Instructions to use beyoru/VibeThinker-3B-FP8-block with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use beyoru/VibeThinker-3B-FP8-block with Transformers:

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

pipe = pipeline("text-generation", model="beyoru/VibeThinker-3B-FP8-block")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("beyoru/VibeThinker-3B-FP8-block")
model = AutoModelForMultimodalLM.from_pretrained("beyoru/VibeThinker-3B-FP8-block")
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 beyoru/VibeThinker-3B-FP8-block with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "beyoru/VibeThinker-3B-FP8-block"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "beyoru/VibeThinker-3B-FP8-block",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/beyoru/VibeThinker-3B-FP8-block

SGLang

How to use beyoru/VibeThinker-3B-FP8-block 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 "beyoru/VibeThinker-3B-FP8-block" \
    --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": "beyoru/VibeThinker-3B-FP8-block",
		"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 "beyoru/VibeThinker-3B-FP8-block" \
        --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": "beyoru/VibeThinker-3B-FP8-block",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use beyoru/VibeThinker-3B-FP8-block with Docker Model Runner:
```
docker model run hf.co/beyoru/VibeThinker-3B-FP8-block
```

VibeThinker-3B-FP8-block

FP8 (block-wise) quantization of a VibeThinker-3B / Qwen2.5-3B reasoning model, ready for inference with vLLM on NVIDIA Hopper/Ada GPUs (H100/H200/L40/4090).

Quantized by Hert4.

What this is

Weights and activations are quantized from BF16 to FP8 (E4M3) using LLM Compressor with the FP8_BLOCK scheme:


Weights	FP8 E4M3, block-wise scaling, `block_structure = [128, 128]`
Activations	FP8 E4M3, dynamic, per-group `group_size = 128`
Kept in original precision	`lm_head`, `embed_tokens`
Format	`compressed-tensors` (`float-quantized`)
Size	~3.2 GB (≈50% of BF16)

This is the same recipe used by DeepSeek-V3-style FP8 checkpoints and pairs with DeepGEMM block-FP8 kernels. The checkpoint is self-contained — the original BF16 model is not required at runtime.

Note: derived from an abliterated (uncensored) VibeThinker-3B finetune — safety behavior differs from the original base. Use responsibly; you are responsible for outputs.

Usage (vLLM)

vllm serve <your-namespace>/VibeThinker-3B-FP8-block \
  --tensor-parallel-size 1 \
  --max-model-len 32768

vLLM auto-detects the FP8 quantization from config.json — do not pass --quantization.

Creation

from llmcompressor import model_free_ptq

model_free_ptq(
    model_stub="<source-bf16-model>",
    save_directory="VibeThinker-3B-FP8-block",
    scheme="FP8_BLOCK",
    ignore=["lm_head", "re:.*embed_tokens.*"],
    device="cpu",
)

Credits

Base lineage: WeiboAI/VibeThinker-3B → Qwen/Qwen2.5-3B
Quantization toolkit: vllm-project/llm-compressor

Downloads last month: 12

Safetensors

Model size

3B params

Tensor type

BF16

F8_E4M3

Model tree for beyoru/VibeThinker-3B-FP8-block

Base model

Qwen/Qwen2.5-3B

Finetuned

Qwen/Qwen2.5-Coder-3B

Finetuned

WeiboAI/VibeThinker-3B

Quantized

(45)

this model