fix typos (#11041)

Author: co63oc

llm/tools/preprocess/create_pretraining_data.py

@@ -176,7 +176,7 @@ def get_whole_word_mask_tokens(tokens, words, max_word_length=6):
             i += 1
             continue
 
-        # add "##" mark on the middel tokens of Chinese words
+        # add "##" mark on the middle tokens of Chinese words
         # such as ["通过", "利用"] -> ["通", "##过"， "利", "##用"]
         has_add = False
         for length in range(max_word_length, 0, -1):

llm/utils/fused_layers.py

@@ -106,11 +106,11 @@ def sp_async_reducesctter(x_grad):
 def sync_mp_allreduce(task, dist_tensor):
     mp_placement_index = dist_tensor.process_mesh.dim_names.index("mp")
     new_placments = list()
-    for idx, placment in enumerate(dist_tensor.placements):
+    for idx, placement in enumerate(dist_tensor.placements):
         if idx == mp_placement_index:
             new_placments.append(dist.Replicate())
         else:
-            new_placments.append(placment)
+            new_placments.append(placement)
     place = paddle.framework._current_expected_place()
     place = paddle.framework._get_paddle_place(place)
 

llm/utils/sp_async_reduce_scatter.py

@@ -172,7 +172,7 @@ def forward_pre_hook(layer, input):
     ipp = id2ipp[id(layer)]
 
 
-def forward_post_hook(layer, input, ouput):
+def forward_post_hook(layer, input, output):
     paddle.nn.functional.linear = paddle_nn_functional_linear
     if is_fused_matmul_bias_supported():
         paddle.incubate.nn.functional.fused_linear = paddle_incubate_nn_functional_fused_linear

paddlenlp/transformers/clipseg/modeling.py

@@ -340,7 +340,7 @@ def forward(
         attn_weights = nn.functional.softmax(attn_weights, axis=-1)
 
         if output_attentions:
-            # this operation is a bit akward, but it's required to
+            # this operation is a bit awkward, but it's required to
             # make sure that attn_weights keeps its gradient.
             # In order to do so, attn_weights have to reshaped
             # twice and have to be reused in the following

paddlenlp/transformers/llama/modeling_auto.py

@@ -1146,7 +1146,7 @@ def forward(
             inputs_embeds = paddle.transpose(inputs_embeds, [1, 0, 2])
 
         if self.config.context_parallel_degree > 1 and (attention_mask is not None or self.config.alibi):
-            raise NotImplementedError("Ring FlashAttention dosen't support attention_mask or alibi")
+            raise NotImplementedError("Ring FlashAttention doesn't support attention_mask or alibi")
 
         global_mesh = global_mesh_starts_with_pp()
         if position_ids is None and self.config.sep_parallel_degree > 1:

paddlenlp/transformers/model_utils.py

@@ -2267,7 +2267,7 @@ def _fuse_or_split_keys(
                             post_quantize=post_quantize,
                         )
                     if post_quantize:
-                        # Split -> quantize(Not support mdoel save)
+                        # Split -> quantize(Not support model save)
                         state_dict = load_state_dict(
                             shard_file,
                             tp_actions if pre_tensor_parallel_split else None,
@@ -2280,7 +2280,7 @@ def _fuse_or_split_keys(
                             dtype,
                         )
                     else:
-                        # quantize -> split(Support mdoel save)
+                        # quantize -> split(Support model save)
                         state_dict = load_state_dict(
                             shard_file,
                             tp_actions if pre_tensor_parallel_split else None,

paddlenlp/utils/downloader.py

@@ -257,7 +257,7 @@ def _decompress(fname):
 
     # For protecting decompressing interrupted,
     # decompress to fpath_tmp directory firstly, if decompress
-    # successed, move decompress files to fpath and delete
+    # succeeded, move decompress files to fpath and delete
     # fpath_tmp and remove download compress file.
 
     if tarfile.is_tarfile(fname):

paddlenlp/utils/pdc_sdk.py

@@ -534,7 +534,7 @@ def pdc_backup_to_flash_device(self, persistent_path: str, flash_device_path: st
         # step 2: copy persistent data to flash device
         try:
             copy_tree(persistent_path, flash_device_path)
-            logger.info(f"backup {persistent_path} to {flash_device_path} successed.")
+            logger.info(f"backup {persistent_path} to {flash_device_path} succeeded.")
         except Exception as e:
             logger.error(f"[Error] [pdc_sdk] copy tree {persistent_path} to {flash_device_path} failed, error: {e}")
             self._pdc_backup_failed_directory(flash_device_path)

slm/applications/question_answering/unsupervised_qa/tools/dev_qq_pair_creation.py

@@ -36,7 +36,7 @@ def extract_q_from_json_file(json_file, out_file=None, test_sample_num=None, que
         if out_file:
             wf = open(os.path.join(out_file), "w", encoding="utf-8")
         if query_answer_path:
-            qeury_answer_wf = open(query_answer_path, "w", encoding="utf-8")
+            query_answer_wf = open(query_answer_path, "w", encoding="utf-8")
         q_list = []
         for i, json_line in enumerate(rf.readlines()):
             line_dict = json.loads(json_line)
@@ -47,7 +47,7 @@ def extract_q_from_json_file(json_file, out_file=None, test_sample_num=None, que
             answer = line_dict["answer"]
             if not test_sample_num or i < test_sample_num:
                 if query_answer_path:
-                    qeury_answer_wf.write(
+                    query_answer_wf.write(
                         question.replace("\n", " ").replace("\t", " ").strip()
                         + "\t"
                         + answer.replace("\n", " ").replace("\t", " ").strip()
@@ -59,9 +59,9 @@ def extract_q_from_json_file(json_file, out_file=None, test_sample_num=None, que
             else:
                 break
         if query_answer_path:
-            qeury_answer_wf.close()
+            query_answer_wf.close()
         if out_file:
-            wf.colse()
+            wf.close()
         return q_list
 
 

slm/applications/text_classification/hierarchical/analysis/word_interpret.ipynb

@@ -197,7 +197,7 @@
      "name": "stdout",
      "output_type": "stream",
      "text": [
-      "Start token level interpretion, it will take some time...\n",
+      "Start token level interpretation, it will take some time...\n",
       "Building prefix dict from the default dictionary ...\n",
       "Loading model from cache /tmp/jieba.cache\n",
       "Loading model cost 0.746 seconds.\n",
@@ -219,7 +219,7 @@
     "    interpreter = GradShapInterpreter(model)\n",
     "\n",
     "# Use interpreter to get the importance scores for all data\n",
-    "print(\"Start token level interpretion, it will take some time...\")\n",
+    "print(\"Start token level interpretation, it will take some time...\")\n",
     "analysis_result = []\n",
     "for batch in interpret_data_loader:\n",
     "    analysis_result += interpreter(tuple(batch))\n",