add support for include_tokens_per_second in training args (#2269)

* add support for include_tokens_per_second in training args

* Update docs/config.qmd

Co-authored-by: NanoCode012 <nano@axolotl.ai>

* Update src/axolotl/core/trainer_builder.py

Co-authored-by: NanoCode012 <nano@axolotl.ai>

---------

Co-authored-by: NanoCode012 <nano@axolotl.ai>

src/axolotl/core/trainer_builder.py

@@ -330,6 +330,12 @@ class HFCausalTrainerBuilder(TrainerBuilderBase):
         )
 
         training_arguments_kwargs = {}
+
+        if self.cfg.include_tokens_per_second is not None:
+            training_arguments_kwargs[
+                "include_tokens_per_second"
+            ] = self.cfg.include_tokens_per_second
+
         if self.cfg.bf16 == "full":
             training_arguments_kwargs["bf16_full_eval"] = True
         else:

src/axolotl/utils/config/models/input/v0_4_1/__init__.py

@@ -844,6 +844,7 @@ class AxolotlInputConfig(
     save_only_model: Optional[bool] = False
     use_tensorboard: Optional[bool] = None
     profiler_steps: Optional[int] = None
+    include_tokens_per_second: Optional[bool] = None
 
     neftune_noise_alpha: Optional[float] = None