Docs / Quantization: Replace all occurences of load_in_8bit with bn…

…b config (#31136)

Replace all occurences of `load_in_8bit` with bnb config

docs/source/de/peft.md

@@ -86,10 +86,10 @@ model.load_adapter(peft_model_id)
 Die `bitsandbytes`-Integration unterstützt Datentypen mit 8bit und 4bit Genauigkeit, was für das Laden großer Modelle nützlich ist, weil es Speicher spart (lesen Sie den `bitsandbytes`-Integrations [guide](./quantization#bitsandbytes-integration), um mehr zu erfahren). Fügen Sie die Parameter `load_in_8bit` oder `load_in_4bit` zu [`~PreTrainedModel.from_pretrained`] hinzu und setzen Sie `device_map="auto"`, um das Modell effektiv auf Ihre Hardware zu verteilen:
 
 ```py
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
 peft_model_id = "ybelkada/opt-350m-lora"
-model = AutoModelForCausalLM.from_pretrained(peft_model_id, device_map="auto", load_in_8bit=True)
+model = AutoModelForCausalLM.from_pretrained(peft_model_id, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 ```
 
 ## Einen neuen Adapter hinzufügen

docs/source/en/peft.md

@@ -88,10 +88,10 @@ Check out the [API documentation](#transformers.integrations.PeftAdapterMixin) s
 The `bitsandbytes` integration supports 8bit and 4bit precision data types, which are useful for loading large models because it saves memory (see the `bitsandbytes` integration [guide](./quantization#bitsandbytes-integration) to learn more). Add the `load_in_8bit` or `load_in_4bit` parameters to [`~PreTrainedModel.from_pretrained`] and set `device_map="auto"` to effectively distribute the model to your hardware:
 
 ```py
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
 peft_model_id = "ybelkada/opt-350m-lora"
-model = AutoModelForCausalLM.from_pretrained(peft_model_id, device_map="auto", load_in_8bit=True)
+model = AutoModelForCausalLM.from_pretrained(peft_model_id, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 ```
 
 ## Add a new adapter

docs/source/en/perf_infer_gpu_one.md

@@ -354,20 +354,20 @@ If you're curious and interested in learning more about the concepts underlying
 To load a model in 8-bit for inference, use the `load_in_8bit` parameter. The `device_map` parameter is optional, but we recommend setting it to `"auto"` to allow 🤗 Accelerate to automatically and efficiently allocate the model given the available resources in the environment:
 
 ```py
-from transformers import AutoModelForCausalLM
+from transformers import AutoModelForCausalLM, BitsAndBytesConfig
 
 model_name = "bigscience/bloom-2b5"
-model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_8bit=True)
+model_8bit = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 ```
 
 If you're loading a model in 8-bit for text generation, you should use the [`~transformers.GenerationMixin.generate`] method instead of the [`Pipeline`] function which is not optimized for 8-bit models and will be slower. Some sampling strategies, like nucleus sampling, are also not supported by the [`Pipeline`] for 8-bit models. You should also place all inputs on the same device as the model:
 
 ```py
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
 model_name = "bigscience/bloom-2b5"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
-model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_8bit=True)
+model_8bit = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 
 prompt = "Hello, my llama is cute"
 inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

docs/source/it/perf_infer_gpu_one.md

@@ -55,10 +55,10 @@ Di seguito sono riportate alcune note per aiutarvi a utilizzare questo modulo, o
 Dopo aver installato le librerie necessarie, per caricare il tuo modello mixed 8-bit è il seguente:
 
 ```py
-from transformers import AutoModelForCausalLM
+from transformers import AutoModelForCausalLM, BitsAndBytesConfig
 
 model_name = "bigscience/bloom-2b5"
-model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_8bit=True)
+model_8bit = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 ```
 
 Per la generazione di testo, si consiglia di:
@@ -69,11 +69,11 @@ Per la generazione di testo, si consiglia di:
 Ecco un semplice esempio:
 
