From b9f8f863d9f0cb31e64f7cb5f1a11e1441cea9e5 Mon Sep 17 00:00:00 2001
From: eustlb <94853470+eustlb@users.noreply.github.com>
Date: Tue, 27 May 2025 17:03:55 +0200
Subject: [PATCH] [CSM] update model id (#38211)
* update model id
* codec_model eval
* add processor img
* use ungated repo for processor tests
---
docs/source/en/model_doc/csm.md | 14 +++++++++-----
src/transformers/models/csm/configuration_csm.py | 4 ++--
src/transformers/models/csm/generation_csm.py | 2 +-
src/transformers/models/csm/modeling_csm.py | 2 +-
src/transformers/models/csm/modular_csm.py | 2 +-
src/transformers/models/csm/processing_csm.py | 2 +-
tests/models/csm/test_modeling_csm.py | 2 +-
tests/models/csm/test_processor_csm.py | 3 +--
8 files changed, 17 insertions(+), 14 deletions(-)
diff --git a/docs/source/en/model_doc/csm.md b/docs/source/en/model_doc/csm.md
index 53c24a5eba..833ddb697b 100644
--- a/docs/source/en/model_doc/csm.md
+++ b/docs/source/en/model_doc/csm.md
@@ -39,7 +39,7 @@ CSM can be used to simply generate speech from a text prompt:
import torch
from transformers import CsmForConditionalGeneration, AutoProcessor
-model_id = "eustlb/csm-1b"
+model_id = "sesame/csm-1b"
device = "cuda" if torch.cuda.is_available() else "cpu"
# load the model and the processor
@@ -74,7 +74,7 @@ import torch
from transformers import CsmForConditionalGeneration, AutoProcessor
from datasets import load_dataset, Audio
-model_id = "eustlb/csm-1b"
+model_id = "sesame/csm-1b"
device = "cuda" if torch.cuda.is_available() else "cpu"
# load the model and the processor
@@ -119,7 +119,7 @@ import torch
from transformers import CsmForConditionalGeneration, AutoProcessor
from datasets import load_dataset, Audio
-model_id = "eustlb/csm-1b"
+model_id = "sesame/csm-1b"
device = "cuda" if torch.cuda.is_available() else "cpu"
# load the model and the processor
@@ -176,7 +176,7 @@ import copy
from transformers import CsmForConditionalGeneration, AutoProcessor
from datasets import load_dataset
-model_id = "eustlb/csm-1b"
+model_id = "sesame/csm-1b"
device = "cuda"
# set logs to ensure no recompilation and graph breaks
@@ -308,7 +308,7 @@ CSM Transformers integration supports training!
from transformers import CsmForConditionalGeneration, AutoProcessor
from datasets import load_dataset, Audio
-model_id = "eustlb/csm-1b"
+model_id = "sesame/csm-1b"
device = "cuda"
# load the model and the processor
@@ -356,6 +356,10 @@ The original code can be found [here](https://github.com/SesameAILabs/csm).
## CsmProcessor
+
+

+
+
[[autodoc]] CsmProcessor
- __call__
diff --git a/src/transformers/models/csm/configuration_csm.py b/src/transformers/models/csm/configuration_csm.py
index e6d6d2e27c..b13b9d2a87 100644
--- a/src/transformers/models/csm/configuration_csm.py
+++ b/src/transformers/models/csm/configuration_csm.py
@@ -28,7 +28,7 @@ class CsmDepthDecoderConfig(PretrainedConfig):
model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield
a similar configuration to that of the csm-1b.
- e.g. [eustlb/csm-1b](https://huggingface.co/eustlb/csm-1b)
+ e.g. [sesame/csm-1b](https://huggingface.co/sesame/csm-1b)
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
documentation from [`PretrainedConfig`] for more information.
@@ -210,7 +210,7 @@ class CsmConfig(PretrainedConfig):
model according to the specified arguments, defining the model architecture. Instantiating a configuration
with the defaults will yield a similar configuration to that of the csm-1b.
