Add default LoRA target modules for Gemma (#3936)

ludwig/schema/model_types/utils.py

@@ -323,6 +323,11 @@ def set_llm_parameters(config: "ModelConfig") -> None:
     # PEFT PR: https://github.com/huggingface/peft/pull/1375
     _set_phi2_target_modules(config)
 
+    # HACK(Arnav): Set Gemma target modules when using LoRA
+    # GitHub issue: https://github.com/ludwig-ai/ludwig/issues/3937
+    # PEFT PR: https://github.com/huggingface/peft/pull/1499
+    _set_gemma_target_modules(config)
+
 
 def _set_llm_tokenizers(config: "ModelConfig") -> None:
     """Sets the tokenizers for the LLM model to the pretrained model name or path. This ensures that they use the
@@ -451,6 +456,22 @@ def _set_phi2_target_modules(config: "ModelConfig") -> None:
     config.adapter.target_modules = target_modules
 
 
+def _set_gemma_target_modules(config: "ModelConfig") -> None:
+    """If the base model is Gemma, LoRA is enabled and the target modules are not set, set the target modules to
+    maximize performance."""
+    if config.base_model not in {"google/gemma-2b", "google/gemma-2b-it", "google/gemma-7b", "google/gemma-7b-it"}:
+        return
+
+    if not config.adapter:
+        return
+
+    if config.adapter.type != "lora" or config.adapter.target_modules:
+        return
+
+    target_modules = ["q_proj", "v_proj"]
+    config.adapter.target_modules = target_modules
+
+
 @DeveloperAPI
 def contains_grid_search_parameters(hyperopt_config: HyperoptConfigDict) -> bool:
     """Returns True if any hyperopt parameter in the config is using the grid_search space."""