[pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

applications/ColossalChat/coati/dataset/loader.py

@@ -89,15 +89,19 @@ def __call__(self, instances: Sequence[Dict[str, List[int]]]) -> Dict[str, torch
 
         # `List[torch.Tensor]`
         batch_input_ids = [
-            torch.LongTensor(instance["input_ids"][: self.max_length])
-            if len(instance["input_ids"]) > self.max_length
-            else torch.LongTensor(instance["input_ids"])
+            (
+                torch.LongTensor(instance["input_ids"][: self.max_length])
+                if len(instance["input_ids"]) > self.max_length
+                else torch.LongTensor(instance["input_ids"])
+            )
             for instance in instances
         ]
         batch_labels = [
-            torch.LongTensor(instance["labels"][: self.max_length])
-            if len(instance["labels"]) > self.max_length
-            else torch.LongTensor(instance["labels"])
+            (
+                torch.LongTensor(instance["labels"][: self.max_length])
+                if len(instance["labels"]) > self.max_length
+                else torch.LongTensor(instance["labels"])
+            )
             for instance in instances
         ]
         if self.tokenizer.padding_side == "right":

applications/ColossalChat/coati/models/loss.py

@@ -1,6 +1,7 @@
 """
 loss functions
 """
+
 from typing import Optional, Tuple
 
 import torch

applications/ColossalChat/coati/models/reward_model.py

@@ -1,6 +1,7 @@
 """
 reward model
 """
+
 from typing import Optional
 
 import torch

applications/ColossalChat/coati/trainer/utils.py

@@ -1,6 +1,7 @@
 """
 Training utilities for Coati.
 """
+
 from typing import Any
 
 import torch

applications/ColossalEval/colossal_eval/dataset/agieval.py

@@ -78,7 +78,9 @@ def get_prompt(line: Dict, dataset_name: str, logger: DistributedLogger) -> Dict
             option_string = "ABCDEFG"
             count = len(line["options"])
 
-            input = "问题：" + line["question"] + " " + "从以下选项中选择：" + " ".join(line["options"]) + "\n" + "答案："
+            input = (
+                "问题：" + line["question"] + " " + "从以下选项中选择：" + " ".join(line["options"]) + "\n" + "答案："
+            )
 
             all_classes = list(option_string[0:count])
 
@@ -150,7 +152,15 @@ def combine_prompt(prompt_path, dataset_name, load_explanation=True, chat_mode=F
             )
         elif dataset_name in chinese_qa_datasets:
             question_input = (
-                "问题：" + passage + " " + question + "\n" + "从以下选项中选择：" + " ".join(options) + "\n" + "答案：{}".format(label)
+                "问题："
+                + passage
+                + " "
+                + question
+                + "\n"
+                + "从以下选项中选择："
+                + " ".join(options)
+                + "\n"
+                + "答案：{}".format(label)
             )
         elif dataset_name in english_cloze_datasets:
             question_input = "Question: ".format(idx + 1) + question + "\n" + "Answer: {}".format(answer)

applications/ColossalEval/colossal_eval/dataset/ceval.py

@@ -57,7 +57,11 @@
     "urban_and_rural_planner": ["Urban and Rural Planner", "注册城乡规划师", "Other"],
     "accountant": ["Accountant", "注册会计师", "Other"],
     "fire_engineer": ["Fire Engineer", "注册消防工程师", "Other"],
-    "environmental_impact_assessment_engineer": ["Environmental Impact Assessment Engineer", "环境影响评价工程师", "Other"],
+    "environmental_impact_assessment_engineer": [
+        "Environmental Impact Assessment Engineer",
+        "环境影响评价工程师",
+        "Other",
+    ],
     "tax_accountant": ["Tax Accountant", "税务师", "Other"],
     "physician": ["Physician", "医师资格", "Other"],
 }

