chore: remove unused conveyor code

conveyor/plugin/calculator_plugin.py

@@ -32,4 +32,3 @@ def compute(self, data: dict):
         self.answer = eval(query)
         end = time.perf_counter()
         self.time += end - start
-

conveyor/plugin/scheduler.py

@@ -136,7 +136,6 @@ def poll_finished(self, client_id: str) -> Optional[List]:
         for plugin in plugin_list:
             if plugin.local_pipe.poll():
                 res = plugin.local_pipe.recv()
-                # logging.debug(f"[PluginScheduler:{client_id}] Finished: {res}")
                 self.join_queue.append(plugin.process)
                 logging.debug(f"[PluginScheduler:{client_id}] Process joined")
                 plugin_list.remove(plugin)

conveyor/scheduling/context.py

@@ -14,7 +14,7 @@
 from conveyor.models.config import ModelConfig
 
 from conveyor.scheduling.cache_manager import CacheManager
-from conveyor.scheduling.parsing import BaseParser, FunctionaryParser
+from conveyor.scheduling.parsing import BaseParser
 from conveyor.utils import getLogger
 
 logging = getLogger(__name__)
@@ -102,9 +102,6 @@ def new(
                     config.head_dim,
                     1,
                 )
-        # logging.debug(
-        #     f"InferenceContext::new(): state={state}, req_ids={req_ids}, seq_lens={seq_lens}, filling_start_offset={filling_start_offset}, kv_indptr={kv_indptr}, kv_page_index={kv_page_index}, kv_last_page_lens={kv_last_page_lens}, qo_indptr={qo_indptr}"
-        # )
 
         return cls(
             state=state,
@@ -147,10 +144,7 @@ def __init__(self, req_id: int, input_text: str, tokenizer, parser: BaseParser):
         self.parser = parser
 
     def evaluate_parser(self, token: int):
-        tokens = self.parser.enqueue(token)
-        # if tokens is not None:
-        #     print(f"::: Evaluate Raw: !!@ {tokens} @!!")
-        #     print(f"::: Evaluate Decoded: !!@ {self.tokenizer.decode(tokens)} @!!")
+        self.parser.enqueue(token)
 
     def decode(self) -> str:
         return self.tokenizer.decode(self.tokens)

conveyor/scheduling/parsing.py

@@ -57,7 +57,6 @@ def enqueue(self, token) -> Optional[Dict | List]:
                     self.in_progress = False
                 self.buffer = []
                 self.string = ""
-                # print(f"The string is: ::>{self.string}<::")
                 return buf
             case _:
                 new_str: str = self.tokenizer.convert_ids_to_tokens(token)
@@ -99,7 +98,6 @@ def enqueue(self, token) -> Optional[Dict | List]:
                     self.in_progress = False
                 self.buffer = []
                 self.string = ""
-                # print(f"The string is: ::>{self.string}<::")
                 return buf
             case _:
                 new_str: str = self.tokenizer.convert_ids_to_tokens(token)
@@ -121,7 +119,6 @@ class FunctionaryParser(BaseParser):
     def __init__(self, tokenizer, client_id, start_cb, update_cb, finish_cb):
         self.buffer = []
         self.string = ""
-        # self.left_bracket_pos = []
         self.tokenizer = tokenizer
         self.start_cb = start_cb
         self.update_cb = update_cb

conveyor/scheduling/scheduler.py

@@ -280,7 +280,6 @@ def iteration_step(
         self.roundrobin_policy()
         # self.fcfs_policy()
         next_operation = self.schedule_next_operation()
-        # logging.debug(f"Scheduler: next operation={next_operation}")
         match next_operation:
             case InferenceState.APPEND:
                 logits, _ = self.forward_append(self.context)
@@ -321,7 +320,6 @@ def prepare_kv_page(
         page_needed, page_idx_start = compute_page_needed(
             sched_ctx.seq_lens, sched_ctx.completed_lens, cache_manager.page_size
         )
-        # logging.debug(f"page_needed={page_needed}, page_idx_start={page_idx_start}")
 
         page_num_required = int(page_needed.sum().item())
         if page_num_required > 0:
@@ -332,9 +330,6 @@ def prepare_kv_page(
                 (page_needed.size(0) + 1,), dtype=torch.int64, device="cpu"
             )
             range_idx[1:] = page_needed.cumsum(dim=0)
-            # logging.debug(
-            #     f"Allocated pages: new_page_idx={new_page_idx}, range_idx={range_idx}"
-            # )
             for i in range(page_needed.size(0)):
                 cache_manager.req_page_mapping[
                     req_ids[i],
@@ -391,7 +386,6 @@ def forward_decode(self, sched_ctx: SchedulerContext) -> None:
         req_ids = torch.tensor(
             [req.req_id for req in sched_ctx.requests], dtype=torch.int64, device="cuda"
         )
-        # sched_ctx.seq_lens.add_(1)
 
         ScheduleEngine.prepare_kv_page(req_ids, sched_ctx, self.cache_manager)
 
@@ -468,7 +462,6 @@ def schedule_next_operation(self) -> InferenceState:
 
     def fcfs_policy(self) -> None:
         while self.new_request_available():
-            # TODO: more than one request can be added
             new_request = self.request_pool.pop_request()
             logging.info(f"Scheduler: new request={new_request.req_id}")
             self.context.add_active_request(new_request)