add basic support for the optimi adamw optimizer (#1727)

* add support for optimi_adamw optimizer w kahan summation

* pydantic validator for optimi_adamw

* workaround for setting optimizer for fsdp

* make sure to install optimizer packages

* make sure to have parity for model parameters passed to optimizer

* add smoke test for optimi_adamw optimizer

* don't use foreach optimi by default

tests/e2e/test_lora_llama.py

@@ -34,8 +34,8 @@ class TestLoraLlama(unittest.TestCase):
                 "sequence_len": 1024,
                 "load_in_8bit": True,
                 "adapter": "lora",
-                "lora_r": 32,
-                "lora_alpha": 64,
+                "lora_r": 8,
+                "lora_alpha": 16,
                 "lora_dropout": 0.05,
                 "lora_target_linear": True,
                 "val_set_size": 0.1,
@@ -50,7 +50,7 @@ class TestLoraLlama(unittest.TestCase):
                         "type": "alpaca",
                     },
                 ],
-                "num_epochs": 2,
+                "num_epochs": 1,
                 "micro_batch_size": 8,
                 "gradient_accumulation_steps": 1,
                 "output_dir": temp_dir,