support no apllication of chat template for instructions

tests/test_sft_trainer.py

@@ -508,12 +508,12 @@ def test_sft_trainer_with_model(self):
                     packing=False,
                 )
 
-            # if we have input/output cols, then things should work without issue
-            dataset_with_input_output = self.dummy_dataset.rename_column("question", "input").rename_column("answer", "output")
+            # with instruction format, things should work without issue
+            dataset_with_prompt_collection = self.dummy_instruction_dataset
             trainer = SFTTrainer(
                 model=self.model,
                 args=training_args,
-                train_dataset=dataset_with_input_output,
+                train_dataset=dataset_with_prompt_collection,
                 dataset_text_field=None,
                 formatting_func=None,
                 max_seq_length=16,

trl/extras/dataset_formatting.py

@@ -58,7 +58,7 @@ def format_dataset(examples):
 
 
 def get_formatting_func_from_dataset(
-    dataset: Union[Dataset, ConstantLengthDataset], tokenizer: AutoTokenizer
+    dataset: Union[Dataset, ConstantLengthDataset], tokenizer: AutoTokenizer, apply_chat_instruction_template=True
 ) -> Optional[Callable]:
     r"""
     Finds the correct formatting function based on the dataset structure. Currently supported datasets are:
@@ -83,6 +83,7 @@ def get_formatting_func_from_dataset(
                 return conversations_formatting_function(tokenizer, "conversations")
         elif dataset.features == FORMAT_MAPPING["instruction"]:
             logging.info("Formatting dataset with instruction format")
-            return instructions_formatting_function(tokenizer)
+            if apply_chat_instruction_template:
+                return instructions_formatting_function(tokenizer)
 
     return None

trl/trainer/sft_trainer.py

@@ -145,6 +145,7 @@ def __init__(
         dataset_num_proc: Optional[int] = None,
         dataset_batch_size: int = 1000,
         neftune_noise_alpha: Optional[float] = None,
+        apply_chat_instruction_template = True,
         model_init_kwargs: Optional[Dict] = None,
         dataset_kwargs: Optional[Dict] = None,
     ):
@@ -250,15 +251,15 @@ def make_inputs_require_grad(module, input, output):
         if formatting_func is None and dataset_text_field is None:
             # check if dataset has ChatML format or instruction format and is supported
             # if not stays #None
-            formatting_func = get_formatting_func_from_dataset(train_dataset, tokenizer)
+            formatting_func = get_formatting_func_from_dataset(train_dataset, tokenizer, apply_chat_instruction_template)
 
-        requires_input_output_keys = False
+        requires_prompt_completion_keys = False
         if not packing:
-            requires_input_output_keys = (dataset_text_field is None and formatting_func is None)
+            requires_prompt_completion_keys = (dataset_text_field is None and formatting_func is None)
             if data_collator is None:
                 # Fall back to the appropriate collator type based on the input_output_keys
                 data_collator = (DataCollatorForSeq2Seq(tokenizer=tokenizer, padding=True)
-                if requires_input_output_keys
+                if requires_prompt_completion_keys
                 else DataCollatorForLanguageModeling(tokenizer=tokenizer, mlm=False))
 
         # Pre-process the datasets only once per node. The remaining processes will use the cache.
@@ -276,7 +277,7 @@ def make_inputs_require_grad(module, input, output):
                     num_of_sequences,
                     chars_per_token,
                     remove_unused_columns=args.remove_unused_columns if args is not None else True,
-                    requires_input_output_keys=requires_input_output_keys,
+                    requires_prompt_completion_keys=requires_prompt_completion_keys,
                     **dataset_kwargs,
                 )
             if eval_dataset is not None:
@@ -373,7 +374,7 @@ def _prepare_dataset(
         num_of_sequences,
         chars_per_token,
         remove_unused_columns=True,
-        requires_input_output_keys=False,
+        requires_prompt_completion_keys=False,
         append_concat_token=True,
         add_special_tokens=True,
     ):
@@ -393,7 +394,7 @@ def _prepare_dataset(
                 formatting_func,
                 add_special_tokens,
                 remove_unused_columns,
-                requires_input_output_keys,
+                requires_prompt_completion_keys,
             )
 
         else:
@@ -418,14 +419,13 @@ def _prepare_non_packed_dataloader(
         formatting_func=None,
         add_special_tokens=True,
         remove_unused_columns=True,
-        requires_input_output_keys=False,
+        requires_prompt_completion_keys=False,
     ):
         use_formatting_func = formatting_func is not None and dataset_text_field is None
         self._dataset_sanity_checked = False
 
-        # TODO : fix how EOS tokens are handled
         # Inspired from https://github.com/allenai/open-instruct/blob/main/open_instruct/finetune.py#L266
-        def tokenize_input_output(element):
+        def tokenize_prompt_completion(element):
 
             # It is difficult to add special tokens here, as separator / EOS tokens that may be added while tokenizing
             # input texts may differ from concatenated text, making masking on input length incorrect.
@@ -437,7 +437,7 @@ def tokenize_input_output(element):
                 )
 
             new_source = []
-            for (input_element, output_element) in zip(element['input'], element['output']):
+            for (input_element, output_element) in zip(element['prompt'], element['completion']):
                 if not input_element.endswith((' ', '\n', '\t')) and not output_element.startswith((' ', '\n', '\t')):
                     new_source.append(input_element + ' ' + output_element)
                 else:
@@ -491,15 +491,15 @@ def tokenize(element):
                 f"inspect dataset other columns (in this case {extra_columns}), you can subclass `DataCollatorForLanguageModeling` in case you used the default collator and create your own data collator in order to inspect the unused dataset columns."
             )
 
-        if requires_input_output_keys:
-            if "input" in dataset.column_names and "output" in dataset.column_names:
+        if requires_prompt_completion_keys:
+            if "prompt" in dataset.column_names and "completion" in dataset.column_names:
                 # TODO: if we execute this input path, it is expected that we are using a seq2seq
                 # collator. If that is the case, the tokenizer should had a pad_token; this is set
                 # to eos automatically if it's unset and no tokenizer is provided, but we should
                 # properly handle if a tokenizer with no padding token is given.
-                tokenize_func = tokenize_input_output
+                tokenize_func = tokenize_prompt_completion
             else:
-                raise KeyError("Missing input / output keys")
+                raise KeyError("Missing prompt / completion keys")
         else:
             tokenize_func = tokenize