Merge branch 'main' into feature/huggingface#35425

bzantium · Feb 15, 2025 · 23fb756 · 23fb756
2 parents e0f1c2d + dd16acb
commit 23fb756
Show file tree

Hide file tree

Showing 17 changed files with 2,521 additions and 225 deletions.
diff --git a/docs/source/en/_toctree.yml b/docs/source/en/_toctree.yml
@@ -463,6 +463,8 @@
         title: Granite
       - local: model_doc/granitemoe
         title: GraniteMoe
+      - local: model_doc/granitemoeshared
+        title: GraniteMoeShared
       - local: model_doc/granitevision
         title: GraniteVision
       - local: model_doc/helium

diff --git a/docs/source/en/index.md b/docs/source/en/index.md
@@ -174,6 +174,7 @@ Flax), PyTorch, and/or TensorFlow.
 |               [GPTSAN-japanese](model_doc/gptsan-japanese)               |       ✅        |         ❌         |      ❌      |
 |                       [Granite](model_doc/granite)                       |       ✅        |         ❌         |      ❌      |
 |                  [GraniteMoeMoe](model_doc/granitemoe)                   |       ✅        |         ❌         |      ❌      |
+|            [GraniteMoeSharedMoe](model_doc/granitemoeshared)             |       ✅        |         ❌         |      ❌      |
 |                    [Graphormer](model_doc/graphormer)                    |       ✅        |         ❌         |      ❌      |
 |                [Grounding DINO](model_doc/grounding-dino)                |       ✅        |         ❌         |      ❌      |
 |                      [GroupViT](model_doc/groupvit)                      |       ✅        |         ✅         |      ❌      |

diff --git a/docs/source/en/model_doc/granitemoeshared.md b/docs/source/en/model_doc/granitemoeshared.md
@@ -0,0 +1,66 @@
+<!--Copyright 2025 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# GraniteMoeShared
+
+## Overview
+
+
+The GraniteMoe model was proposed in [Power Scheduler: A Batch Size and Token Number Agnostic Learning Rate Scheduler](https://arxiv.org/abs/2408.13359) by Yikang Shen, Matthew Stallone, Mayank Mishra, Gaoyuan Zhang, Shawn Tan, Aditya Prasad, Adriana Meza Soria, David D. Cox and Rameswar Panda.
+
+Additionally this class GraniteMoeSharedModel adds shared experts for Moe.
+
+```python
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+model_path = "ibm-research/moe-7b-1b-active-shared-experts"
+tokenizer = AutoTokenizer.from_pretrained(model_path)
+
+# drop device_map if running on CPU
+model = AutoModelForCausalLM.from_pretrained(model_path, device_map="auto")
+model.eval()
+
+# change input text as desired
+prompt = "Write a code to find the maximum value in a list of numbers."
+
+# tokenize the text
+input_tokens = tokenizer(prompt, return_tensors="pt")
+# generate output tokens
+output = model.generate(**input_tokens, max_new_tokens=100)
+# decode output tokens into text
+output = tokenizer.batch_decode(output)
+# loop over the batch to print, in this example the batch size is 1
+for i in output:
+    print(i)
+```
+
+This HF implementation is contributed by [Mayank Mishra](https://huggingface.co/mayank-mishra), [Shawn Tan](https://huggingface.co/shawntan) and [Sukriti Sharma](https://huggingface.co/SukritiSharma).
+
+
+## GraniteMoeSharedConfig
+
+[[autodoc]] GraniteMoeSharedConfig
+
+## GraniteMoeSharedModel
+
+[[autodoc]] GraniteMoeSharedModel
+    - forward
+
+## GraniteMoeSharedForCausalLM
+
+[[autodoc]] GraniteMoeSharedForCausalLM
+    - forward
diff --git a/docs/source/en/perf_infer_gpu_one.md b/docs/source/en/perf_infer_gpu_one.md
@@ -61,6 +61,7 @@ FlashAttention-2 is currently supported for the following architectures:
 * [GPT-J](https://huggingface.co/docs/transformers/model_doc/gptj#transformers.GPTJModel)
 * [Granite](https://huggingface.co/docs/transformers/model_doc/granite#transformers.GraniteModel)
 * [GraniteMoe](https://huggingface.co/docs/transformers/model_doc/granitemoe#transformers.GraniteMoeModel)
+* [GraniteMoeShared](https://huggingface.co/docs/transformers/model_doc/granitemoeshared#transformers.GraniteMoeSharedModel)
 * [Idefics2](https://huggingface.co/docs/transformers/model_doc/idefics2#transformers.Idefics2Model)
 * [Idefics3](https://huggingface.co/docs/transformers/model_doc/idefics3#transformers.Idefics3Model)
 * [Falcon](https://huggingface.co/docs/transformers/model_doc/falcon#transformers.FalconModel)
@@ -268,6 +269,7 @@ For now, Transformers supports SDPA inference and training for the following arc
 * [Idefics3](https://huggingface.co/docs/transformers/model_doc/idefics3#transformers.Idefics3Model)
 * [I-JEPA](https://huggingface.co/docs/transformers/model_doc/ijepa#transformers.IJepaModel)
 * [GraniteMoe](https://huggingface.co/docs/transformers/model_doc/granitemoe#transformers.GraniteMoeModel)
+* [GraniteMoeShared](https://huggingface.co/docs/transformers/model_doc/granitemoeshared#transformers.GraniteMoeSharedModel)
 * [JetMoe](https://huggingface.co/docs/transformers/model_doc/jetmoe#transformers.JetMoeModel)
 * [Jamba](https://huggingface.co/docs/transformers/model_doc/jamba#transformers.JambaModel)
 * [Llama](https://huggingface.co/docs/transformers/model_doc/llama#transformers.LlamaModel)