 ```py
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
 model_name = "bigscience/bloom-2b5"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
-model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_8bit=True)
+model_8bit = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 
 text = "Hello, my llama is cute"
 inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
@@ -87,7 +87,7 @@ outputs = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
 Usare il seguente modo caricare il modello mixed-8bit su più GPU (stesso comando della configurazione a GPU singola):
 ```py
 model_name = "bigscience/bloom-2b5"
-model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_8bit=True)
+model_8bit = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 ```
 Puoi controllare la RAM della GPU che si vuole allocare su ogni GPU usando `accelerate`. Utilizzare l'argomento `max_memory` come segue:
 

docs/source/ja/main_classes/quantization.md

@@ -245,12 +245,12 @@ model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", load_i
 
 ```python
 # pip install transformers accelerate bitsandbytes
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
 model_id = "bigscience/bloom-1b7"
 
 tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", load_in_8bit=True)
+model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 ```
 
 次に、通常 [`PreTrainedModel`] を使用するのと同じようにモデルを使用します。
@@ -321,9 +321,9 @@ model_double_quant = AutoModelForCausalLM.from_pretrained(model_id, quantization
 この機能を使用できるようにするには、必ず `bitsandbytes>0.37.2` を使用してください (この記事の執筆時点では、`bitsandbytes==0.38.0.post1` でテストしました)。
 
 ```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
-model = AutoModelForCausalLM.from_pretrained("bigscience/bloom-560m", device_map="auto", load_in_8bit=True)
+model = AutoModelForCausalLM.from_pretrained("bigscience/bloom-560m", quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 tokenizer = AutoTokenizer.from_pretrained("bigscience/bloom-560m")
 
 model.push_to_hub("bloom-560m-8bit")

docs/source/ja/peft.md

@@ -91,10 +91,10 @@ model.load_adapter(peft_model_id)
 `bitsandbytes` 統合は、8ビットおよび4ビットの精度データ型をサポートしており、大規模なモデルを読み込む際にメモリを節約するのに役立ちます（詳細については `bitsandbytes` 統合の[ガイド](./quantization#bitsandbytes-integration)を参照してください）。[`~PreTrainedModel.from_pretrained`] に `load_in_8bit` または `load_in_4bit` パラメータを追加し、`device_map="auto"` を設定してモデルを効果的にハードウェアに分散配置できます：
 
 ```py
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
 peft_model_id = "ybelkada/opt-350m-lora"
-model = AutoModelForCausalLM.from_pretrained(peft_model_id, device_map="auto", load_in_8bit=True)
+model = AutoModelForCausalLM.from_pretrained(peft_model_id, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 ```
 
 ## Add a new adapter

docs/source/ja/perf_infer_gpu_one.md

@@ -357,10 +357,10 @@ Int8混合精度行列分解は、行列乗算を2つのストリームに分割
 必要なライブラリをインストールした後、ミックス 8 ビットモデルを読み込む方法は次の通りです：
 
 ```py
-from transformers import AutoModelForCausalLM
+from transformers import AutoModelForCausalLM, BitsAndBytesConfig
 
 model_name = "bigscience/bloom-2b5"
-model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_8bit=True)
+model_8bit = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 ```
 
 以下はシンプルな例です：
@@ -370,11 +370,11 @@ model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto",
 
 
 ```py
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
 model_name = "bigscience/bloom-2b5"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
-model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_8bit=True)
+model_8bit = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 
 prompt = "Hello, my llama is cute"
 inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
@@ -388,7 +388,7 @@ outputs = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
 
 ```py
 model_name = "bigscience/bloom-2b5"
-model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_8bit=True)
+model_8bit = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 ```
 
 `accelerate`を使用して各GPUに割り当てるGPU RAMを制御する際には、以下のように`max_memory`引数を使用します：

docs/source/ko/peft.md

@@ -86,10 +86,10 @@ model.load_adapter(peft_model_id)
 `bitsandbytes` 통합은 8비트와 4비트 정밀도 데이터 유형을 지원하므로 큰 모델을 가져올 때 유용하면서 메모리도 절약합니다. 모델을 하드웨어에 효과적으로 분배하려면 [`~PreTrainedModel.from_pretrained`]에 `load_in_8bit` 또는 `load_in_4bit` 매개변수를 추가하고 `device_map="auto"`를 설정하세요:
 
