Merge pull request #23 from TianyiQ/main

fix(abstractions): data manipulation for dialogue generation

examples/abstractions/finetuning_datamanip.py

@@ -119,19 +119,29 @@ def dialogue_manipulation():
             }
         ]
     )
-    dialogue_data = llama8b_instruct.inference(
-        dialogue_data, "dialogue_data2", backend="sglang"
-    )
-    dialogue_data = dialogue_data.switch_role_to_user()
-    dialogue_data = llama8b_instruct.inference(
-        dialogue_data, "dialogue_data3", backend="sglang"
-    )
-    dialogue_data = dialogue_data.switch_role_to_assistant()
+
+    def converse():
+        nonlocal dialogue_data
+
+        dialogue_data = llama8b_instruct.inference(
+            dialogue_data, "dialogue_data2", backend="sglang"
+        )
+        dialogue_data = dialogue_data.switch_role_to_user()
+
+        dialogue_data = llama8b_instruct.inference(
+            dialogue_data, "dialogue_data3", backend="sglang"
+        )
+        dialogue_data = dialogue_data.switch_role_to_assistant()
+
+    for i in range(5):
+        converse()
+
     print(list(dialogue_data.all_passages()))
+    print(list(dialogue_data.to_openai_format()))
 
 
 if __name__ == "__main__":
-    # continue_pretrain()
-    # supervised_finetune()
-    # direct_preference_optimization()
+    continue_pretrain()
+    supervised_finetune()
+    direct_preference_optimization()
     dialogue_manipulation()

src/abstractions/backends.py

@@ -288,22 +288,23 @@ def vllm_process_batch(
                 temperature=temperature, top_p=0.95, max_tokens=max_tokens
             )
 
-            if not os.environ.get("ALLOW_EMPTY_INPUT") or not eval(
-                os.environ.get("ALLOW_EMPTY_INPUT")
+            if not os.environ.get("ALLOW_EMPTY_INSTRUCTION") or not eval(
+                os.environ.get("ALLOW_EMPTY_INSTRUCTION")
             ):
                 found = 0
                 for dic in sample_dicts:
-                    if not dic.get("input"):
+                    if not dic.get("instruction"):
                         if not found:
                             warnings.warn(
-                                'In at least one sample, "input" field is missing or empty. Content from the "instruction" field will be copied to the "input" field. This behavior can be disabled by ALLOW_EMPTY_INPUT=1.'
+                                'In at least one sample, "instruction" field is missing or empty. Content from the "input" field will be moved to the "instruction" field. This behavior can be disabled by ALLOW_EMPTY_INSTRUCTION=1.'
                             )
                             found = 1
-                        dic["input"] = dic["instruction"]
+                        dic["instruction"] = dic["input"]
+                        del dic["input"]
 
             prompts = [
                 fill_in_QA_template(
-                    dic["instruction"], dic["input"], model_repoid_or_path=template_type
+                    dic.get("instruction"), dic.get("input"), model_repoid_or_path=template_type
                 )
                 for dic in sample_dicts
             ]
@@ -518,18 +519,19 @@ def sglang_process_batch(
             """
             nonlocal purpose
 
-            if not os.environ.get("ALLOW_EMPTY_INPUT") or not eval(
-                os.environ.get("ALLOW_EMPTY_INPUT")
+            if not os.environ.get("ALLOW_EMPTY_INSTRUCTION") or not eval(
+                os.environ.get("ALLOW_EMPTY_INSTRUCTION")
             ):
                 found = 0
                 for dic in sample_dicts:
-                    if not dic.get("input"):
+                    if not dic.get("instruction"):
                         if not found:
                             warnings.warn(
-                                'In at least one sample, "input" field is missing or empty. Content from the "instruction" field will be copied to the "input" field. This behavior can be disabled by ALLOW_EMPTY_INPUT=1.'
+                                'In at least one sample, "instruction" field is missing or empty. Content from the "input" field will be moved to the "instruction" field. This behavior can be disabled by ALLOW_EMPTY_INSTRUCTION=1.'
                             )
                             found = 1
-                        dic["input"] = dic["instruction"]
+                        dic["instruction"] = dic["input"]
+                        del dic["input"]
 
             dialogues = dict_to_dialogue_list(sample_dicts, purpose)
             options_lists = [(dic["predict"] if "predict" in dic and isinstance(dic["predict"], list) else []) for dic in sample_dicts]
@@ -675,6 +677,8 @@ def dict_to_dialogue_list(
 
         if purpose == "logprobs" and "predict" in dic and isinstance(dic["predict"], str):
             res.append({"role": "assistant", "content": dic["predict"]})
+        elif "output" in dic:
+            res.append({"role": "assistant", "content": dic["output"]})
 
         return res
 

src/abstractions/data.py

@@ -15,9 +15,11 @@
 import os
 import json
 import warnings
+from functools import partial
 import src.utils.text_utils as tu
 from tqdm import tqdm
 from src.abstractions.configs.templates_configs import *
+from src.abstractions.backends import dict_to_dialogue_list
 
 
 # helper function, escape spaces in paths
@@ -184,6 +186,15 @@ def copy(self, data_name: str = None) -> "Data":
 
         cp.key_fields = self.key_fields.copy()
         return cp
+
+    def to_openai_format(self) -> Iterable[List[Dict[str, str]]]:
+        """
+        Convert the data to OpenAI format, where each dialogue is a list of dictionaries with string keys and string values.
+        Each dictionary represents a dialogue turn.
+        """
+        convert_fn: Callable[[Dict], List[Dict]] = partial(dict_to_dialogue_list, purpose="logprobs")
+        for element in self.all_passages():
+            yield convert_fn(element)
 
     def transform(
         self,
@@ -357,7 +368,7 @@ def switch_role_to_user_fn(sample_dict: Dict) -> Dict:
             all_histories = [h[i] for h in sample_dict.get("history", []) for i in range(2)]
             all_histories = [dialogue_starter] + all_histories
             assert len(all_histories) % 2 == 1
-            sample_dict["history"] = [[all_histories[i], all_histories[i + 1]] for i in range(len(all_histories)-1, 2)]
+            sample_dict["history"] = [[all_histories[i], all_histories[i + 1]] for i in range(0, len(all_histories)-1, 2)]
             sample_dict["instruction"] = all_histories[-1]
             sample_dict["system"] = user_system_prompt
             return sample_dict