Rename trainer arg tokenizer to processing_class

README.md

@@ -133,7 +133,7 @@ training_args = RewardConfig(output_dir="Qwen2.5-0.5B-Reward", per_device_train_
 trainer = RewardTrainer(
     args=training_args,
     model=model,
-    tokenizer=tokenizer,
+    processing_class=tokenizer,
     train_dataset=dataset,
 )
 trainer.train()
@@ -166,7 +166,7 @@ dataset = dataset.map(lambda x: tokenizer(x["prompt"]), remove_columns="prompt")
 training_args = RLOOConfig(output_dir="Qwen2.5-0.5B-RL")
 trainer = RLOOTrainer(
     config=training_args,
-    tokenizer=tokenizer,
+    processing_class=tokenizer,
     policy=policy,
     ref_policy=ref_policy,
     reward_model=reward_model,
@@ -189,7 +189,7 @@ model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-0.5B-Instruct")
 tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-0.5B-Instruct")
 dataset = load_dataset("trl-lib/Capybara-Preferences", split="train")
 training_args = DPOConfig(output_dir="Qwen2.5-0.5B-DPO")
-trainer = DPOTrainer(model=model, args=training_args, train_dataset=dataset, tokenizer=tokenizer)
+trainer = DPOTrainer(model=model, args=training_args, train_dataset=dataset, processing_class=tokenizer)
 trainer.train()
 ```
 

docs/source/bco_trainer.mdx

@@ -32,7 +32,7 @@ bco_trainer = BCOTrainer(
     model_ref,
     args=training_args,
     train_dataset=train_dataset,
-    tokenizer=tokenizer,
+    processing_class=tokenizer,
 )
 ```
 After this one can then call:
@@ -75,7 +75,7 @@ bco_trainer = BCOTrainer(
     model_ref,
     args=training_args,
     train_dataset=train_dataset,
-    tokenizer=tokenizer,
+    processing_class=tokenizer,
     embedding_func=embedding_func,
     embedding_tokenizer=self.embedding_tokenizer,
 )

docs/source/cpo_trainer.mdx

@@ -32,7 +32,7 @@ tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2-0.5B-Instruct")
 train_dataset = load_dataset("trl-lib/Capybara-Preferences", split="train")
 
 training_args = CPOConfig(output_dir="Qwen2-0.5B-CPO", logging_steps=10)
-trainer = CPOTrainer(model=model, args=training_args, tokenizer=tokenizer, train_dataset=train_dataset)
+trainer = CPOTrainer(model=model, args=training_args, processing_class=tokenizer, train_dataset=train_dataset)
 trainer.train()
 ```
 

docs/source/dpo_trainer.mdx

@@ -47,7 +47,7 @@ tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2-0.5B-Instruct")
 train_dataset = load_dataset("trl-lib/Capybara-Preferences", split="train")
 
 training_args = DPOConfig(output_dir="Qwen2-0.5B-DPO", logging_steps=10)
-trainer = DPOTrainer(model=model, args=training_args, tokenizer=tokenizer, train_dataset=train_dataset)
+trainer = DPOTrainer(model=model, args=training_args, processing_class=tokenizer, train_dataset=train_dataset)
 trainer.train()
 ```
 
@@ -100,8 +100,8 @@ Additionally, unlike standard text-based models where a `tokenizer` is used, for
       model,
       args=training_args,
       train_dataset=train_dataset,
--     tokenizer=tokenizer,
-+     tokenizer=processor,
+-     processing_class=tokenizer,
++     processing_class=processor,
 )
 ```
 
@@ -194,7 +194,7 @@ First install `unsloth` according to the [official documentation](https://github
 
 - training_args = DPOConfig(output_dir="Qwen2-0.5B-DPO", logging_steps=10)
 + training_args = DPOConfig(output_dir="Qwen2-0.5B-DPO", logging_steps=10, bf16=True)
-  trainer = DPOTrainer(model=model, args=training_args, tokenizer=tokenizer, train_dataset=train_dataset)
+  trainer = DPOTrainer(model=model, args=training_args, processing_class=tokenizer, train_dataset=train_dataset)
   trainer.train()
 