applications/ColossalEval/colossal_eval/dataset/mtbench.py

@@ -56,9 +56,11 @@ def load(path: str, logger: DistributedLogger, few_shot: bool) -> List[Dict]:
                     "instruction": question["turns"],
                     "input": "",
                     "output": [],
-                    "target": [""] * turn_number
-                    if question["question_id"] not in reference
-                    else reference[question["question_id"]],
+                    "target": (
+                        [""] * turn_number
+                        if question["question_id"] not in reference
+                        else reference[question["question_id"]]
+                    ),
                 }
 
                 if category in dataset["test"]:

applications/ColossalEval/colossal_eval/models/huggingface.py

@@ -77,7 +77,9 @@ def _get_choices_indices(self, language: str):
             self.indices_for_choices[0].append(
                 self.tokenizer(f"Answer: {choice}", add_special_tokens=False).input_ids[-1]
             )
-            self.indices_for_choices[1].append(self.tokenizer(f"答案：{choice}", add_special_tokens=False).input_ids[-1])
+            self.indices_for_choices[1].append(
+                self.tokenizer(f"答案：{choice}", add_special_tokens=False).input_ids[-1]
+            )
 
     def _load_tokenizer(self, path: str, tokenizer_path: Optional[str], tokenizer_kwargs: dict):
         """

applications/ColossalQA/colossalqa/chain/retrieval_qa/base.py

@@ -7,6 +7,7 @@
 https://github.com/langchain-ai/langchain
 The original code is licensed under the MIT license.
 """
+
 from __future__ import annotations
 
 import copy

applications/ColossalQA/colossalqa/chain/retrieval_qa/load_chain.py

@@ -8,6 +8,7 @@
 https://github.com/langchain-ai/langchain
 The original code is licensed under the MIT license.
 """
+
 import copy
 from typing import Any, Mapping, Optional, Protocol
 

applications/ColossalQA/colossalqa/chain/retrieval_qa/stuff.py

@@ -7,6 +7,7 @@
 https://github.com/langchain-ai/langchain
 The original code is licensed under the MIT license.
 """
+
 import copy
 from typing import Any, List
 

applications/ColossalQA/colossalqa/data_loader/table_dataloader.py

@@ -2,7 +2,6 @@
 Class for loading table type data. please refer to Pandas-Input/Output for file format details.
 """
 
-
 import glob
 import os
 

applications/ColossalQA/colossalqa/local/colossalcloud_llm.py

@@ -20,6 +20,7 @@
 print(resp)  # super-heavyweight awesome-natured yawning Australian creature!
 
 """
+
 import json
 from typing import Any, Mapping
 

applications/ColossalQA/colossalqa/local/llm.py

@@ -12,6 +12,7 @@
 logger.info(llm(TEST_PROMPT_CHATGLM, max_new_tokens=100), verbose=True)
 
 """
+
 from typing import Any, List, Mapping, Optional
 
 import torch

applications/ColossalQA/colossalqa/local/utils.py

@@ -1,6 +1,7 @@
 """
 Generation utilities
 """
+
 import json
 from typing import List
 

applications/ColossalQA/colossalqa/memory.py

@@ -2,6 +2,7 @@
 Implement a memory class for storing conversation history
 Support long term and short term memory
 """
+
 from typing import Any, Dict, List
 
 from colossalqa.chain.memory.summary import ConversationSummaryMemory

applications/ColossalQA/colossalqa/mylogging.py

@@ -1,6 +1,7 @@
 """
 Class for logging with extra control for debugging
 """
+
 import logging
 
 

applications/ColossalQA/colossalqa/retrieval_conversation_en.py

@@ -1,6 +1,7 @@
 """
 Script for Chinese retrieval based conversation system backed by ChatGLM
 """
+
 from typing import Tuple
 
 from colossalqa.chain.retrieval_qa.base import RetrievalQA

applications/ColossalQA/colossalqa/retrieval_conversation_universal.py

@@ -1,6 +1,7 @@
 """
 Multilingual retrieval based conversation system
 """
+
 from typing import List
 
