fix slow CI

huggingface · Nov 26, 2024 · 4229c03 · 4229c03
1 parent 43df3a4
commit 4229c03
Showing 1 changed file with 10 additions and 6 deletions.
diff --git a/tests/slow/test_sft_slow.py b/tests/slow/test_sft_slow.py
@@ -104,7 +104,7 @@ def test_sft_trainer_transformers(self, model_name, packing):
 
             model = AutoModelForCausalLM.from_pretrained(model_name)
             tokenizer = AutoTokenizer.from_pretrained(model_name)
-
+            tokenizer.pad_token = tokenizer.eos_token if tokenizer.pad_token is None else tokenizer.pad_token
             trainer = SFTTrainer(
                 model,
                 args=training_args,
@@ -138,6 +138,7 @@ def test_sft_trainer_peft(self, model_name, packing):
 
             model = AutoModelForCausalLM.from_pretrained(model_name)
             tokenizer = AutoTokenizer.from_pretrained(model_name)
+            tokenizer.pad_token = tokenizer.eos_token if tokenizer.pad_token is None else tokenizer.pad_token
 
             trainer = SFTTrainer(
                 model,
@@ -174,6 +175,7 @@ def test_sft_trainer_transformers_mp(self, model_name, packing):
 
             model = AutoModelForCausalLM.from_pretrained(model_name)
             tokenizer = AutoTokenizer.from_pretrained(model_name)
+            tokenizer.pad_token = tokenizer.eos_token if tokenizer.pad_token is None else tokenizer.pad_token
 
             trainer = SFTTrainer(
                 model,
@@ -209,7 +211,7 @@ def test_sft_trainer_transformers_mp_gc(self, model_name, packing, gradient_chec
 
             model = AutoModelForCausalLM.from_pretrained(model_name)
             tokenizer = AutoTokenizer.from_pretrained(model_name)
-
+            tokenizer.pad_token = tokenizer.eos_token if tokenizer.pad_token is None else tokenizer.pad_token
             trainer = SFTTrainer(
                 model,
                 args=training_args,
@@ -245,7 +247,7 @@ def test_sft_trainer_transformers_mp_gc_peft(self, model_name, packing, gradient
 
             model = AutoModelForCausalLM.from_pretrained(model_name)
             tokenizer = AutoTokenizer.from_pretrained(model_name)
-
+            tokenizer.pad_token = tokenizer.eos_token if tokenizer.pad_token is None else tokenizer.pad_token
             trainer = SFTTrainer(
                 model,
                 args=training_args,
@@ -288,6 +290,7 @@ def test_sft_trainer_transformers_mp_gc_device_map(
 
             model = AutoModelForCausalLM.from_pretrained(model_name, device_map=device_map)
             tokenizer = AutoTokenizer.from_pretrained(model_name)
+            tokenizer.pad_token = tokenizer.eos_token if tokenizer.pad_token is None else tokenizer.pad_token
 
             trainer = SFTTrainer(
                 model,
@@ -327,7 +330,7 @@ def test_sft_trainer_transformers_mp_gc_peft_qlora(self, model_name, packing, gr
 
             model = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=quantization_config)
             tokenizer = AutoTokenizer.from_pretrained(model_name)
-
+            tokenizer.pad_token = tokenizer.eos_token if tokenizer.pad_token is None else tokenizer.pad_token
             trainer = SFTTrainer(
                 model,
                 args=training_args,
@@ -369,8 +372,9 @@ def test_sft_trainer_with_chat_format_qlora(self, model_name, packing):
 
             model = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=quantization_config)
             tokenizer = AutoTokenizer.from_pretrained(model_name)
-
-            model, tokenizer = setup_chat_format(model, tokenizer)
+            tokenizer.pad_token = tokenizer.eos_token if tokenizer.pad_token is None else tokenizer.pad_token
+            if tokenizer.chat_template is None:
+                model, tokenizer = setup_chat_format(model, tokenizer)
 
             trainer = SFTTrainer(
                 model,