[Port] TensorFlow implementation of Mistral

docs/source/en/index.md

@@ -200,7 +200,7 @@ Flax), PyTorch, and/or TensorFlow.
 |                 [Megatron-BERT](model_doc/megatron-bert)                 |       ✅        |         ❌         |      ❌      |
 |                 [Megatron-GPT2](model_doc/megatron_gpt2)                 |       ✅        |         ✅         |      ✅      |
 |                       [MGP-STR](model_doc/mgp-str)                       |       ✅        |         ❌         |      ❌      |
-|                       [Mistral](model_doc/mistral)                       |       ✅        |         ❌         |      ✅      |
+|                       [Mistral](model_doc/mistral)                       |       ✅        |         ✅         |      ✅      |
 |                       [Mixtral](model_doc/mixtral)                       |       ✅        |         ❌         |      ❌      |
 |                         [mLUKE](model_doc/mluke)                         |       ✅        |         ❌         |      ❌      |
 |                           [MMS](model_doc/mms)                           |       ✅        |         ✅         |      ✅      |

docs/source/en/model_doc/mistral.md

@@ -216,4 +216,19 @@ A list of official Hugging Face and community (indicated by 🌎) resources to h
 ## FlaxMistralForCausalLM
 
 [[autodoc]] FlaxMistralForCausalLM
-    - __call__
+    - __call__
+
+## TFMistralModel
+
+[[autodoc]] TFMistralModel
+    - call
+
+## TFMistralForCausalLM
+
+[[autodoc]] TFMistralForCausalLM
+    - call
+
+## TFMistralForSequenceClassification
+
+[[autodoc]] TFMistralForSequenceClassification
+    - call

src/transformers/__init__.py

@@ -3964,6 +3964,9 @@
     _import_structure["models.mbart"].extend(
         ["TFMBartForConditionalGeneration", "TFMBartModel", "TFMBartPreTrainedModel"]
     )
+    _import_structure["models.mistral"].extend(
+        ["TFMistralForCausalLM", "TFMistralForSequenceClassification", "TFMistralModel", "TFMistralPreTrainedModel"]
+    )
     _import_structure["models.mobilebert"].extend(
         [
             "TFMobileBertForMaskedLM",
@@ -8047,6 +8050,12 @@
             TFMBartModel,
             TFMBartPreTrainedModel,
         )
+        from .models.mistral import (
+            TFMistralForCausalLM,
+            TFMistralForSequenceClassification,
+            TFMistralModel,
+            TFMistralPreTrainedModel,
+        )
         from .models.mobilebert import (
             TFMobileBertForMaskedLM,
             TFMobileBertForMultipleChoice,

src/transformers/models/auto/modeling_tf_auto.py

@@ -65,6 +65,7 @@
         ("lxmert", "TFLxmertModel"),
         ("marian", "TFMarianModel"),
         ("mbart", "TFMBartModel"),
+        ("mistral", "TFMistralModel"),
         ("mobilebert", "TFMobileBertModel"),
         ("mobilevit", "TFMobileViTModel"),
         ("mpnet", "TFMPNetModel"),
@@ -178,6 +179,7 @@
         ("gpt-sw3", "TFGPT2LMHeadModel"),
         ("gpt2", "TFGPT2LMHeadModel"),
         ("gptj", "TFGPTJForCausalLM"),
+        ("mistral", "TFMistralForCausalLM"),
         ("openai-gpt", "TFOpenAIGPTLMHeadModel"),
         ("opt", "TFOPTForCausalLM"),
         ("rembert", "TFRemBertForCausalLM"),
@@ -320,6 +322,7 @@
         ("layoutlm", "TFLayoutLMForSequenceClassification"),
         ("layoutlmv3", "TFLayoutLMv3ForSequenceClassification"),
         ("longformer", "TFLongformerForSequenceClassification"),
+        ("mistral", "TFMistralForSequenceClassification"),
         ("mobilebert", "TFMobileBertForSequenceClassification"),
         ("mpnet", "TFMPNetForSequenceClassification"),
         ("openai-gpt", "TFOpenAIGPTForSequenceClassification"),

src/transformers/models/mistral/__init__.py

@@ -13,7 +13,13 @@
 # limitations under the License.
 from typing import TYPE_CHECKING
 
-from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_flax_available, is_torch_available
+from ...utils import (
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    is_flax_available,
+    is_tf_available,
+    is_torch_available,
+)
 
 
 _import_structure = {
@@ -47,6 +53,19 @@
         "FlaxMistralPreTrainedModel",
     ]
 
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_tf_mistral"] = [
+        "TFMistralModel",
+        "TFMistralForCausalLM",
+        "TFMistralForSequenceClassification",
+        "TFMistralPreTrainedModel",
+    ]
+
 
 if TYPE_CHECKING:
     from .configuration_mistral import MistralConfig
@@ -77,6 +96,19 @@
             FlaxMistralPreTrainedModel,
         )
 
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_tf_mistral import (
+            TFMistralForCausalLM,
+            TFMistralForSequenceClassification,
+            TFMistralModel,
+            TFMistralPreTrainedModel,
+        )
+
 
 else:
     import sys