Bittensor 7.0.0, vllm 0.4.2 upgrade

prompting/base/validator.py

@@ -142,6 +142,9 @@ def run(self):
                     self.loop.run_until_complete(
                         asyncio.wait_for(task, timeout=forward_timeout)
                     )
+                except torch.cuda.OutOfMemoryError as e:
+                    bt.logging.error(f"Out of memory error: {e}")
+                    continue
                 except MaxRetryError as e:
                     bt.logging.error(f"MaxRetryError: {e}")
                     continue

prompting/forward.py

@@ -16,28 +16,30 @@
 # DEALINGS IN
 #  THE SOFTWARE.
 
-import asyncio
-import random
 import sys
 import time
+import random
+import asyncio
 import traceback
-from dataclasses import dataclass
-from typing import Awaitable, Dict, List
-
-import bittensor as bt
 import numpy as np
-import torch
-
+import bittensor as bt
+from typing import List, Dict, Awaitable
 from prompting.agent import HumanAgent
-from prompting.conversation import create_task
 from prompting.dendrite import DendriteResponseEvent
+from prompting.conversation import create_task
 from prompting.protocol import StreamPromptingSynapse
 from prompting.rewards import RewardResult
 from prompting.tasks import QuestionAnsweringTask
+from prompting.utils.uids import get_random_uids
 from prompting.utils.logging import log_event
 from prompting.utils.misc import async_log, serialize_exception_to_string
-from prompting.utils.uids import get_random_uids
+from dataclasses import dataclass
 
+@async_log
+async def generate_reference(agent):
+    loop = asyncio.get_running_loop()
+    result = await loop.run_in_executor(None, agent.task.generate_reference, agent.llm_pipeline)
+    return result
 
 @async_log
 async def execute_dendrite_call(dendrite_call):
@@ -165,6 +167,7 @@ def log_stream_results(stream_results: List[StreamResult]):
             f"Failed response for uid {failed_response.uid}: {formatted_exception}"
         )
 
+
 async def run_step(
     self, agent: HumanAgent, roles: List[str], messages: List[str], k: int, timeout: float, exclude: list = None
 ):
@@ -271,8 +274,6 @@ async def forward(self):
     """
     Encapsulates a full conversation between the validator and miners. Contains one or more rounds of request-response.
 
-    Raises:
-        torch.cuda.OutOfMemoryError: CUDA out of memory error.
     """
     bt.logging.info("🚀 Starting forward loop...")
     forward_start_time = time.time()
@@ -353,10 +354,6 @@ async def forward(self):
             messages.append(agent.challenge)
             turn += 1
 
-        except torch.cuda.OutOfMemoryError as err:
-            bt.logging.error("CUDA out of memory", str(err))
-            raise err
-
         except BaseException as e:
             unexpected_errors = serialize_exception_to_string(e)
             bt.logging.error(

requirements.txt

@@ -1,28 +1,34 @@
-angle_emb
-bittensor==6.10.1
-bs4
+angle_emb==0.4.4
+bittensor==7.0.0
+bs4==0.0.2
 click==8.1.3
 datasets==2.14.6
 deprecation==2.1.0
-torch==2.1.1
+torch==2.3.0
 torchmetrics
 pre-commit==3.3.2
 git+https://github.com/synapse-alpha/mathgenerator.git@main#egg=mathgenerator
 numpy==1.22.0
-rouge
+rouge==1.0.1
 scipy==1.10.1
 sentencepiece
 wandb==0.15.10
 tenacity
 antlr4-python3-runtime==4.11
-wikipedia
+wikipedia==1.4.0
 openai==1.9.0
-langchain==0.1.5
+langchain==0.2.1
+langchain_core==0.2.3
 langchainhub==0.1.14
+python-dotenv==1.0.1
+wikipedia_sections==2.0.0
+vllm==0.4.3
+loguru==0.7.2
+argostranslate==1.9.6
 python-dotenv
 wikipedia_sections
 vllm
 loguru
 argostranslate
-transformers==4.41.1
+transformers==4.41.2
 autoawq==0.2.5