diff --git a/src/transformers/__init__.py b/src/transformers/__init__.py
@@ -497,6 +497,7 @@
     "models.gptj": ["GPTJConfig"],
     "models.granite": ["GraniteConfig"],
     "models.granitemoe": ["GraniteMoeConfig"],
+    "models.granitemoeshared": ["GraniteMoeSharedConfig"],
     "models.grounding_dino": [
         "GroundingDinoConfig",
         "GroundingDinoProcessor",
@@ -2548,6 +2549,14 @@
             "GraniteMoePreTrainedModel",
         ]
     )
+    _import_structure["models.granitemoeshared"].extend(
+        [
+            "GraniteMoeSharedForCausalLM",
+            "GraniteMoeSharedModel",
+            "GraniteMoeSharedPreTrainedModel",
+        ]
+    )
+
     _import_structure["models.grounding_dino"].extend(
         [
             "GroundingDinoForObjectDetection",
@@ -5617,6 +5626,7 @@
     from .models.gptj import GPTJConfig
     from .models.granite import GraniteConfig
     from .models.granitemoe import GraniteMoeConfig
+    from .models.granitemoeshared import GraniteMoeSharedConfig
     from .models.grounding_dino import (
         GroundingDinoConfig,
         GroundingDinoProcessor,
@@ -7497,6 +7507,11 @@
             GraniteMoeModel,
             GraniteMoePreTrainedModel,
         )
+        from .models.granitemoeshared import (
+            GraniteMoeSharedForCausalLM,
+            GraniteMoeSharedModel,
+            GraniteMoeSharedPreTrainedModel,
+        )
         from .models.grounding_dino import (
             GroundingDinoForObjectDetection,
             GroundingDinoModel,

diff --git a/src/transformers/models/__init__.py b/src/transformers/models/__init__.py
@@ -119,6 +119,7 @@
     gptj,
     granite,
     granitemoe,
+    granitemoeshared,
     grounding_dino,
     groupvit,
     helium,

diff --git a/src/transformers/models/auto/configuration_auto.py b/src/transformers/models/auto/configuration_auto.py
@@ -138,6 +138,7 @@
         ("gptsan-japanese", "GPTSanJapaneseConfig"),
         ("granite", "GraniteConfig"),
         ("granitemoe", "GraniteMoeConfig"),
+        ("granitemoeshared", "GraniteMoeSharedConfig"),
         ("granitevision", "LlavaNextConfig"),
         ("graphormer", "GraphormerConfig"),
         ("grounding-dino", "GroundingDinoConfig"),
@@ -469,6 +470,7 @@
         ("gptsan-japanese", "GPTSAN-japanese"),
         ("granite", "Granite"),
         ("granitemoe", "GraniteMoeMoe"),
+        ("granitemoeshared", "GraniteMoeSharedMoe"),
         ("granitevision", "LLaVA-NeXT"),
         ("graphormer", "Graphormer"),
         ("grounding-dino", "Grounding DINO"),

diff --git a/src/transformers/models/auto/modeling_auto.py b/src/transformers/models/auto/modeling_auto.py
@@ -133,6 +133,7 @@
         ("gptsan-japanese", "GPTSanJapaneseForConditionalGeneration"),
         ("granite", "GraniteModel"),
         ("granitemoe", "GraniteMoeModel"),
+        ("granitemoeshared", "GraniteMoeSharedModel"),
         ("graphormer", "GraphormerModel"),
         ("grounding-dino", "GroundingDinoModel"),
         ("groupvit", "GroupViTModel"),
@@ -528,6 +529,7 @@
         ("gptj", "GPTJForCausalLM"),
         ("granite", "GraniteForCausalLM"),
         ("granitemoe", "GraniteMoeForCausalLM"),
+        ("granitemoeshared", "GraniteMoeSharedForCausalLM"),
         ("helium", "HeliumForCausalLM"),
         ("jamba", "JambaForCausalLM"),
         ("jetmoe", "JetMoeForCausalLM"),

diff --git a/src/transformers/models/granitemoeshared/__init__.py b/src/transformers/models/granitemoeshared/__init__.py
@@ -0,0 +1,27 @@
+# Copyright 2024 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from typing import TYPE_CHECKING
+
+from ...utils import _LazyModule
+from ...utils.import_utils import define_import_structure
+
+
+if TYPE_CHECKING:
+    from .configuration_granitemoeshared import *
+    from .modeling_granitemoeshared import *
+else:
+    import sys
+
+    _file = globals()["__file__"]
+    sys.modules[__name__] = _LazyModule(__name__, _file, define_import_structure(_file), module_spec=__spec__)