Falcon embeddings (#1149) [skip docker]

* also fix multipack for falcon and add smoke tests

* make sure to handle special tokens and added tokens for lora

* fix reference to model_type

* fix tests for falcon

* fix stray typo

* fixes for smoke tests

tests/e2e/patched/test_mixtral_samplepack.py

@@ -32,6 +32,7 @@ class TestMixtral(unittest.TestCase):
                 "base_model": "hf-internal-testing/Mixtral-tiny",
                 "tokenizer_config": "mistralai/Mixtral-8x7B-v0.1",
                 "flash_attention": True,
+                "sample_packing": True,
                 "sequence_len": 2048,
                 "load_in_4bit": True,
                 "adapter": "qlora",
@@ -57,7 +58,6 @@ class TestMixtral(unittest.TestCase):
                 "max_steps": 20,
                 "save_steps": 10,
                 "eval_steps": 10,
-                "sample_packing": True,
                 "bf16": "auto",
             }
         )
@@ -76,6 +76,7 @@ class TestMixtral(unittest.TestCase):
                 "base_model": "hf-internal-testing/Mixtral-tiny",
                 "tokenizer_config": "mistralai/Mixtral-8x7B-v0.1",
                 "flash_attention": True,
+                "sample_packing": True,
                 "sequence_len": 2048,
                 "val_set_size": 0.1,
                 "special_tokens": {},
@@ -95,7 +96,6 @@ class TestMixtral(unittest.TestCase):
                 "max_steps": 20,
                 "save_steps": 10,
                 "eval_steps": 10,
-                "sample_packing": True,
                 "bf16": "auto",
             }
         )