Convert PaliGemma2

- Set version 2 as default in conversion and verifications
- Reduce parameters of fake model configs
- PaliGemma unittests are still crashing <50% times which has been improved from crashing 70% times before. Reducing some parameters seem helpful.
- Updated README.md accordingly

PiperOrigin-RevId: 708838151

ai_edge_torch/generative/examples/README.md

@@ -7,12 +7,14 @@ Gemma is Google's open-source LLM. The model has both a 2B and 7B versions. See
 ## PaliGemma
 PaliGemma is a multimodal LLM which gets images and text as input, then
 generates text as output. See
-[model's Kaggle page](https://www.kaggle.com/models/google/paligemma).
-The example we provide is PaliGemma 3B with 224 image size. Since Kaggle has
-only Jax-version of PaliGemma, PyTorch model can be download from
-[here](https://huggingface.co/google/paligemma-3b-mix-224/tree/main).
-
-Note that PaliGemma can be converted to TfLite only with [ODML Torch conversion
+[model's Kaggle page](https://www.kaggle.com/models/google/paligemma2).
+The examples we provide are PaliGemma2 and 1 of 3B with 224 image size.
+The checkpoint for PaliGemma2 can be downloaded from
+[here](https://www.kaggle.com/models/google/paligemma-2/transformers/paligemma2-3b-pt-224).
+Since Kaggle has only Jax-version of PaliGemma1, PyTorch model of PaliGemma1 can
+be download from [here](https://huggingface.co/google/paligemma-3b-mix-224/tree/main).
+
+Note that PaliGemma models can be converted to TfLite only with [ODML Torch conversion
 backend](https://github.com/google-ai-edge/ai-edge-torch/blob/main/docs/pytorch_converter/README.md#use-odml-torch-conversion-backend-experimental)
 
 ## Llama

ai_edge_torch/generative/examples/paligemma/convert_to_tflite.py

@@ -29,9 +29,15 @@
 from ai_edge_torch.generative.utilities.model_builder import ExportConfig
 import torch
 
+_VERSION = flags.DEFINE_enum(
+    'version',
+    '2',
+    ['1', '2'],
+    'The version of PaliGemma model to verify.',
+)
 _CHECKPOINT_PATH = flags.DEFINE_string(
     'checkpoint_path',
-    os.path.join(pathlib.Path.home(), 'Downloads/llm_data/paligemma-3b-224'),
+    os.path.join(pathlib.Path.home(), 'Downloads/llm_data/paligemma2-3b-224'),
     'The path to the model checkpoint, or directory holding the checkpoint.',
 )
 _TFLITE_PATH = flags.DEFINE_string(
@@ -63,10 +69,12 @@
 
 def main(_):
   pytorch_model = paligemma.build_model(
-      _CHECKPOINT_PATH.value, kv_cache_max_len=_KV_CACHE_MAX_LEN.value
+      _CHECKPOINT_PATH.value,
+      version=int(_VERSION.value),
+      kv_cache_max_len=_KV_CACHE_MAX_LEN.value,
   )
   quant_suffix = 'q8' if _QUANTIZE.value else 'f32'
-  output_filename = f'paligemma_{quant_suffix}_seq{_PREFILL_SEQ_LEN.value}_ekv{_KV_CACHE_MAX_LEN.value}.tflite'
+  output_filename = f'paligemma{_VERSION.value}_{quant_suffix}_seq{_PREFILL_SEQ_LEN.value}_ekv{_KV_CACHE_MAX_LEN.value}.tflite'
   converter.convert_to_tflite(
       pytorch_model,
       tflite_path=os.path.join(_TFLITE_PATH.value, output_filename),

ai_edge_torch/generative/examples/paligemma/decoder.py

@@ -137,6 +137,8 @@ def get_fake_decoder_config(kv_cache_max_len: int = 128) -> cfg.ModelConfig:
   config.vocab_size = 128
   config.num_layers = 2
   config.max_seq_len = 2 * kv_cache_max_len
+  config.embedding_dim = 128
+  config.embedding_scale = 128**0.5
   return config
 
