Instructions to use 0xSero/Hy3-299B-FP8 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use 0xSero/Hy3-299B-FP8 with Transformers:

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

pipe = pipeline("text-generation", model="0xSero/Hy3-299B-FP8")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("0xSero/Hy3-299B-FP8")
model = AutoModelForCausalLM.from_pretrained("0xSero/Hy3-299B-FP8")
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=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use 0xSero/Hy3-299B-FP8 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "0xSero/Hy3-299B-FP8"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "0xSero/Hy3-299B-FP8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/0xSero/Hy3-299B-FP8

SGLang

How to use 0xSero/Hy3-299B-FP8 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 "0xSero/Hy3-299B-FP8" \
    --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": "0xSero/Hy3-299B-FP8",
		"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 "0xSero/Hy3-299B-FP8" \
        --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": "0xSero/Hy3-299B-FP8",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use 0xSero/Hy3-299B-FP8 with Docker Model Runner:
```
docker model run hf.co/0xSero/Hy3-299B-FP8
```

Support this work → · X · GitHub · REAP paper · Cerebras REAP

Hy3-299B-FP8

FP8 quantization of tencent/Hy3-preview.

At a glance


Base model	tencent/Hy3-preview
Format	FP8
Total params	299B
Active / token	—
Experts / layer	192
Layers	80
Hidden size	4096
Context	262,144
On-disk size	300 GB

Which variant should I pick?

Variant	Format	Link
`Hy3-299B-FP8` (this)	FP8	link
`Hy3-299B-NVFP4`	NVFP4	link

This is a checkpoint-only FP8_BLOCK quantization of tencent/Hy3-preview, produced with llmcompressor.entrypoints.model_free.model_free_ptq.

Base model: tencent/Hy3-preview
Quantization scheme: FP8_BLOCK
Ignored modules/patterns: lm_head, model.embed_tokens, re:.*router.gate$, re:.*expert_bias$
Source snapshot: recorded in QUANTIZATION_MANIFEST.json
License: inherits Tencent Hy Community License Agreement from the base model; original LICENSE is included.

Notes

This release quantizes safetensors weights without importing the custom HYV3 model class. Router gates, expert bias tensors, embeddings, and lm_head are preserved unquantized for compatibility/conservatism.

License & citation

License inherited from the base model.

@misc{lasby2025reap,
  title  = {REAP the Experts: Why Pruning Prevails for One-Shot MoE Compression},
  author = {Mike Lasby and Ivan Lazarevich and Nish Sinnadurai and Sean Lie and Yani Ioannou and Vithursan Thangarasa},
  year   = {2025}, eprint = {2510.13999}, archivePrefix = {arXiv}
}

Model tree for 0xSero/Hy3-299B-FP8

Base model

tencent/Hy3-preview

Quantized

(10)

this model

Collection including 0xSero/Hy3-299B-FP8

Other models

Collection

Hy3, Kimi-K2.5, INTELLECT-3, NousCoder. • 8 items • Updated 1 day ago

Paper for 0xSero/Hy3-299B-FP8

REAP the Experts: Why Pruning Prevails for One-Shot MoE compression

Paper • 2510.13999 • Published Oct 15, 2025 • 20

0xSero
/

Hy3-299B-FP8