Instructions to use HuggingFaceBio/Carbon-X-Base with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use HuggingFaceBio/Carbon-X-Base with Transformers:

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

pipe = pipeline("text-generation", model="HuggingFaceBio/Carbon-X-Base")
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("HuggingFaceBio/Carbon-X-Base")
model = AutoModelForCausalLM.from_pretrained("HuggingFaceBio/Carbon-X-Base", 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 HuggingFaceBio/Carbon-X-Base with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/HuggingFaceBio/Carbon-X-Base

SGLang

How to use HuggingFaceBio/Carbon-X-Base 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 "HuggingFaceBio/Carbon-X-Base" \
    --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": "HuggingFaceBio/Carbon-X-Base",
		"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 "HuggingFaceBio/Carbon-X-Base" \
        --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": "HuggingFaceBio/Carbon-X-Base",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use HuggingFaceBio/Carbon-X-Base with Docker Model Runner:
```
docker model run hf.co/HuggingFaceBio/Carbon-X-Base
```

Use standard tokenizer serialization to preserve configuration

by edbeeching HF Staff - opened 12 days ago

base: refs/heads/main

←

from: refs/pr/1

Discussion Files changed

+56

-67

edbeeching

Hugging Face Biology Research org 12 days ago

HybridDNATokenizer currently saves the underlying BPE tokenizer's configuration and manually restores only a few fields. Consequently, changing EOS to <|im_end|> is lost after saving/reloading, and other standard tokenizer settings can be dropped too.

This change uses the standard Transformers persistence path used by Qwen:

Inherit save_pretrained() for the wrapper's configuration, special-token map, AddedToken metadata, chat template, and custom-code export.
Delegate only BPE asset writing through _save_pretrained(), retaining dna_config.json and the existing DNA vocabulary-layout checks.
Register k, tail, and auto_dna_tags as constructor configuration, and delegate loading to the parent implementation. Existing bio_config.json / dna_config.json options remain fallback defaults for older checkpoints.
Preserve restored added-token metadata instead of overwriting it with base-tokenizer defaults. Explicit caller overrides, including eos_token=None / pad_token=None, retain standard Transformers semantics.

DNA encoding and vocabulary IDs are unchanged. This PR changes only tokenizer.py; it does not change model weights or the saved model configuration.

Reproduction on current main

from transformers import AutoTokenizer

tok = AutoTokenizer.from_pretrained("HuggingFaceBio/Carbon-X-Base", trust_remote_code=True)
tok.eos_token = "<|im_end|>"
tok.save_pretrained("./checkpoint")

tok = AutoTokenizer.from_pretrained("./checkpoint", trust_remote_code=True)
print(tok.eos_token)  # Currently <|endoftext|>; should remain <|im_end|>.

Validation

Eight tokenizer-only tests passed with both Transformers 5.8.1 and 5.12.1:

Existing Base checkpoint compatibility and unchanged vocabulary / representative DNA token IDs.
EOS, PAD, BOS, UNK, configured padding/truncation sides, model length, cleanup settings, and chat-template persistence.
AddedToken flags and extra-special-token metadata.
Explicit reload overrides and nullable EOS/PAD settings.
DNA constructor options across multiple k-mer sizes and tail modes.
Legacy DNA configuration fallbacks and vocabulary-layout mismatch rejection.
Independent AutoTokenizer loading and re-saving in a fresh process, without importing the Carbon package.

After this upstream fix is merged, the temporary tokenizer copy in Carbon SFT PR #33 can be removed.

Use standard tokenizer serialization to preserve configuratione81b8c10

Use double quotes for special-token defaultsfe5eb686

edbeeching changed pull request status to closed 11 days ago

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