 ```py
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
 peft_model_id = "ybelkada/opt-350m-lora"
-model = AutoModelForCausalLM.from_pretrained(peft_model_id, device_map="auto", load_in_8bit=True)
+model = AutoModelForCausalLM.from_pretrained(peft_model_id, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 ```
 
 ## 새 어댑터 추가 [[add-a-new-adapter]]

docs/source/ko/perf_infer_gpu_one.md

@@ -127,10 +127,10 @@ Int8 혼합 정밀도 행렬 분해는 행렬 곱셈을 두 개의 스트림으
 필요한 라이브러리를 설치한 후 혼합 8비트 모델을 가져오는 방법은 다음과 같습니다:
 
 ```py
-from transformers import AutoModelForCausalLM
+from transformers import AutoModelForCausalLM, BitsAndBytesConfig
 
 model_name = "bigscience/bloom-2b5"
-model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_8bit=True)
+model_8bit = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 ```
 
 텍스트 생성의 경우:
@@ -141,11 +141,11 @@ model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto",
 다음은 간단한 예입니다:
 
 ```py
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
 model_name = "bigscience/bloom-2b5"
 tokenizer = AutoTokenizer.from_pretrained(model_name)
-model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_8bit=True)
+model_8bit = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 
 prompt = "Hello, my llama is cute"
 inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
@@ -159,7 +159,7 @@ outputs = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)
 다중 GPU에서 혼합 8비트 모델을 로드하는 방법은 단일 GPU 설정과 동일합니다(동일한 명령어 사용):
 ```py
 model_name = "bigscience/bloom-2b5"
-model_8bit = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", load_in_8bit=True)
+model_8bit = AutoModelForCausalLM.from_pretrained(model_name, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 ```
 하지만 `accelerate`를 사용하여 각 GPU에 할당할 GPU RAM을 제어할 수 있습니다. 다음과 같이 `max_memory` 인수를 사용하세요:
 

docs/source/zh/main_classes/quantization.md

@@ -360,12 +360,12 @@ model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", load_i
 
 ```python
 # pip install transformers accelerate bitsandbytes
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
 model_id = "bigscience/bloom-1b7"
 
 tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", load_in_8bit=True)
+model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 ```
 
 然后，像通常使用 `PreTrainedModel` 一样使用您的模型。
@@ -441,9 +441,9 @@ model_double_quant = AutoModelForCausalLM.from_pretrained(model_id, quantization
 
 
 ```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
-model = AutoModelForCausalLM.from_pretrained("bigscience/bloom-560m", device_map="auto", load_in_8bit=True)
+model = AutoModelForCausalLM.from_pretrained("bigscience/bloom-560m", quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 tokenizer = AutoTokenizer.from_pretrained("bigscience/bloom-560m")
 
 model.push_to_hub("bloom-560m-8bit")

docs/source/zh/peft.md

@@ -86,10 +86,10 @@ model.load_adapter(peft_model_id)
 `bitsandbytes`集成支持8bit和4bit精度数据类型，这对于加载大模型非常有用，因为它可以节省内存（请参阅`bitsandbytes`[指南](./quantization#bitsandbytes-integration)以了解更多信息）。要有效地将模型分配到您的硬件，请在[`~PreTrainedModel.from_pretrained`]中添加`load_in_8bit`或`load_in_4bit`参数，并将`device_map="auto"`设置为：
 
 ```py
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
 peft_model_id = "ybelkada/opt-350m-lora"
-model = AutoModelForCausalLM.from_pretrained(peft_model_id, device_map="auto", load_in_8bit=True)
+model = AutoModelForCausalLM.from_pretrained(peft_model_id, quantization_config=BitsAndBytesConfig(load_in_8bit=True))
 ```
 
 ## 添加新的adapter