fix: multiple typos of different value (#782)

* Update quicktour.mdx

* Update peft_model.yaml

* Update evaluation_tracker.py

* Update bert_scorer.py

Author: DeVikingMark

docs/source/quicktour.mdx

@@ -179,6 +179,6 @@ To evaluate a model trained with nanotron on a single gpu.
  ```
 
 The `nproc-per-node` argument should match the data, tensor and pipeline
-parallelism confidured in the `lighteval_config_template.yaml` file.
+parallelism configured in the `lighteval_config_template.yaml` file.
 That is: `nproc-per-node = data_parallelism * tensor_parallelism *
 pipeline_parallelism`.

examples/model_configs/peft_model.yaml

@@ -1,6 +1,6 @@
 model_parameters:
   model_name: "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B" # pretrained=model_name,trust_remote_code=boolean,revision=revision_to_use,model_parallel=True ... For a PEFT model, the pretrained model should be the one trained with PEFT and the base model below will contain the original model on which the adapters will be applied.
-  tokenizer: null # name of tokenier to use if defferent from the model's default
+  tokenizer: null # name of tokenizer to use if different from the model's default
   subfolder: null # subfolder in the model's directory to use
   dtype: "float16"  # Specifying the model to be loaded in 4 bit uses BitsAndBytesConfig. The other option is to use "8bit" quantization.
   compile: true

src/lighteval/logging/evaluation_tracker.py

@@ -579,7 +579,7 @@ def recreate_metadata_card(self, repo_id: str) -> None:  # noqa: C901
         new_dictionary.update(results_dict)
         results_string = json.dumps(new_dictionary, indent=4)
 
-        # If we are pushing to the Oppen LLM Leaderboard, we'll store specific data in the model card.
+        # If we are pushing to the Open LLM Leaderboard, we'll store specific data in the model card.
         is_open_llm_leaderboard = repo_id.split("/")[0] == "open-llm-leaderboard"
         if is_open_llm_leaderboard:
             org_string = (
@@ -695,7 +695,7 @@ def push_to_tensorboard(  # noqa: C901
         # We are doing parallel evaluations of multiple checkpoints and recording the steps not in order
         # This messes up with tensorboard, so the easiest is to rename files in the order of the checkpoints
         # See: https://github.com/tensorflow/tensorboard/issues/5958
-        # But tensorboardX don't let us control the prefix of the files (only the suffix), so we need to do it ourselves before commiting the files
+        # But tensorboardX don't let us control the prefix of the files (only the suffix), so we need to do it ourselves before committing the files
 
         # tb_context.close()  # flushes the unfinished write operations
         time.sleep(5)

src/lighteval/metrics/imports/bert_scorer.py

@@ -74,7 +74,7 @@ def bert_encode(model, x, attention_mask, all_layers=False):
 
 def collate_idf(arr, tokenizer, idf_dict, device="cuda:0"):
     """
-    Helper function that pads a list of sentences to hvae the same length and
+    Helper function that pads a list of sentences to have the same length and
     loads idf score for words in the sentences.
 
     Args:
@@ -161,20 +161,20 @@ def greedy_cos_idf(
     Args:
         - :param: `ref_embedding` (torch.Tensor):
                    embeddings of reference sentences, BxKxd,
-                   B: batch size, K: longest length, d: bert dimenison
+                   B: batch size, K: longest length, d: bert dimension
         - :param: `ref_lens` (list of int): list of reference sentence length.
         - :param: `ref_masks` (torch.LongTensor): BxKxK, BERT attention mask for
                    reference sentences.
         - :param: `ref_idf` (torch.Tensor): BxK, idf score of each word
                    piece in the reference sentence
         - :param: `hyp_embedding` (torch.Tensor):
                    embeddings of candidate sentences, BxKxd,
-                   B: batch size, K: longest length, d: bert dimenison
+                   B: batch size, K: longest length, d: bert dimension
         - :param: `hyp_lens` (list of int): list of candidate sentence length.
         - :param: `hyp_masks` (torch.LongTensor): BxKxK, BERT attention mask for
                    candidate sentences.
         - :param: `hyp_idf` (torch.Tensor): BxK, idf score of each word
-                   piece in the candidate setence
+                   piece in the candidate sentence
     """
     ref_embedding.div_(torch.norm(ref_embedding, dim=-1).unsqueeze(-1))
     hyp_embedding.div_(torch.norm(hyp_embedding, dim=-1).unsqueeze(-1))
@@ -255,7 +255,7 @@ def bert_cos_score_idf(
         - :param: `model` : a BERT model in `pytorch_pretrained_bert`
         - :param: `refs` (list of str): reference sentences
         - :param: `hyps` (list of str): candidate sentences
-        - :param: `tokenzier` : a BERT tokenizer corresponds to `model`
+        - :param: `tokenizer` : a BERT tokenizer corresponds to `model`
         - :param: `idf_dict` : a dictionary mapping a word piece index to its
                                inverse document frequency
         - :param: `verbose` (bool): turn on intermediate status update
@@ -339,7 +339,7 @@ def __init__(
     ):
         """
         Args:
-            - :param: `model_type` (str): contexual embedding model specification, default using the suggested
+            - :param: `model_type` (str): contextual embedding model specification, default using the suggested
                       model for the target langauge; has to specify at least one of
                       `model_type` or `lang`
             - :param: `num_layers` (int): the layer of representation to use.