 ```

docs/source/gkd_trainer.md

@@ -74,7 +74,7 @@ trainer = GKDTrainer(
     model=model,
     teacher_model=teacher_model,
     args=training_args,
-    tokenizer=tokenizer,
+    processing_class=tokenizer,
     train_dataset=train_dataset,
     eval_dataset=eval_dataset,
 )

docs/source/kto_trainer.mdx

@@ -83,7 +83,7 @@ kto_trainer = KTOTrainer(
     ref_model,
     args=training_args,
     train_dataset=train_dataset,
-    tokenizer=tokenizer,
+    processing_class=tokenizer,
 )
 ```
 After this one can then call:

docs/source/nash_md_trainer.md

@@ -41,7 +41,7 @@ trainer = NashMDTrainer(
     model=model,
     reward_model=reward_model,
     args=training_args,
-    tokenizer=tokenizer,
+    processing_class=tokenizer,
     train_dataset=train_dataset,
 )
 trainer.train()

docs/source/online_dpo_trainer.md

@@ -40,7 +40,7 @@ train_dataset = load_dataset("trl-lib/ultrafeedback-prompt", split="train")
 
 training_args = OnlineDPOConfig(output_dir="online-dpo-qwen2", logging_steps=10)
 trainer = OnlineDPOTrainer(
-    model=model, reward_model=reward_model, args=training_args, tokenizer=tokenizer, train_dataset=train_dataset
+    model=model, reward_model=reward_model, args=training_args, processing_class=tokenizer, train_dataset=train_dataset
 )
 trainer.train()
 ```

docs/source/orpo_trainer.md

@@ -66,7 +66,7 @@ orpo_trainer = ORPOTrainer(
     model,
     args=training_args,
     train_dataset=train_dataset,
-    tokenizer=tokenizer,
+    processing_class=tokenizer,
 )
 ```
 After this one can then call:

docs/source/reward_trainer.mdx

@@ -41,7 +41,7 @@ peft_config = LoraConfig(
 trainer = RewardTrainer(
     model=model,
     args=training_args,
-    tokenizer=tokenizer,
+    processing_class=tokenizer,
     train_dataset=dataset,
     peft_config=peft_config,
 )

docs/source/sft_trainer.mdx

@@ -744,7 +744,7 @@ trainer = SFTTrainer(
     args=training_args,
     data_collator=collate_fn,
     train_dataset=train_dataset,
-    tokenizer=processor.tokenizer,
+    processing_class=processor.tokenizer,
 )
 ```
 

docs/source/xpo_trainer.mdx

@@ -41,7 +41,7 @@ trainer = XPOTrainer(
     model=model,
     reward_model=reward_model,
     args=training_args,
-    tokenizer=tokenizer,
+    processing_class=tokenizer,
     train_dataset=train_dataset,
 )
 trainer.train()

examples/research_projects/stack_llama_2/scripts/dpo_llama2.py

@@ -237,7 +237,7 @@ def return_prompt_and_responses(samples) -> Dict[str, str]:
         beta=script_args.beta,
         train_dataset=train_dataset,
         eval_dataset=eval_dataset,
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
         peft_config=peft_config,
         max_prompt_length=script_args.max_prompt_length,
         max_length=script_args.max_length,

examples/research_projects/stack_llama_2/scripts/sft_llama2.py

@@ -187,7 +187,7 @@ def create_datasets(tokenizer, args, seed=None):
     peft_config=peft_config,
     max_seq_length=None,
     formatting_func=prepare_sample_text,
-    tokenizer=tokenizer,
+    processing_class=tokenizer,
     args=training_args,
 )
 trainer.train()

examples/scripts/bco.py

@@ -152,7 +152,7 @@ def mean_pooling(model_output, attention_mask):
         args=training_args,
         train_dataset=dataset["train"],
         eval_dataset=dataset["test"],
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
         peft_config=get_peft_config(model_args),
         embedding_func=embedding_func,
         embedding_tokenizer=embedding_tokenizer,

examples/scripts/cpo.py

@@ -100,7 +100,7 @@ class ScriptArguments:
         args=training_args,
         train_dataset=dataset["train"],
         eval_dataset=dataset["test"],
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
         peft_config=get_peft_config(model_config),
     )
 

examples/scripts/dpo.py

@@ -121,7 +121,7 @@
         args=training_args,
         train_dataset=dataset[script_args.dataset_train_split],
         eval_dataset=dataset[script_args.dataset_test_split],
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
         peft_config=peft_config,
     )
 

examples/scripts/dpo_online.py

@@ -107,7 +107,7 @@
         args=training_args,
         train_dataset=dataset[script_args.dataset_train_split],
         eval_dataset=dataset[script_args.dataset_test_split],
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
         peft_config=get_peft_config(model_config),
     )
     generation_config = GenerationConfig(

examples/scripts/dpo_vlm.py

@@ -126,7 +126,7 @@ def process(row):
         args=training_args,
         train_dataset=dataset[script_args.dataset_train_split],
         eval_dataset=dataset[script_args.dataset_test_split],
-        tokenizer=processor,
+        processing_class=processor,
         peft_config=peft_config,
     )
 

