hpcaitech · binmakeswell · Mar 12, 2024 · Mar 6, 2024
@@ -117,8 +117,8 @@ def _call(
             ) = copy.deepcopy(buffered_history_backup), copy.deepcopy(summarized_history_temp_backup)
 
         # if rejection_trigger_keywords is not given, return the response from LLM directly
-        rejection_trigger_keywrods = inputs.get('rejection_trigger_keywrods', [])
-        answer = answer if all([rej not in answer for rej in rejection_trigger_keywrods]) else None
+        rejection_trigger_keywords = inputs.get('rejection_trigger_keywords', [])
+        answer = answer if all([rej not in answer for rej in rejection_trigger_keywords]) else None
         if answer is None: 
             answer = inputs.get('rejection_answer', "抱歉，根据提供的信息无法回答该问题。")
         if self.combine_documents_chain.memory is not None:
@@ -161,8 +161,8 @@ async def _acall(
             input_documents=docs, question=question, callbacks=_run_manager.get_child(), **kwargs
         )
         # if rejection_trigger_keywords is not given, return the response from LLM directly
-        rejection_trigger_keywrods = inputs.get('rejection_trigger_keywrods', [])
-        answer = answer if all([rej not in answer for rej in rejection_trigger_keywrods]) or len(rejection_trigger_keywrods)==0 else None
+        rejection_trigger_keywords = inputs.get('rejection_trigger_keywords', [])
+        answer = answer if all([rej not in answer for rej in rejection_trigger_keywords]) or len(rejection_trigger_keywords)==0 else None
         if answer is None:
             answer = inputs.get('rejection_answer', "抱歉，根据提供的信息无法回答该问题。")
         self.combine_documents_chain.memory.save_context({"question": question}, {"output": answer})

@@ -75,7 +75,7 @@
 # Below are English retrieval qa prompts
 
 _EN_RETRIEVAL_QA_PROMPT = """[INST] <<SYS>>Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist content.
-If the answer cannot be infered based on the given context, please say "I cannot answer the question based on the information given.".<</SYS>>
+If the answer cannot be inferred based on the given context, please say "I cannot answer the question based on the information given.".<</SYS>>
 Use the context and chat history to answer the question.
 
 context:
@@ -97,8 +97,8 @@
 Human: I have a friend, Mike. Do you know him?
 Assistant: Yes, I know a person named Mike
 
-sentence: What's his favorate food?
-disambiguated sentence: What's Mike's favorate food?
+sentence: What's his favorite food?
+disambiguated sentence: What's Mike's favorite food?
 [/INST]
 Chat history:
 {chat_history}

@@ -80,7 +80,7 @@ def run(self, user_input: str, memory: ConversationBufferWithSummary) -> Tuple[s
             self.retrieval_chain.run(
                 query=user_input,
                 stop=[self.memory.human_prefix + ": "],
-                rejection_trigger_keywrods=["cannot answer the question"],
+                rejection_trigger_keywords=["cannot answer the question"],
                 rejection_answer="Sorry, this question cannot be answered based on the information provided.",
             ).split("\n")[0],
             self.memory,

@@ -103,7 +103,7 @@ def load_supporting_docs(self, files: List[List[str]] = None, text_splitter: Tex
                     break
                 data_name = input("Enter a short description of the data:")
                 separator = input(
-                    "Enter a separator to force separating text into chunks, if no separator is given, the defaut separator is '\\n\\n', press ENTER directly to skip:"
+                    "Enter a separator to force separating text into chunks, if no separator is given, the default separator is '\\n\\n', press ENTER directly to skip:"
                 )
                 separator = separator if separator != "" else "\n\n"
                 retriever_data = DocumentLoader([[file, data_name.replace(" ", "_")]]).all_data

@@ -87,7 +87,7 @@ def run(self, user_input: str, memory: ConversationBufferWithSummary) -> Tuple[s
                 query=user_input,
                 stop=["</答案>"],
                 doc_prefix="支持文档",
-                rejection_trigger_keywrods=["无法回答该问题"],
+                rejection_trigger_keywords=["无法回答该问题"],
                 rejection_answer="抱歉，根据提供的信息无法回答该问题。",
             ).split("\n")[0],
             self.memory,

@@ -61,7 +61,7 @@
     information_retriever.add_documents(docs=documents, cleanup="incremental", mode="by_source", embedding=embedding)
 
     prompt_template = """Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.
-    If the answer cannot be infered based on the given context, please don't share false information.
+    If the answer cannot be inferred based on the given context, please don't share false information.
     Use the context and chat history to respond to the human's input at the end or carry on the conversation. You should generate one response only. No following up is needed.
 
     context:

@@ -67,7 +67,7 @@ def disambiguity(input):
             break
         data_name = input("Enter a short description of the data:")
         separator = input(
-            "Enter a separator to force separating text into chunks, if no separator is given, the defaut separator is '\\n\\n'. Note that"
+            "Enter a separator to force separating text into chunks, if no separator is given, the default separator is '\\n\\n'. Note that"
             + "we use neural text spliter to split texts into chunks, the seperator only serves as a delimiter to force split long passage into"
             + " chunks before passing to the neural network. Press ENTER directly to skip:"
         )
@@ -112,7 +112,7 @@ def disambiguity(input):
         agent_response = retrieval_chain.run(
             query=user_input,
             stop=["Human: "],
-            rejection_trigger_keywrods=EN_RETRIEVAL_QA_TRIGGER_KEYWORDS,
+            rejection_trigger_keywords=EN_RETRIEVAL_QA_TRIGGER_KEYWORDS,
             rejection_answer=EN_RETRIEVAL_QA_REJECTION_ANSWER,
         )
         agent_response = agent_response.split("\n")[0]

