Added test plots

app.py

@@ -1,5 +1,5 @@
 # Code adapted from: https://huggingface.co/spaces/RaoFoundation/pretraining-leaderboard/blob/main/app.py
-'''
+
 import argparse
 import functools
 import traceback
@@ -22,30 +22,23 @@ import pandas as pd
 
 load_dotenv()
 
-'''
-import gradio as gr
+
+
 FONT = (
     """<link href="https://fonts.cdnfonts.com/css/jmh-typewriter" rel="stylesheet">"""
 )
 TITLE = """<h1 align="center" id="space-title" class="typewriter">Subnet 9 Leaderboard</h1>"""
 HEADER = """<h2 align="center" class="typewriter"><a href="https://github.com/macrocosm-os/pretraining" target="_blank">Subnet 9</a> is a <a href="https://bittensor.com/" target="_blank">Bittensor</a> subnet that rewards miners for producing pretrained Foundation-Models on the <a href="https://huggingface.co/datasets/tiiuae/falcon-refinedweb" target="_blank">Falcon Refined Web dataset</a>. It acts like a continuous benchmark whereby miners are rewarded for attaining the best losses on randomly sampled pages of Falcon.<br/>The models with the best head-to-head loss on the evaluation data receive a steady emission of TAO.</h3>"""
 
-demo = gr.Blocks(css=".typewriter {font-family: 'JMH Typewriter', sans-serif;}")
-with demo:
-    gr.HTML(FONT)
-    gr.HTML(TITLE)
-    gr.HTML(HEADER)
-    demo.launch(share=True)
-'''
 EVALUATION_DETAILS = """<ul><li><b>Name:</b> the 🤗 Hugging Face model name (click to go to the model card)</li><li><b>Rewards / Day:</b> the expected rewards per day based on current ranking.</li><li><b>Last Average Loss:</b> the last loss value on the evaluation data for the model as calculated by a validator (lower is better)</li><li><b>UID:</b> the Bittensor UID of the miner</li><li><b>Block:</b> the Bittensor block that the model was submitted in</li></ul><br/>More stats on <a href="https://taostats.io/subnets/netuid-9/" target="_blank">taostats</a>."""
 EVALUATION_HEADER = """<h3 align="center">Shows the latest internal evaluation statistics as calculated by the Opentensor validator</h3>"""
 VALIDATOR_WANDB_PROJECT = "opentensor-dev/pretraining-subnet"
-BENCHMARK_WANDB_PROJECT = "raofoundation/pretraining-leaderboard-data"
-H4_TOKEN = os.environ.get("H4_TOKEN", None)
-API = HfApi(token=H4_TOKEN)
+#BENCHMARK_WANDB_PROJECT = "raofoundation/pretraining-leaderboard-data"
+H4_TOKEN = os.environ.get("HF_TOKEN", None)
+API = HfApi(token=HF_TOKEN)
 WANDB_TOKEN = os.environ.get("WANDB_API_KEY", None)
 SUBTENSOR_ENDPOINT=os.environ.get("SUBTENSOR_ENDPOINT", None)
-REPO_ID = "RaoFoundation/pretraining-leaderboard"
+REPO_ID = "macrocosm-os/sn9"
 MAX_AVG_LOSS_POINTS = 1
 RETRIES = 5
 DELAY_SECS = 3
@@ -321,7 +314,7 @@ def get_benchmarks() -> Tuple[pd.DataFrame, datetime.datetime]:
 
 
 def restart_space():
-    API.restart_space(repo_id=REPO_ID, token=H4_TOKEN)
+    API.restart_space(repo_id=REPO_ID, token=HF_TOKEN)
 
 
 def main():
@@ -347,7 +340,7 @@ def main():
             for uid, stats in validator_df.items():
                 weight_keys.update(stats[-1].keys())
                 
-            benchmarks, benchmark_timestamp = get_benchmarks()
+            #benchmarks, benchmark_timestamp = get_benchmarks()
             break
         except Exception as e:
             print(f"Failed to get data: {e}")
@@ -371,11 +364,13 @@ def main():
             num_top_classes=10,
         )
         
+        '''
         if benchmarks is not None:
             with gr.Accordion("Top Model Benchmarks"):
                 gr.components.Dataframe(benchmarks)
                 gr.HTML("""<div>PPL computed using a stride of 512. See <a href='https://github.com/RaoFoundation/pretraining/blob/dev/scripts/run_benchmarks.py'>here</a> for the full code.</div>""")
                 gr.HTML(f"""<div>Last Updated: {benchmark_timestamp.strftime("%Y-%m-%d %H:%M:%S")} (UTC)</div>""")
+        '''
 
         with gr.Accordion("Evaluation Stats"):
             gr.HTML(EVALUATION_HEADER)

requirements.txt

@@ -0,0 +1,7 @@
+bittensor==6.9.3
+requests==2.31.0
+wandb==0.16.2
+python-dotenv==1.0.1
+APScheduler==3.10.1
+huggingface-hub>=0.18.0
+pandas==2.2.0