examples/scripts/gkd.py

@@ -122,7 +122,7 @@
         args=training_args,
         train_dataset=dataset[script_args.dataset_train_split],
         eval_dataset=dataset[script_args.dataset_test_split],
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
         peft_config=get_peft_config(model_config),
     )
     completions_callback = LogCompletionsCallback(trainer, trainer.generation_config, num_prompts=8)

examples/scripts/kto.py

@@ -122,7 +122,7 @@ def format_dataset(example):
         args=training_args,
         train_dataset=dataset["train"],
         eval_dataset=dataset["test"],
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
         peft_config=get_peft_config(model_args),
     )
 

examples/scripts/nash_md.py

@@ -110,7 +110,7 @@
         args=training_args,
         train_dataset=dataset[script_args.dataset_train_split],
         eval_dataset=dataset[script_args.dataset_test_split],
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
     )
     generation_config = GenerationConfig(
         max_new_tokens=training_args.max_new_tokens, do_sample=True, temperature=training_args.temperature

examples/scripts/orpo.py

@@ -112,7 +112,7 @@ def process(row):
         args=training_args,
         train_dataset=dataset["train"],
         eval_dataset=dataset["test"],
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
         peft_config=get_peft_config(model_config),
     )
 

examples/scripts/ppo/ppo.py

@@ -120,7 +120,7 @@ def tokenize(element):
     ################
     trainer = PPOv2Trainer(
         config=training_args,
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
         policy=policy,
         ref_policy=ref_policy,
         reward_model=reward_model,

examples/scripts/ppo/ppo_tldr.py

@@ -125,7 +125,7 @@ def tokenize(element):
     ################
     trainer = PPOv2Trainer(
         config=training_args,
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
         policy=policy,
         ref_policy=ref_policy,
         reward_model=reward_model,

examples/scripts/reward_modeling.py

@@ -111,7 +111,7 @@
     ##########
     trainer = RewardTrainer(
         model=model,
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
         args=training_args,
         train_dataset=dataset[script_args.dataset_train_split],
         eval_dataset=dataset[script_args.dataset_test_split],

examples/scripts/rloo/rloo.py

@@ -121,7 +121,7 @@ def tokenize(element):
     ################
     trainer = RLOOTrainer(
         config=training_args,
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
         policy=policy,
         ref_policy=ref_policy,
         reward_model=reward_model,

examples/scripts/rloo/rloo_tldr.py

@@ -125,7 +125,7 @@ def tokenize(element):
     ################
     trainer = RLOOTrainer(
         config=training_args,
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
         policy=policy,
         ref_policy=ref_policy,
         reward_model=reward_model,

examples/scripts/sft.py

@@ -95,7 +95,7 @@
         args=training_args,
         train_dataset=dataset[script_args.dataset_train_split],
         eval_dataset=dataset[script_args.dataset_test_split],
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
         peft_config=get_peft_config(model_config),
     )
 

examples/scripts/sft_vlm.py

@@ -119,7 +119,7 @@ def collate_fn(examples):
         data_collator=collate_fn,
         train_dataset=dataset[script_args.dataset_train_split],
         eval_dataset=dataset[script_args.dataset_test_split],
-        tokenizer=processor.tokenizer,
+        processing_class=processor.tokenizer,
         peft_config=get_peft_config(model_config),
     )
 

examples/scripts/xpo.py

@@ -94,7 +94,7 @@
         args=training_args,
         train_dataset=dataset[script_args.dataset_train_split],
         eval_dataset=dataset[script_args.dataset_test_split],
-        tokenizer=tokenizer,
+        processing_class=tokenizer,
     )
     generation_config = GenerationConfig(
         max_new_tokens=training_args.max_new_tokens, do_sample=True, temperature=training_args.temperature

tests/slow/test_dpo_slow.py

@@ -85,7 +85,7 @@ def test_dpo_bare_model(self, model_id, loss_type, pre_compute_logits):
                 model=model,
                 ref_model=None,
                 args=training_args,
-                tokenizer=tokenizer,
+                processing_class=tokenizer,
                 train_dataset=self.dataset,
                 eval_dataset=self.dataset,
             )
@@ -142,7 +142,7 @@ def test_dpo_peft_model(self, model_id, loss_type, pre_compute_logits, gradient_
                 model=model,
                 ref_model=None,
                 args=training_args,
-                tokenizer=tokenizer,
+                processing_class=tokenizer,
                 train_dataset=self.dataset,
                 eval_dataset=self.dataset,
                 peft_config=self.peft_config,
@@ -206,7 +206,7 @@ def test_dpo_peft_model_qlora(self, model_id, loss_type, pre_compute_logits, gra
                 model=model,
                 ref_model=None,
                 args=training_args,
-                tokenizer=tokenizer,
+                processing_class=tokenizer,
                 train_dataset=self.dataset,
                 eval_dataset=self.dataset,
                 peft_config=self.peft_config,