Merge pull request #4878 from ly863/main

Train the last turing conversation.

src/llamafactory/data/processors/supervised.py

@@ -70,7 +70,11 @@ def _encode_supervised_example(
             source_mask = [IGNORE_INDEX] * source_len
 
         input_ids += source_ids + target_ids
-        labels += source_mask + target_ids
+
+        if data_args.train_last_turn_only and turn_idx != len(encoded_pairs) - 1:
+            labels += source_mask + [IGNORE_INDEX] * len(target_ids)
+        else:
+            labels += source_mask + target_ids
 
     if template.efficient_eos:
         input_ids += [tokenizer.eos_token_id]

src/llamafactory/hparams/data_args.py

@@ -41,6 +41,10 @@ class DataArguments:
         default="data",
         metadata={"help": "Path to the folder containing the datasets."},
     )
+    train_last_turn_only: Optional[bool] = field(
+        default=False,
+        metadata={"help": "Whether or not to train the last turn only."},
+    )
     cutoff_len: int = field(
         default=1024,
         metadata={"help": "The cutoff length of the tokenized inputs in the dataset."},

src/llamafactory/hparams/parser.py

@@ -162,6 +162,9 @@ def get_train_args(args: Optional[Dict[str, Any]] = None) -> _TRAIN_CLS:
     # Check arguments
     if finetuning_args.stage != "pt" and data_args.template is None:
         raise ValueError("Please specify which `template` to use.")
+
+    if finetuning_args.stage == "pt" and data_args.train_last_turn_only:
+        raise ValueError("PT stage does not support `train_last_turn_only`.")
 
     if finetuning_args.stage != "sft" and training_args.predict_with_generate:
         raise ValueError("`predict_with_generate` cannot be set as True except SFT.")

src/llamafactory/webui/components/train.py

@@ -44,10 +44,11 @@ def create_train_tab(engine: "Engine") -> Dict[str, "Component"]:
         )
         dataset_dir = gr.Textbox(value=DEFAULT_DATA_DIR, scale=1)
         dataset = gr.Dropdown(multiselect=True, allow_custom_value=True, scale=4)
+        train_last_turn_only = gr.Checkbox()
         preview_elems = create_preview_box(dataset_dir, dataset)
 
-    input_elems.update({training_stage, dataset_dir, dataset})
-    elem_dict.update(dict(training_stage=training_stage, dataset_dir=dataset_dir, dataset=dataset, **preview_elems))
+    input_elems.update({training_stage, dataset_dir, dataset,train_last_turn_only})
+    elem_dict.update(dict(training_stage=training_stage, dataset_dir=dataset_dir, dataset=dataset,train_last_turn_only=train_last_turn_only, **preview_elems))
 
     with gr.Row():
         learning_rate = gr.Textbox(value="5e-5")

src/llamafactory/webui/locales.py

@@ -536,6 +536,20 @@
             "info": "更改分词器词表和嵌入层的大小。",
         },
     },
+    "train_last_turn_only": {
+        "en": {
+            "label": "Train last turn only",
+            "info": "Train the model with the last turn only in multi turn.",
+        },
+        "ru": {
+            "label": "Обучать только последний поворот",
+            "info": "Обучать модель только последним поворотом в многоповоротном диалоге.",
+        },
+        "zh": {
+            "label": "仅最后一轮参与训练",
+            "info": "多轮对话仅使用最后一轮计算loss。",
+        },
+    },
     "use_llama_pro": {
         "en": {
             "label": "Enable LLaMA Pro",

src/llamafactory/webui/runner.py

@@ -125,6 +125,7 @@ def _parse_train_args(self, data: Dict["Component", Any]) -> Dict[str, Any]:
             visual_inputs=get("top.visual_inputs"),
             dataset_dir=get("train.dataset_dir"),
             dataset=",".join(get("train.dataset")),
+            train_last_turn_only=get("train.train_last_turn_only"),
             cutoff_len=get("train.cutoff_len"),
             learning_rate=float(get("train.learning_rate")),
             num_train_epochs=float(get("train.num_train_epochs")),