@@ -142,7 +142,7 @@ def metadata_func(data_sample, additional_fields):
         agent_response = retrieval_chain.run(
             query=user_input,
             stop=["Human: "],
-            rejection_trigger_keywrods=EN_RETRIEVAL_QA_TRIGGER_KEYWORDS,
+            rejection_trigger_keywords=EN_RETRIEVAL_QA_TRIGGER_KEYWORDS,
             rejection_answer=EN_RETRIEVAL_QA_REJECTION_ANSWER,
         )
         agent_response = agent_response.split("\n")[0]

@@ -11,7 +11,7 @@
     parser.add_argument('--sql_file_path', type=str, default=None, help='path to the a empty folder for storing sql files for indexing')
     args = parser.parse_args()
 
-    # Will ask for documents path in runnning time
+    # Will ask for documents path in running time
     session = UniversalRetrievalConversation(files_en=None, 
                 files_zh=None, 
                 zh_model_path=args.zh_model_path, en_model_path=args.en_model_path,

@@ -107,7 +107,7 @@ def disambiguity(input: str):
             query=user_input,
             stop=["</答案>"],
             doc_prefix="支持文档",
-            rejection_trigger_keywrods=ZH_RETRIEVAL_QA_TRIGGER_KEYWORDS,
+            rejection_trigger_keywords=ZH_RETRIEVAL_QA_TRIGGER_KEYWORDS,
             rejection_answer=ZH_RETRIEVAL_QA_REJECTION_ANSWER,
         )
         print(f"Agent: {agent_response}")
@@ -140,7 +140,7 @@ def run(self, user_input: str, memory: ConversationBufferWithSummary) -> Tuple[s
         result = self.rag_chain.run(
             query=user_input,
             stop=[memory.human_prefix + ": "],
-            rejection_trigger_keywrods=ZH_RETRIEVAL_QA_TRIGGER_KEYWORDS,
+            rejection_trigger_keywords=ZH_RETRIEVAL_QA_TRIGGER_KEYWORDS,
             rejection_answer=ZH_RETRIEVAL_QA_REJECTION_ANSWER,
         )
         return result, memory

@@ -89,7 +89,7 @@ docker pull hpcaitech/colossalai-inference:v2
 docker run -it --gpus all --name ANY_NAME -v $PWD:/workspace -w /workspace hpcaitech/colossalai-inference:v2 /bin/bash
 
 # enter into docker container
-cd /path/to/CollossalAI
+cd /path/to/ColossalAI
 pip install -e .
 
 ```

@@ -86,7 +86,7 @@ docker pull hpcaitech/colossalai-inference:v2
 docker run -it --gpus all --name ANY_NAME -v $PWD:/workspace -w /workspace hpcaitech/colossalai-inference:v2 /bin/bash
 
 # enter into docker container
-cd /path/to/CollossalAI
+cd /path/to/ColossalAI
 pip install -e .
 
 # install lightllm

@@ -46,7 +46,7 @@ class CaiInferEngine:
 
     model = LlamaForCausalLM.from_pretrained("your_path_to_model")
     tokenizer = LlamaTokenizer.from_pretrained("/home/lczyh/share/models/llama-7b-hf")
-    # assume the model is infered with 2 pipeline stages
+    # assume the model is inferred with 2 pipeline stages
     inferengine = CaiInferEngine(pp_size=2, model=model, model_policy=LlamaModelInferPolicy())
 
     input = ["Introduce a landmark in China ","Introduce a landmark in China "]
@@ -70,7 +70,7 @@ def __init__(
         max_input_len: int = 32,
         max_output_len: int = 32,
         verbose: bool = False,
-        # TODO: implement early_stopping, and various gerneration options
+        # TODO: implement early_stopping, and various generation options
         early_stopping: bool = False,
         do_sample: bool = False,
         num_beams: int = 1,

@@ -47,7 +47,7 @@ be optimized jointly to further speed up training.
 
 2. Model Accuracy
     - Communication Efficiency
-      - Reduce Volumn of Comm.
+      - Reduce Volume of Comm.
       - Reduce Frequency of Comm.
     - Memory Efficiency
       - Mix-Precision Training

@@ -164,7 +164,7 @@ def _gen_token_action(self, model: Module):
             self.timestamps[self.mb_manager.idx].append(time.time())
         assert (
             "logits" in logits
-        ), f"When first stage in GENERATE phase, the ouput should have attribute `logits`, but has {logits.keys()}"
+        ), f"When first stage in GENERATE phase, the output should have attribute `logits`, but has {logits.keys()}"
         new_token = self._get_token_id(logits["logits"])
 
         self.mb_manager.step(new_token)
@@ -401,7 +401,7 @@ def generate_step_broadcast(self, model: Module, data_iter: Iterable) -> Union[t
                             self.timestamps[self.mb_manager.idx].append(time.time())
                         assert (
                             "logits" in logits
-                        ), f"When first stage in GENERATE phase, the ouput should have attribute `logits`, but has {logits.keys()}"
+                        ), f"When first stage in GENERATE phase, the output should have attribute `logits`, but has {logits.keys()}"
                         new_token = self._get_token_id(logits["logits"])
                         self.mb_manager.step(new_token)
                         # If the current micro batch is not DONE, go through blocks

@@ -338,7 +338,7 @@ def count_flops_attn(model, _x, y):
 
 class QKVAttentionLegacy(nn.Module):
     """
-    A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping
+    A module which performs QKV attention. Matches legacy QKVAttention + input/output heads shaping
     """
 
     def __init__(self, n_heads):