【开源之夏】add dpo trainer and support dpo training for mindnlp (#1701)

mindspore-lab · Oct 31, 2024 · d250d79 · d250d79
1 parent 47f2de8
commit d250d79
Show file tree

Hide file tree

Showing 14 changed files with 6,322 additions and 0 deletions.
diff --git a/mindnlp/trl/__init__.py b/mindnlp/trl/__init__.py
@@ -13,3 +13,11 @@
 # limitations under the License.
 # ============================================================================
 """MindNLP Transformer Reinforcement Learning."""
+
+from .trainer import (
+    DPOTrainer,
+    DPOConfig,
+    FDivergenceType,
+    _build_tokenized_answer,
+    _truncate_tokens
+)
diff --git a/mindnlp/trl/import_utils.py b/mindnlp/trl/import_utils.py
diff --git a/mindnlp/trl/models/__init__.py b/mindnlp/trl/models/__init__.py
@@ -0,0 +1,5 @@
+"""trl model __init__"""
+from .modeling_base import (
+    PreTrainedModelWrapper,
+    create_reference_model
+)
diff --git a/mindnlp/trl/models/modeling_base.py b/mindnlp/trl/models/modeling_base.py