_dataset())
+
+
 class SeedDataManager:
     @staticmethod
-    def get_prompts(num_prompts=5):
-        return [
-            next_circular(seed_datasets, ALL_OASST_PROMPTS)["text"]
-            for _ in range(num_prompts)
-        ]
+    def get_prompt_and_completions():
+        global seed_dataset
+        try:
+            return SeedDataManager._map_seed_data(next(seed_dataset))
+        except StopIteration:
+            seed_dataset = iter(get_seed_dataset())
+            return SeedDataManager._map_seed_data(next(seed_dataset))
+
+    @staticmethod
+    def _map_seed_data(row: Dict) -> Tuple[str, List[str]]:
+        prompt = row["prompt"]
+        completions = [c["response"] for c in row["responses"]]
+        return prompt, completions
diff --git a/neurons/validator.py b/neurons/validator.py
index fdedd285..dc4c8f9b 100644
--- a/neurons/validator.py
+++ b/neurons/validator.py
@@ -4,6 +4,7 @@
 
 import bittensor as bt
 from commons.data_manager import DataManager
+from commons.dataset import SeedDataManager
 from commons.objects import DendriteQueryResponse
 from commons.consensus import Consensus
 
@@ -54,7 +55,7 @@ async def _update_score_and_send_feedback(self):
             await self._forward_consensus(
                 synapse=d.result, hotkeys=list(d.hotkey_to_scores.keys())
             )
-            asyncio.sleep(5)
+            await asyncio.sleep(5)
 
     async def forward(self):
         """
@@ -65,24 +66,14 @@ async def forward(self):
         - Rewarding the miners
         - Updating the scores
         """
-
-        # TODO change to real data
+        prompt, completions = SeedDataManager.get_prompt_and_completions(n=1)
         request = RankingRequest(
-            n_completions=3,
+            n_completions=len(completions),
             pid=get_new_uuid(),
-            prompt="What is your name?",
-            completions=[
-                Completion(
-                    text="My name is Assistant, and I am a helpful assisstant created by OpenAI.",
-                ),
-                Completion(
-                    text="My name is Llama, and I am an assistant created by Meta.",
-                ),
-                Completion(
-                    text="My name is GPT-3, and I am an AI created by OpenAI.",
-                ),
-            ],
+            prompt=prompt,
+            completions=[Completion(text=c) for c in completions],
         )
+
         miner_uids = get_random_uids(
             metagraph=self.metagraph,
             k=self.config.neuron.sample_size,