Feat: Add support for tokenizer’s or custom jinja chat_template (#1970)

* Allow using tokenizer's default chat template with fallbacks

Summary of changes:

1. Adds `tokenizer_default` as option for `chat_template` in
   `chat_template` prompt strategy that allows using the chat template
   from tokenizer's config.json
2. Allows falling back to chat templates available in axolotl if
   tokenizer does not have a chat template
3. Adds a mistral chat template which supports system message - taken
   from https://github.com/chujiezheng/chat_templates/blob/main/chat_templates/mistral-instruct.jinja

---

Why?

Many popular models are not trained with chatml format. As a result for
the model to correctly learn chatml we have to turn on train_on_inputs
which requires more compute and time. If we can use the model's already
learned chat template we can just learn the output tokens

---

Todo:

- Write tests

* Add tests

* Fix lint and bug post merge from main

* Add option `chat_template_jinja` to provide a jinja template

* remove custom mistral template

* Address review comments and add docs

* Update docs/dataset-formats/conversation.qmd

Co-authored-by: NanoCode012 <kevinvong@rocketmail.com>

* fix: set default to tokenizer template

* Merge branch 'main' into cj_tokenizer_default_prompt_template

* chore: remove redundant function

* fix: re-arrange enum declaration position

* fix: refactor artifact left from main merge

* feat(doc): updated config with chat template options and clarified examples

* chore: clarify doc

* chore: added example for non-default template

* chore: refactor

* fix: test

* fix: config being dropped and unittest to catch that

* chore: lint

* chore: skip duplicate

* fix: rename var after merge

* feat: add test for levy's dpo case

* fix: remove default setting on edge case where chat template overriden in dataset section

* feat: handle sharegpt deprecation better in docs

* feat: add example using fallback

* feat: handles chat_template requiring specific user/assistant order

* fix: update test based on new defaults

* fix: imported name incorrectly updated on merge

* chore: lint

* fix: update dummy message to prevent potential overlap with real content

* fix(doc): formatting

* fix: update bradleyterry to use new chat_template

---------

Co-authored-by: Chirag Jain <jain.chirag925@gmail.com>

tests/prompt_strategies/test_dpo_chat_templates.py

@@ -86,6 +86,20 @@ def fixture_llama3_tokenizer():
     return tokenizer
 
 
+@pytest.fixture(name="phi3_tokenizer")
+def fixture_phi3_tokenizer():
+    tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-medium-128k-instruct")
+
+    return tokenizer
+
+
+@pytest.fixture(name="gemma_tokenizer")
+def fixture_gemma_tokenizer():
+    tokenizer = AutoTokenizer.from_pretrained("unsloth/gemma-2b-it", revision="703fb4a")
+
+    return tokenizer
+
+
 class TestAssistantDPOChatTemplateLlama3:
     """
     Test class for assistant style datasets with llama-3 prompts using the chat_template strategy.
@@ -99,7 +113,7 @@ class TestAssistantDPOChatTemplateLlama3:
                     "chat_template": "llama3",
                     "datasets": [
                         {
-                            "chat_template": "llama3",
+                            "type": "chat_template",
                         }
                     ],
                 }
@@ -124,7 +138,7 @@ class TestAssistantDPOChatTemplateLlama3:
                     "chat_template": "llama3",
                     "datasets": [
                         {
-                            "chat_template": "llama3",
+                            "type": "chat_template",
                             "field_messages": "conversation",
                             "field_chosen": "better",
                             "field_rejected": "worse",
@@ -152,5 +166,65 @@ class TestAssistantDPOChatTemplateLlama3:
         assert result["rejected"] == "party on<|eot_id|>"
 
 
+class TestAssistantDPOChatTemplatePhi3:
+    """
+    Test class for assistant style datasets with phi-3 prompts using the tokenizer's chat_template strategy.
+    """
+
+    def test_phi3_defaults(self, phi3_tokenizer, assistant_dataset):
+        # pylint: disable=duplicate-code
+        transform_fn = default(
+            DictDefault(
+                {
+                    "chat_template": "tokenizer_default",
+                    "datasets": [
+                        {
+                            "type": "chat_template",
+                        }
+                    ],
+                }
+            )
+        )
+        result = transform_fn(assistant_dataset[0], tokenizer=phi3_tokenizer)
+        assert result["prompt"] == (
+            "<|user|>\nhello<|end|>\n"
+            + "<|assistant|>\nhello<|end|>\n"
+            + "<|user|>\ngoodbye<|end|>\n"
+            + "<|assistant|>\n"
+        )
+        assert result["chosen"] == "goodbye<|end|>"
+        assert result["rejected"] == "party on<|end|>"
+
+
+class TestAssistantDPOChatTemplateGemma:
+    """
+    Test class for assistant style datasets with gemma prompts using the tokenizer's chat_template strategy.
+    """
+
+    def test_gemma_defaults(self, gemma_tokenizer, assistant_dataset):
+        # pylint: disable=duplicate-code
+        transform_fn = default(
+            DictDefault(
+                {
+                    "chat_template": "tokenizer_default",
+                    "datasets": [
+                        {
+                            "type": "chat_template",
+                        }
+                    ],
+                }
+            )
+        )
+        result = transform_fn(assistant_dataset[0], tokenizer=gemma_tokenizer)
+        assert result["prompt"] == (
+            "<bos><start_of_turn>user\nhello<end_of_turn>\n"
+            + "<start_of_turn>model\nhello<end_of_turn>\n"
+            + "<start_of_turn>user\ngoodbye<end_of_turn>\n"
+            + "<start_of_turn>model\n"
+        )
+        assert result["chosen"] == "goodbye<end_of_turn>"
+        assert result["rejected"] == "party on<end_of_turn>"
+
+
 if __name__ == "__main__":
     unittest.main()