Update some type hints

src/transformers/configuration_utils.py

@@ -31,6 +31,7 @@
 from .generation.configuration_utils import GenerationConfig
 from .modeling_gguf_pytorch_utils import load_gguf_checkpoint
 from .modeling_rope_utils import RotaryEmbeddingConfigMixin
+from .tokenization_utils_base import PreTrainedTokenizerBase
 from .utils import (
     CONFIG_NAME,
     PushToHubMixin,
@@ -189,7 +190,7 @@ class PreTrainedConfig(PushToHubMixin, RotaryEmbeddingConfigMixin):
     problem_type: Literal["regression", "single_label_classification", "multi_label_classification"] | None = None
 
     # Tokenizer kwargs
-    tokenizer_class: str | None = None
+    tokenizer_class: str | PreTrainedTokenizerBase | None = None
 
     def __post_init__(self, **kwargs):
         # BC for the `torch_dtype` argument instead of the simpler `dtype`
@@ -1265,3 +1266,18 @@ def recursive_diff_dict(dict_a, dict_b, config_obj=None):
 
 # The alias is only here for BC - we did not have the correct CamelCasing before
 PretrainedConfig = PreTrainedConfig
+
+
+def layer_type_validation(layer_types: list[str], num_hidden_layers: int | None = None, attention: bool = True):
+    logger.warning(
+        "`layer_type_validation` is deprecated and will be removed in v5.20. "
+        "Use `PreTrainedConfig.validate_layer_type` instead"
+    )
+
+    if not all(layer_type in ALLOWED_LAYER_TYPES for layer_type in layer_types):
+        raise ValueError(f"The `layer_types` entries must be in {ALLOWED_LAYER_TYPES}")
+    if num_hidden_layers is not None and num_hidden_layers != len(layer_types):
+        raise ValueError(
+            f"`num_hidden_layers` ({num_hidden_layers}) must be equal to the number of layer types "
+            f"({len(layer_types)})"
+        )

src/transformers/models/blenderbot/configuration_blenderbot.py

@@ -73,7 +73,7 @@ class BlenderbotConfig(PreTrainedConfig):
     bos_token_id: int | None = 1
     eos_token_id: int | list[int] | None = 2
     encoder_no_repeat_ngram_size: int = 3
-    forced_eos_token_id: int | None = 2
+    forced_eos_token_id: int | list[int] | None = 2
     is_decoder: bool = False
     tie_word_embeddings: bool = True
 

src/transformers/models/blenderbot_small/configuration_blenderbot_small.py

@@ -69,7 +69,7 @@ class BlenderbotSmallConfig(PreTrainedConfig):
     pad_token_id: int | None = 0
     bos_token_id: int | None = 1
     eos_token_id: int | list[int] | None = 2
-    forced_eos_token_id: int | None = 2
+    forced_eos_token_id: int | list[int] | None = 2
     is_decoder: bool = False
     tie_word_embeddings: bool = True
 

src/transformers/models/clvp/configuration_clvp.py

@@ -174,7 +174,7 @@ class ClvpDecoderConfig(PreTrainedConfig):
     summary_first_dropout: float | int = 0.1
     use_cache: bool = True
     bos_token_id: int | None = 8192
-    eos_token_id: int | None = 8193
+    eos_token_id: int | list[int] | None = 8193
     pad_token_id: int | None = None
     feature_size: int = 80
     use_attention_bias: bool = True

src/transformers/models/deepseek_v3/configuration_deepseek_v3.py

@@ -78,7 +78,7 @@ class DeepseekV3Config(PreTrainedConfig):
     n_routed_experts: int = 256
     routed_scaling_factor: float = 2.5
     kv_lora_rank: int = 512
-    q_lora_rank: int = 1536
+    q_lora_rank: int | None = 1536
     qk_rope_head_dim: int = 64
     v_head_dim: int | None = 128
     qk_nope_head_dim: int = 128

src/transformers/models/dia/configuration_dia.py

@@ -78,7 +78,7 @@ class DiaDecoderConfig(PreTrainedConfig):
     use_cache: bool = True
     is_encoder_decoder: bool = True
     pad_token_id: int | None = 1025
-    eos_token_id: int | None = 1024
+    eos_token_id: int | list[int] | None = 1024
     bos_token_id: int | None = 1026
 
