Instructions to use r0b0tlab/Hy3-295B-NVFP4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use r0b0tlab/Hy3-295B-NVFP4 with Transformers:

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

pipe = pipeline("text-generation", model="r0b0tlab/Hy3-295B-NVFP4")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("r0b0tlab/Hy3-295B-NVFP4")
model = AutoModelForCausalLM.from_pretrained("r0b0tlab/Hy3-295B-NVFP4")
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 Settings

vLLM

How to use r0b0tlab/Hy3-295B-NVFP4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "r0b0tlab/Hy3-295B-NVFP4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "r0b0tlab/Hy3-295B-NVFP4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/r0b0tlab/Hy3-295B-NVFP4

SGLang

How to use r0b0tlab/Hy3-295B-NVFP4 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 "r0b0tlab/Hy3-295B-NVFP4" \
    --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": "r0b0tlab/Hy3-295B-NVFP4",
		"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 "r0b0tlab/Hy3-295B-NVFP4" \
        --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": "r0b0tlab/Hy3-295B-NVFP4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use r0b0tlab/Hy3-295B-NVFP4 with Docker Model Runner:
```
docker model run hf.co/r0b0tlab/Hy3-295B-NVFP4
```

Hy3-295B-NVFP4

A community NVFP4-quantized version of Tencent's Hy3 295B Mixture-of-Experts model, optimized for dual-GB10 (NVIDIA DGX Spark) deployment.

295B total parameters - 21B active per token - NVFP4 W4A4 on routed experts

Quantization Method

Built with NVIDIA Model Optimizer 0.45.0. Shard-by-shard CPU conversion using NVFP4QTensor.quantize() with weight-derived amax scales. No full-model loading required.

Component	Precision	Details
Routed experts (layers 1-79, 192 experts/layer)	NVFP4 W4A4	group_size=16, per-block E4M3 scale, per-tensor FP32 scale_2
MTP / NextN predict (layer 80)	BF16	Preserved at original precision
Shared experts, attention, router, embeddings	BF16	Unchanged from source

File Size

Source	Compressed
605 GB (BF16)	~186 GB (NVFP4 mixed)

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "r0b0tlab/Hy3-295B-NVFP4",
    torch_dtype=torch.bfloat16,
    device_map="auto",
    trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained("r0b0tlab/Hy3-295B-NVFP4")

License

Same as the source model (Tencent Hy3).

Downloads last month: -

Safetensors

Model size

174B params

Tensor type

F8_E4M3

BF16

F32

Model tree for r0b0tlab/Hy3-295B-NVFP4

Base model

tencent/Hy3

Finetuned

(1)

this model