- e.g. [eustlb/csm-1b](https://huggingface.co/eustlb/csm-1b)
+ e.g. [sesame/csm-1b](https://huggingface.co/sesame/csm-1b)
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
documentation from [`PretrainedConfig`] for more information.
diff --git a/src/transformers/models/csm/generation_csm.py b/src/transformers/models/csm/generation_csm.py
index 2fec3ea891..7afc7c2d60 100644
--- a/src/transformers/models/csm/generation_csm.py
+++ b/src/transformers/models/csm/generation_csm.py
@@ -415,7 +415,7 @@ class CsmGenerationMixin(GenerationMixin):
>>> from transformers import CsmProcessor, CsmForConditionalGeneration
>>> from datasets import load_dataset, Audio
- >>> model_id = "eustlb/csm-1b"
+ >>> model_id = "sesame/csm-1b"
>>> torch_device = "cuda" if torch.cuda.is_available() else "cpu"
>>> processor = AutoProcessor.from_pretrained(model_id)
diff --git a/src/transformers/models/csm/modeling_csm.py b/src/transformers/models/csm/modeling_csm.py
index 7771695654..c0c4f5927a 100644
--- a/src/transformers/models/csm/modeling_csm.py
+++ b/src/transformers/models/csm/modeling_csm.py
@@ -1113,7 +1113,7 @@ class CsmForConditionalGeneration(CsmPreTrainedModel, CsmGenerationMixin):
>>> from transformers import CsmForConditionalGeneration, AutoProcessor
>>> from datasets import load_dataset, Audio
- >>> model_id = "eustlb/csm-1b"
+ >>> model_id = "sesame/csm-1b"
>>> torch_device = "cuda" if torch.cuda.is_available() else "cpu"
>>> processor = AutoProcessor.from_pretrained(model_id)
diff --git a/src/transformers/models/csm/modular_csm.py b/src/transformers/models/csm/modular_csm.py
index 86483076d3..4322a2a07f 100644
--- a/src/transformers/models/csm/modular_csm.py
+++ b/src/transformers/models/csm/modular_csm.py
@@ -727,7 +727,7 @@ class CsmForConditionalGeneration(CsmPreTrainedModel, CsmGenerationMixin):
>>> from transformers import CsmForConditionalGeneration, AutoProcessor
>>> from datasets import load_dataset, Audio
- >>> model_id = "eustlb/csm-1b"
+ >>> model_id = "sesame/csm-1b"
>>> torch_device = "cuda" if torch.cuda.is_available() else "cpu"
>>> processor = AutoProcessor.from_pretrained(model_id)
diff --git a/src/transformers/models/csm/processing_csm.py b/src/transformers/models/csm/processing_csm.py
index a0f91a1c3d..ca516d8264 100644
--- a/src/transformers/models/csm/processing_csm.py
+++ b/src/transformers/models/csm/processing_csm.py
@@ -76,7 +76,7 @@ class CsmProcessor(ProcessorMixin):
ds = load_dataset("hf-internal-testing/dailytalk-dummy", split="train")
audio = ds[0]["audio"]["array"]
- processor = CsmProcessor.from_pretrained("eustlb/csm-1b")
+ processor = CsmProcessor.from_pretrained("sesame/csm-1b")
processor(
text=["<|begin_of_text|>[0]What are you working on?<|end_of_text|><|AUDIO|><|audio_eos|><|begin_of_text|>[1]I'm figuring out my budget.<|end_of_text|>"],
diff --git a/tests/models/csm/test_modeling_csm.py b/tests/models/csm/test_modeling_csm.py
index 26442ef845..2425ff35ec 100644
--- a/tests/models/csm/test_modeling_csm.py
+++ b/tests/models/csm/test_modeling_csm.py
@@ -417,7 +417,7 @@ class CsmForConditionalGenerationTest(ModelTesterMixin, GenerationTesterMixin, u
class CsmForConditionalGenerationIntegrationTest(unittest.TestCase):
def setUp(self):
# TODO: @eustlb, update with correct sesame's repo
- self.model_checkpoint = "eustlb/csm-1b"
+ self.model_checkpoint = "sesame/csm-1b"
def tearDown(self):
cleanup(torch_device, gc_collect=True)
diff --git a/tests/models/csm/test_processor_csm.py b/tests/models/csm/test_processor_csm.py
index da96381246..dcd344d120 100644
--- a/tests/models/csm/test_processor_csm.py
+++ b/tests/models/csm/test_processor_csm.py
@@ -37,8 +37,7 @@ class CsmProcessorTest(ProcessorTesterMixin, unittest.TestCase):
@classmethod
def setUpClass(cls):
- # TODO: @eustlb, change for hf-internal-testing/csm-1b
- cls.checkpoint = "eustlb/csm-1b"
+ cls.checkpoint = "hf-internal-testing/namespace-sesame-repo_name_csm-1b"
processor = CsmProcessor.from_pretrained(cls.checkpoint)
cls.audio_token = processor.audio_token
cls.audio_token_id = processor.audio_token_id