 

src/transformers/models/eurobert/configuration_eurobert.py

@@ -61,7 +61,7 @@ class EuroBertConfig(LlamaConfig):
     initializer_range: float = 0.02
     rms_norm_eps: float = 1e-05
     bos_token_id: int | None = 128000
-    eos_token_id: int | None = 128001
+    eos_token_id: int | list[int] | None = 128001
     pad_token_id: int | None = 128001
     mask_token_id: int = 128002
     pretraining_tp: int = 1

src/transformers/models/eurobert/modular_eurobert.py

@@ -63,7 +63,7 @@ class EuroBertConfig(LlamaConfig):
     initializer_range: float = 0.02
     rms_norm_eps: float = 1e-05
     bos_token_id: int | None = 128000
-    eos_token_id: int | None = 128001
+    eos_token_id: int | list[int] | None = 128001
     pad_token_id: int | None = 128001
     mask_token_id: int = 128002
     pretraining_tp: int = 1

src/transformers/models/exaone_moe/configuration_exaone_moe.py

@@ -94,7 +94,7 @@ class ExaoneMoeConfig(PreTrainedConfig):
     rms_norm_eps: float = 1e-5
     use_cache: bool = True
     bos_token_id: int | None = 1
-    eos_token_id: int | None = 53
+    eos_token_id: int | list[int] | None = 53
     pad_token_id: int | None = 0
     tie_word_embeddings: bool = False
     rope_parameters: dict | None = None

src/transformers/models/exaone_moe/modular_exaone_moe.py

@@ -92,7 +92,7 @@ class ExaoneMoeConfig(Exaone4Config):
     rms_norm_eps: float = 1e-5
     use_cache: bool = True
     bos_token_id: int | None = 1
-    eos_token_id: int | None = 53
+    eos_token_id: int | list[int] | None = 53
     pad_token_id: int | None = 0
     tie_word_embeddings: bool = False
     rope_parameters: dict | None = None

src/transformers/models/flaubert/configuration_flaubert.py

@@ -136,7 +136,7 @@ class FlaubertConfig(PreTrainedConfig):
     lang_id: int = 0
     pad_token_id: int | None = 2
     bos_token_id: int | None = 0
-    eos_token_id: int | None = 1
+    eos_token_id: int | list[int] | None = 1
     tie_word_embeddings: bool = True
 
 

src/transformers/models/fnet/configuration_fnet.py

@@ -62,7 +62,7 @@ class FNetConfig(PreTrainedConfig):
     tpu_short_seq_length: int = 512
     pad_token_id: int | None = 3
     bos_token_id: int | None = 1
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
     tie_word_embeddings: bool = True
 
 

src/transformers/models/fsmt/configuration_fsmt.py

@@ -97,8 +97,8 @@ class FSMTConfig(PreTrainedConfig):
     use_cache: bool = True
     pad_token_id: int | None = 1
     bos_token_id: int | None = 0
-    eos_token_id: int | None = 2
-    forced_eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
+    forced_eos_token_id: int | list[int] | None = 2
 
     def __post_init__(self, **kwargs):
         kwargs.pop("decoder", None)  # delete unused kwargs

src/transformers/models/git/configuration_git.py

@@ -98,7 +98,7 @@ class GitConfig(PreTrainedConfig):
     use_cache: bool = True
     tie_word_embeddings: bool = False
     bos_token_id: int | None = 101
-    eos_token_id: int | None = 102
+    eos_token_id: int | list[int] | None = 102
     num_image_with_embedding: int | None = None
 
     def __post_init__(self, **kwargs):

src/transformers/models/glm4_moe_lite/configuration_glm4_moe_lite.py

@@ -84,7 +84,7 @@ class Glm4MoeLiteConfig(PreTrainedConfig):
     n_routed_experts: int = 64
     routed_scaling_factor: float = 1.8
     kv_lora_rank: int = 512
-    q_lora_rank: int = 768
+    q_lora_rank: int | None = 768
     qk_rope_head_dim: int = 64
     v_head_dim: int = 256
     qk_nope_head_dim: int = 192

src/transformers/models/glm4_moe_lite/modular_glm4_moe_lite.py

@@ -92,7 +92,7 @@ class Glm4MoeLiteConfig(PreTrainedConfig):
     n_routed_experts: int = 64
     routed_scaling_factor: float = 1.8
     kv_lora_rank: int = 512
-    q_lora_rank: int = 768
+    q_lora_rank: int | None = 768
     qk_rope_head_dim: int = 64
     v_head_dim: int = 256
     qk_nope_head_dim: int = 192

src/transformers/models/gpt2/configuration_gpt2.py

@@ -95,7 +95,7 @@ class GPT2Config(PreTrainedConfig):
     scale_attn_weights: bool = True
     use_cache: bool = True
     bos_token_id: int | None = 50256
-    eos_token_id: int | None = 50256
+    eos_token_id: int | list[int] | None = 50256
     pad_token_id: int | None = None
     scale_attn_by_inverse_layer_idx: bool = False
     reorder_and_upcast_attn: bool = False

