Instructions to use baidu/ERNIE-4.5-0.3B-PT with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use baidu/ERNIE-4.5-0.3B-PT with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="baidu/ERNIE-4.5-0.3B-PT")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("baidu/ERNIE-4.5-0.3B-PT")
model = AutoModelForCausalLM.from_pretrained("baidu/ERNIE-4.5-0.3B-PT", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use baidu/ERNIE-4.5-0.3B-PT with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "baidu/ERNIE-4.5-0.3B-PT"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "baidu/ERNIE-4.5-0.3B-PT",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/baidu/ERNIE-4.5-0.3B-PT

SGLang

How to use baidu/ERNIE-4.5-0.3B-PT with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "baidu/ERNIE-4.5-0.3B-PT" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "baidu/ERNIE-4.5-0.3B-PT",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "baidu/ERNIE-4.5-0.3B-PT" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "baidu/ERNIE-4.5-0.3B-PT",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use baidu/ERNIE-4.5-0.3B-PT with Docker Model Runner:
```
docker model run hf.co/baidu/ERNIE-4.5-0.3B-PT
```

Update config.json

by AntonV HF Staff - opened Jul 23, 2025

base: refs/heads/main

←

from: refs/pr/4

Discussion Files changed

-1

AntonV

Jul 23, 2025

See https://github.com/huggingface/transformers/blob/0fe03afeb82e1a435a75704d1f434c47e49a0bbb/tests/models/ernie4_5/test_modeling_ernie4_5.py#L103 for the correct class, sorry about that

Didn't use the auto class in the test

Update config.jsonb6da6157

hlfby06

BAIDU org Jul 25, 2025

Exactly, Ernie4_5ForCausalLM is more suitable.

By the way, if we only want to use LlamaTokenizerFast, we can remove "add_prefix_space": false, from tokenizer_config.json and tokenizer.model under the root directory. Could you confirm this?
Furthermore, are there any functional differences between LlamaTokenizerFast and LlamaTokenizer?

AntonV

Jul 25, 2025

By the way, if we only want to use LlamaTokenizerFast, we can remove "add_prefix_space": false, from tokenizer_config.json and tokenizer.model under the root directory.

Not entirely sure there tbh but iirc there were slight difference between False and True so I opted for False to ensure same behavior as before.

Furthermore, are there any functional differences between LlamaTokenizerFast and LlamaTokenizer?

Unfortunately yes and that's why we opted to only support the fast variant as it aligned functionality-wise without any further changes. The "slow" tokenizer had slight deviations on how it behaved around special tokens, i.e. it would need patches like https://huggingface.co/baidu/ERNIE-4.5-0.3B-PT/commit/00fbf72f526289cdb3e1ddb14eb74adf8665c56d - this is mostly due to legacy behavior tbh.

AntonV

Jul 29, 2025

gentle ping @hlfby06

I think we can merge this and https://huggingface.co/baidu/ERNIE-4.5-0.3B-Base-PT/discussions/4 now as vLLM has been updated (see https://github.com/vllm-project/vllm/pull/21735)

hmellor

Aug 7, 2025

I've just found this PR, @hlfby06 please can it be merged ASAP?

vLLM has updated to the latest version of Transformers and this model no longer runs in vLLM because the checkpoint contains the old architecture name.

WYF3634076 changed pull request status to merged Aug 7, 2025

hmellor

Aug 7, 2025

Thanks!

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment