GGBond stands for GGML Bounding: an object-oriented Python wrapper over GGML, written in pure Python on top of ggml-python.
It has two layers. ggbond is a small high-level API where a Session owns the devices and every resource. ggbond.ggml maps GGML's own concepts (Context, Backend, GAllocr, Scheduler, GGUF, …) 1:1. Both keep GGML's op names, argument order and shape order.
pip install -e . # or: uv syncRequires Python 3.10+. There is no build step; CPU always works, and GPU backends depend on how your ggml-python was built.
import numpy as np
from ggbond import Session
a = np.array([[2, 8], [5, 1], [4, 2], [8, 6]], dtype=np.float32)
b = np.array([[10, 5], [9, 9], [5, 4]], dtype=np.float32)
with Session() as sess, sess.graph() as g: # best available device
ta = g.input(a, name="a")
tb = g.input(b, name="b")
result = g.run(ta.mul_mat(tb)) # allocate, upload, compute, download
np.testing.assert_allclose(result, b @ a.T)The same computation with every GGML step spelled out:
import numpy as np
from ggbond.ggml import Backend, Context, F32, GAllocr
a = np.array([[2, 8], [5, 1], [4, 2], [8, 6]], dtype=np.float32)
b = np.array([[10, 5], [9, 9], [5, 4]], dtype=np.float32)
with Backend.cpu_init() as backend, Context(1 << 20, no_alloc=True) as ctx:
ta = ctx.new_tensor_2d(F32, *reversed(a.shape), name="a").set_input()
tb = ctx.new_tensor_2d(F32, *reversed(b.shape), name="b").set_input()
out = ta.mul_mat(tb).set_output()
graph = ctx.new_graph().build_forward_expand(out)
with GAllocr.from_backend(backend) as allocator:
allocator.alloc_graph(graph)
ta.write(a)
tb.write(b)
backend.compute(graph)
result = out.to_numpy()
np.testing.assert_allclose(result, b @ a.T)Tensor dimensions are in GGML order, the reverse of a NumPy shape, which is why ta.mul_mat(tb) is b @ a.T.
examples/ has matrix multiplication, GPT-2, Magika and NAM inference on the high-level API, and examples/ggml/ has the same programs on the 1:1 layer. Model files are not included.
MIT
