Instructions to use entropy/roberta_zinc_decoder with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use entropy/roberta_zinc_decoder with Transformers:

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

pipe = pipeline("text-generation", model="entropy/roberta_zinc_decoder", trust_remote_code=True)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("entropy/roberta_zinc_decoder", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("entropy/roberta_zinc_decoder", trust_remote_code=True)

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use entropy/roberta_zinc_decoder with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "entropy/roberta_zinc_decoder"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "entropy/roberta_zinc_decoder",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/entropy/roberta_zinc_decoder

SGLang

How to use entropy/roberta_zinc_decoder 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 "entropy/roberta_zinc_decoder" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "entropy/roberta_zinc_decoder",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "entropy/roberta_zinc_decoder" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "entropy/roberta_zinc_decoder",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use entropy/roberta_zinc_decoder with Docker Model Runner:
```
docker model run hf.co/entropy/roberta_zinc_decoder
```

entropy commited on Sep 18, 2023

Commit

b2975b7

1 Parent(s): c889db3

Update README.md

Browse files

Files changed (1) hide show

README.md +26 -1

README.md CHANGED Viewed

@@ -8,7 +8,8 @@ tags:
 # Roberta Zinc Decoder
 This model is a GPT2 decoder model designed to reconstruct SMILES strings from embeddings created by the
-[roberta_zinc_480m](https://huggingface.co/entropy/roberta_zinc_480m) model.
 The decoder model conditions generation on mean pooled embeddings from the encoder model. Mean pooled
 embeddings are used to allow for integration with vector databases, which require fixed length embeddings.
@@ -62,6 +63,30 @@ gen = decoder_model.generate(
 reconstructed_smiles = tokenizer.batch_decode(gen, skip_special_tokens=True)
 ```
 ---
 license: mit
 ---

 # Roberta Zinc Decoder
 This model is a GPT2 decoder model designed to reconstruct SMILES strings from embeddings created by the
+[roberta_zinc_480m](https://huggingface.co/entropy/roberta_zinc_480m) model. The decoder model was
+trained on 30m compounds from the [ZINC Database](https://zinc.docking.org/).
 The decoder model conditions generation on mean pooled embeddings from the encoder model. Mean pooled
 embeddings are used to allow for integration with vector databases, which require fixed length embeddings.
 reconstructed_smiles = tokenizer.batch_decode(gen, skip_special_tokens=True)
 ```
+## Model Performance
+The decoder model was evaluated on a test set of 1m compounds from ZINC. Compounds
+were encoded with the [roberta_zinc_480m](https://huggingface.co/entropy/roberta_zinc_480m) model
+and reconstructed with the decoder model.
+The following metrics are computed:
+* `exact_match` - percent of inputs exactly reconstructed
+* `token_accuracy` - percent of output tokens exactly matching input tokens (excluding padding)
+* `valid_structure` - percent of generated outputs that resolved to a valid SMILES string
+* `tanimoto` - tanimoto similarity between inputs and generated outputs. Excludes invalid structures
+* `cos_sim` - cosine similarity between input encoder embeddings and output encoder embeddings
+`eval_type=full` reports metrics for the full 1m compound test set.
+`eval_type=failed` subsets metrics for generated outputs that failed to exactly replicate the inputs.
+|eval_type|exact_match|token_accuracy|valid_structure|tanimoto|cos_sim |
+|---------|-----------|--------------|---------------|--------|--------|
+|full     |0.948277   |0.990704      |0.994278       |0.987698|0.998224|
+|failed   |0.000000   |0.820293      |0.889372       |0.734097|0.965668|
 ---
 license: mit
 ---