chore(dependencies): bump transformers v5 (#2964)

* chore(dependencies): upgrade transformers + hggingface-hub + peft + scipy

* chore(dependencies): bump pi0 family to transformers v5

* chore(dependencies): bump wall x to transformers v5

* chore(dependencies): bump gr00t to transformers v5

* chore(style): fix pre-commit

* fix(policy): xvla forced_bos_token missing

* test(rl): skip ci tests for resnet10

* Fix: full pi models support for transformer v5 (#2967)

* fix(pi): remove loss truncation

* fix(pi): remove state padding before tokenization

* fix(pi): fix image padding value

* fix from_pretrain

* add transformer v5 changes

* remove reference

* more fixes

* make it work

* add support for rest of pi family

* add pifast work

* more changes

* more changes

* more cleanup

* fix torch params

* dtype fix

* torch compile

* embed mismatch fix

* revert groot

* more nit fixes

* remove unused classes

* more fixes

* revert

* nit

* torch dtype warning fix

* but back dynamic renaming

* add tie embedding

---------

Co-authored-by: Yufei Sun <skieyfly@gmail.com>

* chore: fix XVLA in transformers v5 (#3006)

* test(policies): enable wall x CI testing

* style(test): pre-commit check

* style(test): pre-commit

* fix wall x for transformer v5 (#3008)

* tv5 fix

* various wall x fixes

* Delete tests/policies/pi0_pi05/print_pi05_output_logits.py

Signed-off-by: Jade Choghari <chogharijade@gmail.com>

* sync modeling_florence2.py with chore/bump_transformers_v5

* more

* more fixes

* more

* remove comment

* more

---------

Signed-off-by: Jade Choghari <chogharijade@gmail.com>

* chore(dependencies): adjust dependencies versioning after transformers v5 (#3034)

* chore(dependecies): adjust dependecies versioning after transformers v5

* fix(policies): remove deprecated input_embeds

* fix(policies): dict _tied_weights_keys

* chore(depedencies): common qwen-vl-utils

* chore(dependencies): bump transformers to 5.2

* Fix policy testing for tv5 (#3032)

* fix ci logger

* other fix

* fix mypy

* change logits to torch2.10

* skip wallx|

* remove logging

---------

Co-authored-by: Steven Palma <imstevenpmwork@ieee.org>

* feat(ci): log into HF to unblock some CI tests (#3007)

* feat(ci): log into HF to unblock some CI tests

* chore(ci): change hf call + secret name

* fix(ci): temp fix for pi0 rtc test

* test(policies): require_cuda for unblocked tests

* test(policies): require_cuda wall_x

* fic(tests): require_cuda outter most for pi0

* fix(test): return instead of yield

---------

Signed-off-by: Steven Palma <imstevenpmwork@ieee.org>

* style(test): fix pre-commit

* chore(deps): upgrade transformers (#3050)

* chore(test): use lerobot model

* fix(policies): change default action tokenizer for wall x

* sample on cpu

* Revert "Merge branch 'chore/bump_transformers_v5' of https://github.com/huggingface/lerobot into chore/bump_transformers_v5"

This reverts commit d9b76755f7, reversing
changes made to 89359cb0b6.

* Reapply "Merge branch 'chore/bump_transformers_v5' of https://github.com/huggingface/lerobot into chore/bump_transformers_v5"

This reverts commit c9914db78b.

---------

Signed-off-by: Jade Choghari <chogharijade@gmail.com>
Signed-off-by: Steven Palma <imstevenpmwork@ieee.org>
Co-authored-by: Jade Choghari <chogharijade@gmail.com>
Co-authored-by: Yufei Sun <skieyfly@gmail.com>
Co-authored-by: Pepijn <pepijn@huggingface.co>

src/lerobot/policies/xvla/configuration_florence2.py

@@ -276,6 +276,8 @@ class Florence2LanguageConfig(PretrainedConfig):
         )
 
         # ensure backward compatibility for BART CNN models
+        if not hasattr(self, "forced_bos_token_id"):
+            self.forced_bos_token_id = None
         if self.forced_bos_token_id is None and kwargs.get("force_bos_token_to_be_generated", False):
             self.forced_bos_token_id = self.bos_token_id
             warnings.warn(

src/lerobot/policies/xvla/modeling_florence2.py

@@ -1951,7 +1951,10 @@ class Florence2Decoder(Florence2LanguagePreTrainedModel):
 
 
 class Florence2LanguageModel(Florence2LanguagePreTrainedModel):
-    _tied_weights_keys = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight"]
+    _tied_weights_keys = {
+        "encoder.embed_tokens.weight": "shared.weight",
+        "decoder.embed_tokens.weight": "shared.weight",
+    }
 
     def __init__(self, config: Florence2LanguageConfig):
         super().__init__(config)
@@ -2076,7 +2079,10 @@ class Florence2LanguageModel(Florence2LanguagePreTrainedModel):
 
 class Florence2LanguageForConditionalGeneration(Florence2LanguagePreTrainedModel, GenerationMixin):
     base_model_prefix = "model"
-    _tied_weights_keys = ["encoder.embed_tokens.weight", "decoder.embed_tokens.weight", "lm_head.weight"]
+    _tied_weights_keys = {
+        "model.encoder.embed_tokens.weight": "model.shared.weight",
+        "model.decoder.embed_tokens.weight": "model.shared.weight",
+    }
     _keys_to_ignore_on_load_missing = ["final_logits_bias"]
 
     def __init__(self, config: Florence2LanguageConfig):
@@ -2436,11 +2442,10 @@ FLORENCE2_INPUTS_DOCSTRING = r"""
     FLORENCE2_START_DOCSTRING,
 )
 class Florence2ForConditionalGeneration(Florence2PreTrainedModel):
-    _tied_weights_keys = [
-        "language_model.encoder.embed_tokens.weight",
-        "language_model.decoder.embed_tokens.weight",
-        "language_model.lm_head.weight",
-    ]
+    _tied_weights_keys = {
+        "language_model.model.encoder.embed_tokens.weight": "language_model.model.shared.weight",
+        "language_model.model.decoder.embed_tokens.weight": "language_model.model.shared.weight",
+    }
 
     def __init__(self, config: Florence2Config):
         super().__init__(config)