Merge branch 'main' of https://github.com/surcyf123/smart-scrape

docs/running_a_validator.md

@@ -88,7 +88,6 @@ pm2 start neurons/validators/api.py --interpreter /usr/bin/python3  --name valid
 - `--neuron.run_all_miner_syn_qs_interval`: Sets the interval, in seconds, for querying all miners with synthetic questions. Set to a positive value to enable. A value of 0 disables this feature.
 - `--reward.prompt_based_weight`: adjusts the influence of a scoring model that evaluates the accuracy and relevance of a node's responses to given prompts.
 - `--reward.prompt_summary_links_content_based_weight`: Specifies the weight for the reward model that evaluates the relevance and quality of summary text in conjunction with linked content data.
-- `--reward.rlhf_weight`: Weight for the rlhf reward model
 - `--neuron.only_allowed_miners`: A list of miner identifiers, hotkey
 - `--neuron.disable_twitter_links_content_fetch`: Enables the option to skip fetching content data for Twitter links, relying solely on the data provided by miners
 

neurons/miners/config.py

@@ -80,6 +80,24 @@ def get_config() -> "bt.Config":
         default=True
     )
 
+    parser.add_argument(
+        "--miner.openai_summary_model",
+        default="gpt-3.5-turbo-1106",
+        help="OpenAI model used for summarizing content.",
+    )
+
+    parser.add_argument(
+        "--miner.openai_query_model",
+        default="gpt-3.5-turbo-1106",
+        help="OpenAI model used for generating queries.",
+    )
+
+    parser.add_argument(
+        "--miner.openai_fix_query_model",
+        default="gpt-4-1106-preview",
+        help="OpenAI model used for fixing queries.",
+    )
+
     # Adds subtensor specific arguments i.e. --subtensor.chain_endpoint ... --subtensor.network ...
     bt.subtensor.add_args(parser)
 

neurons/miners/miner.py

@@ -48,8 +48,6 @@
 client = AsyncOpenAI(timeout=60.0)
 valid_hotkeys = []
 
-
-
 class StreamMiner(ABC):
     def __init__(self, config=None, axon=None, wallet=None, subtensor=None):
         bt.logging.info("starting stream miner")

neurons/miners/twitter_miner.py

@@ -42,15 +42,15 @@ def __init__(self, miner: any):
         self.miner = miner
 
     async def intro_text(self, model, prompt, send, is_intro_text):
-        bt.logging.info("miner.intro_text => ", self.miner.config.miner.intro_text)
-        bt.logging.info("Synapse.is_intro_text => ", is_intro_text)
+        bt.logging.trace("miner.intro_text => ", self.miner.config.miner.intro_text)
+        bt.logging.trace("Synapse.is_intro_text => ", is_intro_text)
         if not self.miner.config.miner.intro_text:
             return
 
         if not is_intro_text:
             return
 
-        bt.logging.info(f"Run intro text")
+        bt.logging.trace(f"Run intro text")
 
         content = f"""
         Generate introduction for that prompt: "{prompt}",
@@ -93,7 +93,7 @@ async def intro_text(self, model, prompt, send, is_intro_text):
                     }
                 )
                 await asyncio.sleep(0.1)  # Wait for 100 milliseconds
-                bt.logging.info(f"Streamed tokens: {joined_buffer}")
+                bt.logging.trace(f"Streamed tokens: {joined_buffer}")
                 buffer = []
 
         return buffer
@@ -105,7 +105,12 @@ async def fetch_tweets(self, prompt):
             #todo we can find tweets based on twitter_query
             filtered_tweets = get_random_tweets(15)
         else:
-            tw_client  = TwitterAPIClient()
+            openai_query_model = self.miner.config.miner.openai_query_model
+            openai_fix_query_model = self.miner.config.miner.openai_fix_query_model
+            tw_client  = TwitterAPIClient(
+                openai_query_model=openai_query_model,
+                openai_fix_query_model=openai_fix_query_model
+            )
             filtered_tweets, prompt_analysis = await tw_client.analyse_prompt_and_fetch_tweets(prompt)
         return filtered_tweets, prompt_analysis
 
@@ -167,8 +172,11 @@ async def twitter_scraper(self, synapse: TwitterScraperStreaming, send: Send):
             prompt = synapse.messages
             seed = synapse.seed
             is_intro_text = synapse.is_intro_text
-            bt.logging.info(synapse)
-            bt.logging.info(f"question is {prompt} with model {model}, seed: {seed}")
+            bt.logging.trace(synapse)
+
+            bt.logging.info("================================== Prompt ===================================")
+            bt.logging.info(prompt)
+            bt.logging.info("================================== Prompt ====================================")
 
             # buffer.append('Test 2')
             intro_response, (tweets, prompt_analysis) = await asyncio.gather(
@@ -188,7 +196,8 @@ async def twitter_scraper(self, synapse: TwitterScraperStreaming, send: Send):
             # else:
             #     synapse.set_tweets(tweets)
 
-            response = await self.finalize_data(prompt=prompt, model=model, filtered_tweets=tweets, prompt_analysis=prompt_analysis)
+            openai_summary_model = self.miner.config.miner.openai_summary_model
+            response = await self.finalize_data(prompt=prompt, model=openai_summary_model, filtered_tweets=tweets, prompt_analysis=prompt_analysis)
 
             # Reset buffer for finalizing data responses
             buffer = []
@@ -213,7 +222,7 @@ async def twitter_scraper(self, synapse: TwitterScraperStreaming, send: Send):
                             "more_body": True,
                         }
                     )
-                    bt.logging.info(f"Streamed tokens: {joined_buffer}")
+                    bt.logging.trace(f"Streamed tokens: {joined_buffer}")
                     buffer = []  # Clear the buffer for the next set of tokens
 
             joined_full_text = "".join(full_text)  # Join all text chunks

neurons/validators/config.py

@@ -100,13 +100,6 @@ def add_args(cls, parser):
         default=0.05,
     )
 
-    parser.add_argument(
-        "--reward.rlhf_weight",
-        type=float,
-        help="Weight for the rlhf reward model",
-        default=DefaultRewardFrameworkConfig.rlhf_model_weight,
-    )
-
     parser.add_argument(
         "--reward.prompt_based_weight",
         type=float,

neurons/validators/reward/config.py

@@ -24,7 +24,6 @@ class RewardModelType(Enum):
     sentence_match_penalty = "sentence_match_penalty"
     rlhf = "rlhf_reward_model"
     prompt = "prompt_reward_model"
-    dpo = "dpo_reward_model"
 
 class RewardScoringType(Enum):
     twitter_question_answer_score = "twitter_question_answer_score"
@@ -35,7 +34,5 @@ class DefaultRewardFrameworkConfig:
     """Reward framework default configuration.
     Note: All the weights should add up to 1.0.
     """
-    dpo_model_weight: float = 0
-    rlhf_model_weight: float = 0
     prompt_model_weight: float = 1
     prompt_model_summary_links_content_weight: float = 0