 

ai_edge_torch/generative/examples/paligemma/decoder2.py

@@ -160,6 +160,8 @@ def get_fake_decoder2_config(kv_cache_max_len: int = 128) -> cfg.ModelConfig:
   config.vocab_size = 128
   config.num_layers = 2
   config.max_seq_len = 2 * kv_cache_max_len
+  config.embedding_dim = 128
+  config.embedding_scale = 128**0.5
   return config
 
 

ai_edge_torch/generative/examples/paligemma/paligemma.py

@@ -136,8 +136,8 @@ def get_fake_model_config(get_decoder_config, **kwargs) -> PaliGemmaConfig:
   return PaliGemmaConfig(
       image_encoder_config=image_encoder.get_fake_image_encoder_config(),
       decoder_config=get_decoder_config(**kwargs),
-      image_token_id=257152,
-      image_projection_scale=2048**0.5,
+      image_token_id=127,
+      image_projection_scale=128**0.5,
       image_projection_use_bias=True,
   )
 

ai_edge_torch/generative/examples/paligemma/verify.py

@@ -30,7 +30,7 @@
 
 _VERSION = flags.DEFINE_enum(
     "version",
-    "1",
+    "2",
     ["1", "2"],
     "The version of PaliGemma model to verify.",
 )

ai_edge_torch/generative/examples/paligemma/verify_image_encoder.py

@@ -28,7 +28,7 @@
 
 _VERSION = flags.DEFINE_enum(
     "version",
-    "1",
+    "2",
     ["1", "2"],
     "The version of PaliGemma vision model to verify.",
 )

ai_edge_torch/generative/test/test_model_conversion_large.py

@@ -21,6 +21,8 @@
 from ai_edge_torch.generative.examples.gemma import gemma2
 from ai_edge_torch.generative.examples.llama import llama
 from ai_edge_torch.generative.examples.openelm import openelm
+from ai_edge_torch.generative.examples.paligemma import decoder
+from ai_edge_torch.generative.examples.paligemma import decoder2
 from ai_edge_torch.generative.examples.paligemma import paligemma
 from ai_edge_torch.generative.examples.phi import phi2
 from ai_edge_torch.generative.examples.phi import phi3
@@ -171,13 +173,9 @@ def test_amd_llama_135m(self):
     pytorch_model = amd_llama_135m.AmdLlama(config).eval()
     self._test_model(config, pytorch_model, "prefill", atol=1e-5, rtol=1e-5)
 
-  @googletest.skipIf(
-      ai_edge_torch.config.in_oss,
-      reason="tests with custom ops are not supported in oss",
-  )
-  def disabled_test_paligemma(self):
-    config = paligemma.get_fake_model_config()
-    pytorch_model = paligemma.PaliGemma(config).eval()
+  def _test_paligemma_model(self, decoder_class, decoder_config, atol, rtol):
+    config = paligemma.get_fake_model_config(decoder_config)
+    pytorch_model = paligemma.PaliGemma(config, decoder_class).eval()
 
     image_embedding_config = config.image_encoder_config.image_embedding
     num_patches = (
@@ -215,11 +213,32 @@ def disabled_test_paligemma(self):
             kv,
             pixel_values=pixel_values,
             signature_name="prefill_pixel",
-            atol=1e-3,
-            rtol=1e-5,
+            atol=atol,
+            rtol=rtol,
         )
     )
 
+  @googletest.skipIf(
+      ai_edge_torch.config.in_oss,
+      reason="tests with custom ops are not supported in oss",
+  )
+  def disabled_test_paligemma1(self):
+    self._test_paligemma_model(
+        decoder.Decoder, decoder.get_fake_decoder_config, atol=1e-3, rtol=1e-5
+    )
+
+  @googletest.skipIf(
+      ai_edge_torch.config.in_oss,
+      reason="tests with custom ops are not supported in oss",
+  )
+  def disabled_test_paligemma2(self):
+    self._test_paligemma_model(
+        decoder2.Decoder2,
+        decoder2.get_fake_decoder2_config,
+        atol=1e-3,
+        rtol=1e-5,
+    )
+
   @googletest.skipIf(
       ai_edge_torch.config.in_oss,
       reason="tests with custom ops are not supported in oss",