Instructions to use WaveCut/YuE2-3B-OrbitQuant-W4A4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use WaveCut/YuE2-3B-OrbitQuant-W4A4 with Transformers:

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

pipe = pipeline("text-generation", model="WaveCut/YuE2-3B-OrbitQuant-W4A4")

# Load model directly
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained("WaveCut/YuE2-3B-OrbitQuant-W4A4", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use WaveCut/YuE2-3B-OrbitQuant-W4A4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "WaveCut/YuE2-3B-OrbitQuant-W4A4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "WaveCut/YuE2-3B-OrbitQuant-W4A4",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/WaveCut/YuE2-3B-OrbitQuant-W4A4

SGLang

How to use WaveCut/YuE2-3B-OrbitQuant-W4A4 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 "WaveCut/YuE2-3B-OrbitQuant-W4A4" \
    --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": "WaveCut/YuE2-3B-OrbitQuant-W4A4",
		"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 "WaveCut/YuE2-3B-OrbitQuant-W4A4" \
        --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": "WaveCut/YuE2-3B-OrbitQuant-W4A4",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use WaveCut/YuE2-3B-OrbitQuant-W4A4 with Docker Model Runner:
```
docker model run hf.co/WaveCut/YuE2-3B-OrbitQuant-W4A4
```

YuE2-3B-OrbitQuant-W4A4

File size: 2,767 Bytes

9a2611f

#include <torch/library.h>
#include "registration.h"
#include "torch_binding.h"
TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
  ops.def("rmsquant(Tensor! packed, Tensor! norms, Tensor x, Tensor weight, Tensor permutation, Tensor signs, Tensor boundaries, float rms_eps, float quant_eps) -> ()");
  ops.impl("rmsquant", torch::kCUDA, &rmsquant);
  ops.def("orbitquant_gemv(Tensor! out, Tensor x, Tensor w, Tensor xn, Tensor wn, Tensor ac, Tensor wc, Tensor bias, bool has_bias, float activation_scale, float weight_scale) -> ()");
  ops.impl("orbitquant_gemv", torch::kCUDA, &orbitquant_gemv);
  ops.def("yue2_rmsnorm(Tensor! out, Tensor x, Tensor weight, float eps) -> ()");
  ops.impl("yue2_rmsnorm", torch::kCUDA, &yue2_rmsnorm);
  ops.def("yue2_rope(Tensor! out, Tensor x, Tensor cos, Tensor sin, int heads) -> ()");
  ops.impl("yue2_rope", torch::kCUDA, &yue2_rope);
  ops.def("yue2_swiglu(Tensor! out, Tensor x) -> ()");
  ops.impl("yue2_swiglu", torch::kCUDA, &yue2_swiglu);
  ops.def("yue2_qkv_decode(Tensor! q, Tensor! keys, Tensor! values, Tensor x, Tensor qw, Tensor kw, Tensor cos, Tensor sin, Tensor positions, float eps) -> ()");
  ops.impl("yue2_qkv_decode", torch::kCUDA, &yue2_qkv_decode);
  ops.def("yue2_rmsnorm_exact(Tensor! out, Tensor x, Tensor weight, float eps) -> ()");
  ops.impl("yue2_rmsnorm_exact", torch::kCUDA, &yue2_rmsnorm_exact);
  ops.def("fused_gemv(Tensor! out, Tensor x, Tensor gamma, bool has_gamma, Tensor permutation, Tensor signs, Tensor boundaries, float rms_eps, float quant_eps, float inv_sqrt_block, Tensor w, Tensor wn, Tensor ac, Tensor wc, Tensor bias, bool has_bias, float activation_scale, float weight_scale, int virtual_threads) -> ()");
  ops.impl("fused_gemv", torch::kCUDA, &fused_gemv);
  ops.def("fused_gemv2(Tensor! out, Tensor! x_new, bool write_x_new, Tensor x, Tensor h, bool has_h, Tensor gamma, bool has_gamma, Tensor permutation, Tensor signs, Tensor boundaries, float rms_eps, float quant_eps, float inv_sqrt_block, Tensor w, Tensor wn, Tensor ac, Tensor wc, Tensor bias, bool has_bias, float activation_scale, float weight_scale, int virtual_threads) -> ()");
  ops.impl("fused_gemv2", torch::kCUDA, &fused_gemv2);
  ops.def("rmsquant_add(Tensor! packed, Tensor! norms, Tensor! x_new, Tensor x, Tensor h, Tensor weight, Tensor permutation, Tensor signs, Tensor boundaries, float rms_eps, float quant_eps) -> ()");
  ops.impl("rmsquant_add", torch::kCUDA, &rmsquant_add);
  ops.def("quantize_rows_int8(Tensor! out, Tensor! scales, Tensor x) -> ()");
  ops.impl("quantize_rows_int8", torch::kCUDA, &quantize_rows_int8);
  ops.def("gemv_w8a8(Tensor! out, Tensor x, Tensor xs, Tensor w, Tensor ws) -> ()");
  ops.impl("gemv_w8a8", torch::kCUDA, &gemv_w8a8);
}
REGISTER_EXTENSION(TORCH_EXTENSION_NAME)