Fix model integration ci (#26322)

* fix wav2vec2 * nit * stash * one more file to update * fix byt5 * vocab size is 256, don't change that! * use other revision * test persimon in smaller size * style * tests * nits * update add tokens from pretrained * test tokenization * nits * potential fnet fix? * more nits * nits * correct test * assert close * udpate * ouch * fix it * some more nits * FINALLU * use `adept` checkpoints * more adept checkpoints * that was invlved!
2023-10-02 13:55:46 +02:00
parent 6824461f2a
commit 63864e057f
9 changed files with 38 additions and 18 deletions
--- a/tests/models/fnet/test_modeling_fnet.py
+++ b/tests/models/fnet/test_modeling_fnet.py
@@ -532,8 +532,6 @@ class FNetModelIntegrationTest(unittest.TestCase):
    @slow
    @require_tokenizers
    def test_inference_long_sentence(self):
-        model = FNetForMaskedLM.from_pretrained("google/fnet-base")
-        model.to(torch_device)
        tokenizer = FNetTokenizerFast.from_pretrained("google/fnet-base")

        inputs = tokenizer(
@@ -543,8 +541,15 @@ class FNetModelIntegrationTest(unittest.TestCase):
            padding="max_length",
            max_length=512,
        )
+
+        # fmt: off
+        torch.testing.assert_allclose(inputs["input_ids"], torch.tensor([[4, 13, 283, 2479, 106, 8, 6, 845, 5, 168, 65, 367, 6, 845, 5, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3, 3, 3, 3, 3, 3, 3, 3, 3, 3,3]]))
+        # fmt: on
+
        inputs = {k: v.to(torch_device) for k, v in inputs.items()}

+        model = FNetForMaskedLM.from_pretrained("google/fnet-base")
+        model.to(torch_device)
        logits = model(**inputs).logits
        predictions_mask_1 = tokenizer.decode(logits[0, 6].topk(5).indices)
        predictions_mask_2 = tokenizer.decode(logits[0, 12].topk(5).indices)
--- a/tests/models/idefics/test_modeling_idefics.py
+++ b/tests/models/idefics/test_modeling_idefics.py
@@ -503,7 +503,11 @@ class IdeficsForVisionText2TextTest(IdeficsModelTest, unittest.TestCase):
 class IdeficsModelIntegrationTest(TestCasePlus):
    @cached_property
    def default_processor(self):
-        return IdeficsProcessor.from_pretrained("HuggingFaceM4/idefics-9b") if is_vision_available() else None
+        return (
+            IdeficsProcessor.from_pretrained("HuggingFaceM4/idefics-9b", revision="refs/pr/11")
+            if is_vision_available()
+            else None
+        )

    @require_bitsandbytes
    @slow
--- a/tests/models/instructblip/test_modeling_instructblip.py
+++ b/tests/models/instructblip/test_modeling_instructblip.py
@@ -29,7 +29,14 @@ from transformers import (
    InstructBlipQFormerConfig,
    InstructBlipVisionConfig,
 )
-from transformers.testing_utils import require_bitsandbytes, require_torch, require_vision, slow, torch_device
+from transformers.testing_utils import (
+    require_accelerate,
+    require_bitsandbytes,
+    require_torch,
+    require_vision,
+    slow,
+    torch_device,
+)
 from transformers.utils import is_torch_available, is_vision_available

 from ...test_configuration_common import ConfigTester
@@ -522,6 +529,7 @@ def prepare_img():
@slow
 class InstructBlipModelIntegrationTest(unittest.TestCase):
    @require_bitsandbytes
+    @require_accelerate
    def test_inference_vicuna_7b(self):
        processor = InstructBlipProcessor.from_pretrained("Salesforce/instructblip-vicuna-7b")
        model = InstructBlipForConditionalGeneration.from_pretrained(
--- a/tests/models/persimmon/test_modeling_persimmon.py
+++ b/tests/models/persimmon/test_modeling_persimmon.py
@@ -386,11 +386,13 @@ class PersimmonIntegrationTest(unittest.TestCase):
    @slow
    def test_model_8b_chat_logits(self):
        input_ids = [1, 306, 4658, 278, 6593, 310, 2834, 338]
-        model = PersimmonForCausalLM.from_pretrained("ArthurZ/persimmon-8b-chat", device_map="auto")
+        model = PersimmonForCausalLM.from_pretrained(
+            "adept/persimmon-8b-chat", device_map="auto", torch_dtype=torch.float16
+        )
        out = model(torch.tensor([input_ids])).logits

        EXPECTED_MEAN = torch.tensor(
-            [[-11.2879, -11.2628, -11.2498, -11.2534, -11.2676, -11.2638, -11.2501, -11.2431]], dtype=torch.float32
+            [[-11.2879, -11.2628, -11.2498, -11.2534, -11.2676, -11.2638, -11.2501, -11.2431]], dtype=torch.float16
        )
        torch.testing.assert_close(out.cpu().mean(-1), EXPECTED_MEAN, atol=1e-4, rtol=1e-4)
        # fmt: off
@@ -403,9 +405,11 @@ class PersimmonIntegrationTest(unittest.TestCase):
    def test_model_8b_chat_greedy_generation(self):
        EXPECTED_TEXT_COMPLETION = """human: Simply put, the theory of relativity states that?\n\nadept: The theory of relativity states that the laws of physics are the same for all observers, regardless of their relative motion."""
        prompt = "human: Simply put, the theory of relativity states that?\n\nadept:"
-        tokenizer = AutoTokenizer.from_pretrained("ArthurZ/persimmon-8b-chat", use_fast=False)
+        tokenizer = AutoTokenizer.from_pretrained("adept/persimmon-8b-chat", use_fast=False)
        input_ids = tokenizer.encode(prompt, return_tensors="pt").to(torch_device)
-        model = PersimmonForCausalLM.from_pretrained("ArthurZ/persimmon-8b-chat").to(torch_device)
+        model = PersimmonForCausalLM.from_pretrained("adept/persimmon-8b-chat", torch_dtype=torch.float16).to(
+            torch_device
+        )

        # greedy generation outputs
        generated_ids = model.generate(input_ids, max_new_tokens=64)