src/transformers/models/gpt_bigcode/configuration_gpt_bigcode.py

@@ -73,7 +73,7 @@ class GPTBigCodeConfig(PreTrainedConfig):
     scale_attn_weights: bool = True
     use_cache: bool = True
     bos_token_id: int | None = 50256
-    eos_token_id: int | None = 50256
+    eos_token_id: int | list[int] | None = 50256
     pad_token_id: int | None = None
     attention_softmax_in_fp32: bool = True
     scale_attention_softmax_in_fp32: bool = True

src/transformers/models/gpt_neo/configuration_gpt_neo.py

@@ -66,7 +66,7 @@ class GPTNeoConfig(PreTrainedConfig):
     initializer_range: float = 0.02
     use_cache: bool = True
     bos_token_id: int | None = 50256
-    eos_token_id: int | None = 50256
+    eos_token_id: int | list[int] | None = 50256
     pad_token_id: int | None = None
     tie_word_embeddings: bool = True
 

src/transformers/models/gptj/configuration_gptj.py

@@ -64,7 +64,7 @@ class GPTJConfig(PreTrainedConfig):
     initializer_range: float = 0.02
     use_cache: bool = True
     bos_token_id: int | None = 50256
-    eos_token_id: int | None = 50256
+    eos_token_id: int | list[int] | None = 50256
     pad_token_id: int | None = None
     tie_word_embeddings: bool = False
 

src/transformers/models/groupvit/configuration_groupvit.py

@@ -57,7 +57,7 @@ class GroupViTTextConfig(PreTrainedConfig):
     initializer_factor: float = 1.0
     pad_token_id: int | None = 1
     bos_token_id: int | None = 49406
-    eos_token_id: int | None = 49407
+    eos_token_id: int | list[int] | None = 49407
 
 
 @auto_docstring(checkpoint="nvidia/groupvit-gcc-yfcc")

src/transformers/models/higgs_audio_v2/configuration_higgs_audio_v2.py

@@ -86,7 +86,7 @@ class HiggsAudioV2Config(PreTrainedConfig):
     use_cache: bool = True
     pad_token_id: int | None = 128001
     bos_token_id: int | None = 1
-    eos_token_id: int | None = 128009
+    eos_token_id: int | list[int] | None = 128009
     pretraining_tp: int | None = 1
     tie_word_embeddings: bool = False
     rope_parameters: RopeParameters | dict | None = None

src/transformers/models/higgs_audio_v2/modular_higgs_audio_v2.py

@@ -75,7 +75,7 @@ class HiggsAudioV2Config(LlamaConfig):
     num_attention_heads: int = 24
     num_key_value_heads: int = 8
     pad_token_id: int | None = 128001
-    eos_token_id: int | None = 128009
+    eos_token_id: int | list[int] | None = 128009
     head_dim: int | None = 128
     num_codebooks: int = 8
     codebook_size: int = 1024

src/transformers/models/hubert/configuration_hubert.py

@@ -159,7 +159,7 @@ class HubertConfig(PreTrainedConfig):
     classifier_proj_size: int = 256
     pad_token_id: int | None = 0
     bos_token_id: int | None = 1
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
 
     def __post_init__(self, **kwargs):
         self.num_feat_extract_layers = len(self.conv_dim)

src/transformers/models/ibert/configuration_ibert.py

@@ -53,7 +53,7 @@ class IBertConfig(PreTrainedConfig):
     layer_norm_eps: float = 1e-12
     pad_token_id: int | None = 1
     bos_token_id: int | None = 0
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
     quant_mode: bool = False
     force_dequant: str = "none"
 

src/transformers/models/idefics/configuration_idefics.py

@@ -133,7 +133,7 @@ class IdeficsConfig(PreTrainedConfig):
     use_cache: bool = True
     pad_token_id: int | None = 0
     bos_token_id: int | None = 1
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
     tie_word_embeddings: bool = False
     cross_layer_interval: int = 1
     qk_layer_norms: bool = False

src/transformers/models/jamba/configuration_jamba.py

@@ -62,7 +62,7 @@ class JambaConfig(PreTrainedConfig):
     router_aux_loss_coef: float = 0.001
     pad_token_id: int | None = 0
     bos_token_id: int | None = 1
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
     max_position_embeddings: int = 262144
     attention_dropout: float | int = 0.0
     num_experts_per_tok: int = 2

src/transformers/models/jina_embeddings_v3/configuration_jina_embeddings_v3.py

@@ -61,7 +61,7 @@ class JinaEmbeddingsV3Config(PreTrainedConfig):
     layer_norm_eps: float = 1e-5
     pad_token_id: int | None = 1
     bos_token_id: int | None = 0
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
     use_cache: bool = True
     classifier_dropout: float | int | None = None
     tie_word_embeddings: bool = True

src/transformers/models/kosmos2/configuration_kosmos2.py

@@ -56,7 +56,7 @@ class Kosmos2TextConfig(PreTrainedConfig):
     use_cache: bool = True
     pad_token_id: int | None = 1
     bos_token_id: int | None = 0
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
     add_cross_attention: bool = False
 
