Instructions to use euclaise/ReMask-3B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use euclaise/ReMask-3B with Transformers:

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

pipe = pipeline("text-generation", model="euclaise/ReMask-3B")
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("euclaise/ReMask-3B")
model = AutoModelForCausalLM.from_pretrained("euclaise/ReMask-3B", 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 euclaise/ReMask-3B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "euclaise/ReMask-3B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "euclaise/ReMask-3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/euclaise/ReMask-3B

SGLang

How to use euclaise/ReMask-3B 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 "euclaise/ReMask-3B" \
    --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": "euclaise/ReMask-3B",
		"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 "euclaise/ReMask-3B" \
        --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": "euclaise/ReMask-3B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use euclaise/ReMask-3B with Docker Model Runner:
```
docker model run hf.co/euclaise/ReMask-3B
```

euclaise commited on Mar 28, 2024

Commit

febd59e

verified ·

1 Parent(s): c404867

Update README.md

Browse files

Files changed (1) hide show

README.md +1 -1

README.md CHANGED Viewed

@@ -68,7 +68,7 @@ Consider the following chat interaction:
 The model must predict the bolded parts.  So, we randomly mask tokens from the bolded parts, and run the model once on the masked sequence and once on the full sequence.
-We then compute a distance loss `D(p_masked, p_full)` between the two predictions. This approach resembles self-distillation, and MSE tends to perform better than KL Divergence for distillation, along with being easier to tune, so I went with MSE.
 Finally, we add this loss to the standard cross-entropy language modeling losses from each prediction, with a weighting value:
 ```

 The model must predict the bolded parts.  So, we randomly mask tokens from the bolded parts, and run the model once on the masked sequence and once on the full sequence.
+We then compute a distance loss `D(p_masked, p_full)` between the two predictions. This approach resembles self-distillation, and MSE tends to perform better than KL Divergence for distillation, along with being easier to tune, so I went with MSE (note that R-TeaFor uses a mix of reverse and forward KL divergence).
 Finally, we add this loss to the standard cross-entropy language modeling losses from each prediction, with a weighting value:
 ```