Upload 5 files
Browse files- ultravox_config.py +13 -6
ultravox_config.py
CHANGED
@@ -32,6 +32,8 @@ class LossFunction(str, Enum):
|
|
32 |
class LossConfig:
|
33 |
loss_function: LossFunction = LossFunction.CrossEntropy
|
34 |
kl_temperature: float = 2.0
|
|
|
|
|
35 |
|
36 |
@property
|
37 |
def requires_alt_fields(self):
|
@@ -47,7 +49,7 @@ class UltravoxConfig(transformers.PretrainedConfig):
|
|
47 |
documentation from [`PretrainedConfig`] for more information.
|
48 |
|
49 |
Args:
|
50 |
-
audio_config (`
|
51 |
Custom audio config or dict
|
52 |
text_config (`Union[AutoConfig, dict]`, *optional*):
|
53 |
The config object of the text backbone. Can be any of `LlamaConfig` or `MistralConfig`.
|
@@ -72,10 +74,10 @@ class UltravoxConfig(transformers.PretrainedConfig):
|
|
72 |
Example:
|
73 |
|
74 |
```python
|
75 |
-
>>> from transformers import
|
76 |
|
77 |
>>> # Initializing an audio encoder config
|
78 |
-
>>> audio_config =
|
79 |
|
80 |
>>> # Initializing a Llama config
|
81 |
>>> text_config = LlamaConfig()
|
@@ -84,13 +86,13 @@ class UltravoxConfig(transformers.PretrainedConfig):
|
|
84 |
>>> configuration = UltravoxConfig(audio_config, text_config)
|
85 |
|
86 |
>>> # Initializing a completely untrained model from the configuration
|
87 |
-
>>> model =
|
88 |
|
89 |
>>> # Accessing the model configuration
|
90 |
>>> configuration = model.config
|
91 |
|
92 |
>>> # Initialize a model from pretrained checkpoints and random projector weights
|
93 |
-
>>> config = UltravoxConfig(audio_model_id="
|
94 |
```"""
|
95 |
|
96 |
model_type = "ultravox"
|
@@ -140,7 +142,7 @@ class UltravoxConfig(transformers.PretrainedConfig):
|
|
140 |
else:
|
141 |
audio_config = audio_config or {}
|
142 |
self.audio_config = transformers.CONFIG_MAPPING[
|
143 |
-
audio_config.get("model_type", "
|
144 |
](**audio_config)
|
145 |
|
146 |
self.text_model_lora_config = (
|
@@ -167,7 +169,12 @@ class UltravoxConfig(transformers.PretrainedConfig):
|
|
167 |
# remove text_config and audio_config if text_model_id and audio_model_id are present
|
168 |
if self.text_model_id is not None:
|
169 |
diff_dict.pop("text_config", None)
|
|
|
|
|
|
|
170 |
if self.audio_model_id is not None:
|
171 |
diff_dict.pop("audio_config", None)
|
|
|
|
|
172 |
|
173 |
return diff_dict
|
|
|
32 |
class LossConfig:
|
33 |
loss_function: LossFunction = LossFunction.CrossEntropy
|
34 |
kl_temperature: float = 2.0
|
35 |
+
# Number of tokens to ignore from the beginning of the sequence. Only used in LSM
|
36 |
+
initial_tokens_to_ignore: int = 0
|
37 |
|
38 |
@property
|
39 |
def requires_alt_fields(self):
|
|
|
49 |
documentation from [`PretrainedConfig`] for more information.
|
50 |
|
51 |
Args:
|
52 |
+
audio_config (`WhisperConfig`, *optional*):
|
53 |
Custom audio config or dict
|
54 |
text_config (`Union[AutoConfig, dict]`, *optional*):
|
55 |
The config object of the text backbone. Can be any of `LlamaConfig` or `MistralConfig`.
|
|
|
74 |
Example:
|
75 |
|
76 |
```python
|
77 |
+
>>> from transformers import UltravoxModel, WhisperConfig, UltravoxConfig, LlamaConfig
|
78 |
|
79 |
>>> # Initializing an audio encoder config
|
80 |
+
>>> audio_config = WhisperConfig()
|
81 |
|
82 |
>>> # Initializing a Llama config
|
83 |
>>> text_config = LlamaConfig()
|
|
|
86 |
>>> configuration = UltravoxConfig(audio_config, text_config)
|
87 |
|
88 |
>>> # Initializing a completely untrained model from the configuration
|
89 |
+
>>> model = UltravoxModel(configuration)
|
90 |
|
91 |
>>> # Accessing the model configuration
|
92 |
>>> configuration = model.config
|
93 |
|
94 |
>>> # Initialize a model from pretrained checkpoints and random projector weights
|
95 |
+
>>> config = UltravoxConfig(audio_model_id="openai/whisper-tiny", text_model_id="meta-llama/Llama-2-7b-chat-hf")
|
96 |
```"""
|
97 |
|
98 |
model_type = "ultravox"
|
|
|
142 |
else:
|
143 |
audio_config = audio_config or {}
|
144 |
self.audio_config = transformers.CONFIG_MAPPING[
|
145 |
+
audio_config.get("model_type", "whisper")
|
146 |
](**audio_config)
|
147 |
|
148 |
self.text_model_lora_config = (
|
|
|
169 |
# remove text_config and audio_config if text_model_id and audio_model_id are present
|
170 |
if self.text_model_id is not None:
|
171 |
diff_dict.pop("text_config", None)
|
172 |
+
elif "text_config" in diff_dict:
|
173 |
+
diff_dict["text_config"].pop("_attn_implementation_autoset", None)
|
174 |
+
|
175 |
if self.audio_model_id is not None:
|
176 |
diff_dict.pop("audio_config", None)
|
177 |
+
elif "audio_config" in diff_dict:
|
178 |
+
diff_dict["audio_config"].pop("_attn_implementation_autoset", None)
|
179 |
|
180 |
return diff_dict
|