 

src/transformers/models/kosmos2_5/configuration_kosmos2_5.py

@@ -56,7 +56,7 @@ class Kosmos2_5TextConfig(PreTrainedConfig):
     use_cache: bool = True
     pad_token_id: int | None = 1
     bos_token_id: int | None = 0
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
 
 
 @auto_docstring(checkpoint="microsoft/kosmos-2.5")

src/transformers/models/layoutlmv3/configuration_layoutlmv3.py

@@ -80,7 +80,7 @@ class LayoutLMv3Config(PreTrainedConfig):
     layer_norm_eps: float = 1e-5
     pad_token_id: int | None = 1
     bos_token_id: int | None = 0
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
     max_2d_position_embeddings: int = 1024
     coordinate_size: int = 128
     shape_size: int = 128

src/transformers/models/led/configuration_led.py

@@ -78,7 +78,7 @@ class LEDConfig(PreTrainedConfig):
     classifier_dropout: float | int = 0.0
     pad_token_id: int | None = 1
     bos_token_id: int | None = 0
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
     attention_window: list[int] | int = 512
     tie_word_embeddings: bool = True
 

src/transformers/models/llama4/configuration_llama4.py

@@ -151,7 +151,7 @@ class Llama4TextConfig(PreTrainedConfig):
     use_cache: bool = True
     pad_token_id: int | None = None
     bos_token_id: int | None = 1
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
     tie_word_embeddings: bool = False
     attention_dropout: float | int = 0.0
     num_experts_per_tok: int = 1

src/transformers/models/longcat_flash/configuration_longcat_flash.py

@@ -91,7 +91,7 @@ class LongcatFlashConfig(PreTrainedConfig):
     attention_bias: bool = False
     attention_dropout: float | int = 0.0
     ffn_hidden_size: int = 12288
-    q_lora_rank: int = 1536
+    q_lora_rank: int | None = 1536
     kv_lora_rank: int = 512
     qk_nope_head_dim: int = 128
     qk_rope_head_dim: int = 64

src/transformers/models/longt5/configuration_longt5.py

@@ -72,7 +72,7 @@ class LongT5Config(PreTrainedConfig):
     encoder_attention_type: str = "local"
     use_cache: bool = True
     pad_token_id: int | None = 0
-    eos_token_id: int | None = 1
+    eos_token_id: int | list[int] | None = 1
     bos_token_id: int | None = None
     is_decoder: bool = False
     tie_word_embeddings: bool = True

src/transformers/models/luke/configuration_luke.py

@@ -68,7 +68,7 @@ class LukeConfig(PreTrainedConfig):
     classifier_dropout: float | int | None = None
     pad_token_id: int | None = 1
     bos_token_id: int | None = 0
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
     tie_word_embeddings: bool = True
 
 

src/transformers/models/m2m_100/configuration_m2m_100.py

@@ -68,7 +68,7 @@ class M2M100Config(PreTrainedConfig):
     scale_embedding: int = True
     pad_token_id: int | None = 1
     bos_token_id: int | None = 0
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
     tie_word_embeddings: bool = True
 
 

src/transformers/models/mamba2/configuration_mamba2.py

@@ -68,7 +68,7 @@ class Mamba2Config(PreTrainedConfig):
     layer_norm_epsilon: float = 1e-5
     pad_token_id: int | None = 1
     bos_token_id: int | None = 0
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
     expand: int = 2
     conv_kernel: int = 4
     n_groups: int = 8

src/transformers/models/marian/configuration_marian.py

@@ -73,9 +73,9 @@ class MarianConfig(PreTrainedConfig):
     decoder_start_token_id: int = 58100
     scale_embedding: bool = False
     pad_token_id: int | None = 58100
-    eos_token_id: int | None = 0
+    eos_token_id: int | list[int] | None = 0
     bos_token_id: int | None = None
-    forced_eos_token_id: int | None = 0
+    forced_eos_token_id: int | list[int] | None = 0
     share_encoder_decoder_embeddings: bool = True
     is_decoder: bool = False
     tie_word_embeddings: bool = True

src/transformers/models/markuplm/configuration_markuplm.py

@@ -75,7 +75,7 @@ class MarkupLMConfig(PreTrainedConfig):
     layer_norm_eps: float = 1e-12
     pad_token_id: int | None = 0
     bos_token_id: int | None = 0
-    eos_token_id: int | None = 2
+    eos_token_id: int | list[int] | None = 2
     max_xpath_tag_unit_embeddings: int = 256
     max_xpath_subs_unit_embeddings: int = 1024
     tag_pad_id: int = 216