Transformers Patch¶

This module patches HuggingFace Transformers to work with MindSpore as the backend.

Overview¶

When you import mindnlp, it automatically patches the transformers library to use MindSpore operations instead of PyTorch. This enables you to use all HuggingFace models directly.

Usage¶

import mindspore
import mindnlp  # Patches are applied automatically

# Now use transformers as usual
from transformers import AutoModel, AutoTokenizer, pipeline

# Load any model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
model = AutoModel.from_pretrained("bert-base-uncased")

# Use pipelines
pipe = pipeline("text-generation", model="Qwen/Qwen2-0.5B", ms_dtype=mindspore.float16)

Key Features¶

Model Loading¶

All from_pretrained() methods work seamlessly:

from transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen2-0.5B",
    ms_dtype=mindspore.bfloat16,  # MindSpore dtype
    device_map="auto"
)

Pipelines¶

All transformers pipelines are supported:

text-generation
text-classification
token-classification
question-answering
fill-mask
summarization
translation
And more...

Special Parameters¶

MindNLP adds support for MindSpore-specific parameters:

ms_dtype: Specify MindSpore data type (e.g., mindspore.float16, mindspore.bfloat16)
device_map: Device placement strategy ("auto", "cuda", "cpu")

Patched Components¶

The patch system modifies:

Model classes (Auto*, *Model, *ForCausalLM, etc.)
Tokenizers
Pipelines
Generation utilities
Configuration classes

Notes¶

For training, consider using mindnlp.engine.Trainer for optimal gradient computation
The patch is automatically applied on import mindnlp
All 200,000+ models on HuggingFace Hub are supported