 from colossalqa.data_loader.document_loader import DocumentLoader

applications/ColossalQA/colossalqa/retrieval_conversation_zh.py

@@ -1,6 +1,7 @@
 """
 Script for Chinese retrieval based conversation system backed by ChatGLM
 """
+
 from typing import Tuple
 
 from colossalqa.chain.retrieval_qa.base import RetrievalQA

applications/ColossalQA/colossalqa/retriever.py

@@ -1,6 +1,7 @@
 """
 Code for custom retriver with incremental update
 """
+
 import copy
 import hashlib
 import os

applications/ColossalQA/colossalqa/text_splitter/chinese_text_splitter.py

@@ -1,6 +1,7 @@
 """
 Code for Chinese text splitter
 """
+
 from typing import Any, List, Optional
 
 from colossalqa.text_splitter.utils import get_cleaned_paragraph

applications/ColossalQA/examples/retrieval_conversation_en.py

@@ -1,6 +1,7 @@
 """
 Script for English retrieval based conversation system backed by LLaMa2
 """
+
 import argparse
 import os
 

applications/ColossalQA/examples/retrieval_conversation_en_customer_service.py

@@ -1,6 +1,7 @@
 """
 Script for English retrieval based conversation system backed by LLaMa2
 """
+
 import argparse
 import json
 import os

applications/ColossalQA/examples/retrieval_conversation_zh.py

@@ -1,6 +1,7 @@
 """
 Script for Chinese retrieval based conversation system backed by ChatGLM
 """
+
 import argparse
 import os
 

applications/ColossalQA/examples/retrieval_intent_classification_zh_customer_service.py

@@ -1,6 +1,7 @@
 """
 Script for English retrieval based conversation system backed by LLaMa2
 """
+
 import argparse
 import os
 

colossalai/auto_parallel/meta_profiler/meta_registry/conv.py

@@ -107,20 +107,22 @@ def convnd_meta_info(*args, **kwargs) -> Tuple[TrainCycleItem, TrainCycleItem, L
     # NOTE: currently in SPMD solver we always believe that there will be a new tensor created in forward
     fwd_memory_cost = MemoryCost(
         activation=compute_size_in_bytes([input_tensor, output_tensor]),
-        parameter=compute_size_in_bytes([weight_tensor, bias_tensor])
-        if has_bias
-        else compute_size_in_bytes(weight_tensor),
+        parameter=(
+            compute_size_in_bytes([weight_tensor, bias_tensor]) if has_bias else compute_size_in_bytes(weight_tensor)
+        ),
         temp=0,
         buffer=0,
     )
 
     bwd_memory_cost = MemoryCost(
-        activation=compute_size_in_bytes([input_tensor, weight_tensor, bias_tensor])
-        if has_bias
-        else compute_size_in_bytes([input_tensor, weight_tensor]),
-        parameter=compute_size_in_bytes([weight_tensor, bias_tensor])
-        if has_bias
-        else compute_size_in_bytes(weight_tensor),
+        activation=(
+            compute_size_in_bytes([input_tensor, weight_tensor, bias_tensor])
+            if has_bias
+            else compute_size_in_bytes([input_tensor, weight_tensor])
+        ),
+        parameter=(
+            compute_size_in_bytes([weight_tensor, bias_tensor]) if has_bias else compute_size_in_bytes(weight_tensor)
+        ),
         temp=0,
         buffer=0,
     )

colossalai/inference/batch_bucket.py

@@ -247,16 +247,16 @@ def add_seqs(
                 self._sequences_dict[seq.request_id] = seq
                 self._sequences_indexes[seq.request_id] = self._current_batch_size + i
             # TODO external (rename): modify Sequence.sentence_len to seq_len
-            self._sequence_lengths[
-                self._current_batch_size : self._current_batch_size + num_seqs_to_add
-            ] = torch.tensor([seq.sentence_len for seq in seqs[:num_seqs_to_add]], dtype=torch.int32)
+            self._sequence_lengths[self._current_batch_size : self._current_batch_size + num_seqs_to_add] = (
+                torch.tensor([seq.sentence_len for seq in seqs[:num_seqs_to_add]], dtype=torch.int32)
+            )
             # NOTE block tables to be updated by kvcache manager
             block_tables = self._block_tables[self._current_batch_size : self._current_batch_size + num_seqs_to_add]
             if alloc_block_tables is not None:
                 # copy block ids from provided block tables
-                self._block_tables[
-                    self._current_batch_size : self._current_batch_size + num_seqs_to_add
-                ] = alloc_block_tables
+                self._block_tables[self._current_batch_size : self._current_batch_size + num_seqs_to_add] = (
+                    alloc_block_tables
+                )
             elif alloc_block_tables_fn:
                 alloc_block_tables_fn(
                     block_tables,

