Support for Falcon2-11B (#30771)

* remove unrelated changes

* remove unrelated changes on phi and stable LM

* add: Test for Falcon 10B

* fix: formatting

* fix: loading the falcon 10B in 8 bit precision using bitsanbytes.

* fix: device placement

* fix: broken tests.

* fix: backwards compatibility for falcon 1B architecture.

* chore: updated test.

* chore: test_modeling_falcon.py to use the 11B model.

* chore: minor edit

* chore: formating.

---------

Co-authored-by: Pablo Montalvo <39954772+molbap@users.noreply.github.com>
Co-authored-by: ArthurZucker <arthur.zucker@gmail.com>

tests/models/falcon/test_modeling_falcon.py

@@ -602,6 +602,25 @@ class FalconLanguageGenerationTest(unittest.TestCase):
 
         self.assertEqual(output_str, EXPECTED_OUTPUT)
 
+    @slow
+    @require_bitsandbytes
+    def test_lm_generate_falcon_11b(self):
+        tokenizer = AutoTokenizer.from_pretrained("tiiuae/falcon-11B", padding_side="left")
+        model = FalconForCausalLM.from_pretrained(
+            "tiiuae/falcon-11B", device_map={"": torch_device}, load_in_8bit=True
+        )
+        model.eval()
+        inputs = tokenizer(
+            "Two roads diverged in a yellow wood,", return_tensors="pt", return_token_type_ids=False
+        ).to(torch_device)
+
+        EXPECTED_OUTPUT = "Two roads diverged in a yellow wood,\nAnd sorry I could not travel both\n"
+
+        output_ids = model.generate(**inputs, do_sample=False, max_new_tokens=9)
+        output_str = tokenizer.batch_decode(output_ids)[0]
+
+        self.assertEqual(output_str, EXPECTED_OUTPUT)
+
     @slow
     def test_lm_generation_big_models(self):
         # The big models are way too big for the CI, so we use tiny random models that resemble their
@@ -647,7 +666,7 @@ class FalconLanguageGenerationTest(unittest.TestCase):
         tokenizer.pad_token = tokenizer.eos_token
         model = AutoModelForCausalLM.from_pretrained(
             "tiiuae/falcon-7b",
-            device_map="auto",
+            device_map={"": torch_device},
             load_in_4bit=True,
         )