colossalai/inference/config.py

@@ -1,6 +1,7 @@
 """
 Our config contains various options for inference optimization, it is a unified API that wraps all the configurations for inference.
 """
+
 import logging
 from abc import ABC, abstractmethod
 from dataclasses import dataclass, fields
@@ -82,9 +83,9 @@ class InputMetaData(RPC_PARAM):
     dtype: torch.dtype = torch.float32
     use_spec_dec: bool = False
     num_tokens_to_verify: int = 0
-    batch_token_ids: Optional[
-        List[List[int]]
-    ] = None  # for `repetition_penalty`, `no_repeat_ngram_size` in sampler process
+    batch_token_ids: Optional[List[List[int]]] = (
+        None  # for `repetition_penalty`, `no_repeat_ngram_size` in sampler process
+    )
 
     def to_rpc_param(self) -> Dict[str, any]:
         return {
@@ -202,9 +203,9 @@ class InferenceConfig(RPC_PARAM):
     prompt_template: Optional[str] = None
     do_sample: bool = False
     beam_width: int = 1  # TODO: beam search is not support for now
-    prefill_ratio: Optional[
-        float
-    ] = 1.2  # the ratio of prefill sequences to decoding sequences, we do prefill step once the actual value exceeds ratio
+    prefill_ratio: Optional[float] = (
+        1.2  # the ratio of prefill sequences to decoding sequences, we do prefill step once the actual value exceeds ratio
+    )
     pad_input: bool = False
     early_stopping: Optional[bool] = False
     top_k: Optional[int] = 50
@@ -234,7 +235,9 @@ class InferenceConfig(RPC_PARAM):
     high_precision: Optional[bool] = False
 
     # cuda_graph
-    use_cuda_graph: bool = False  # NOTE only when we have the graph for specific decoding batch size can we use the cuda graph for inference
+    use_cuda_graph: bool = (
+        False  # NOTE only when we have the graph for specific decoding batch size can we use the cuda graph for inference
+    )
     max_context_len_to_capture: int = 512
 
     # StreamingLLM (sliding window attention with attention sinks)

colossalai/inference/core/engine.py

@@ -46,7 +46,6 @@
 
 
 class InferenceEngine:
-
     """
     InferenceEngine which manages the inference process..
 

colossalai/inference/core/rpc_engine.py

@@ -34,7 +34,6 @@ def run_server(host, port, event: mp.Event = None):
 
 
 class RPCInferenceEngine(InferenceEngine):
-
     """
     InferenceEngine which manages the inference process..
 

colossalai/inference/executor/rpc_worker.py

@@ -41,7 +41,6 @@
 
 
 class rpcWorkerService(rpyc.Service):
-
     """
     Execute the computation tasks and manage its own kv cache
 

colossalai/inference/kv_cache/kvcache_manager.py

@@ -279,9 +279,11 @@ def allocate_context_from_block_tables(self, block_tables: torch.Tensor, context
             block.add_ref()
             self._allocate_on_block(
                 block,
-                block.block_size
-                if context_lengths[i] % block.block_size == 0
-                else context_lengths[i].item() % block.block_size,
+                (
+                    block.block_size
+                    if context_lengths[i] % block.block_size == 0
+                    else context_lengths[i].item() % block.block_size
+                ),
             )
         for block_id in alloc_block_ids:
             if block_id in alloc_block_ids[last_block_locs]: