next ollama runner (#7913)

feat: add new Ollama engine using ggml through cgo

This change introduces a new way to run pretrained models. It introduces 3 high level interfaces and a bunch of smaller helper interfaces to facilitate this.

- `model.Model` defines the interface for a model architecture. Models such as `llama` and `mllama`, which are provided as examples, can implement the model's forward propagation in the `Forward` method. This method will be called to generate completions. This interface can be found in `model/model.go`
- `ml.Backend` defines the interface for a backend tensor library, in this case `ggml`. Among other things, a Backend is responsible for loading a pretrained model into hardware (GPU, CPU, etc) and providing an interface for Models to access loaded tensors. This interface can be found in `ml/backend.go`
- `ml.Tensor` defines the interface for a tensor and tensor operations

This is the first implementation of the new engine. Follow up PRs will implement more features:

- non-greedy sampling (#8410)
- integration with Ollama and KV caching (#8301)
- more model support (#9080) with more coming soon

Co-authored-by: Bruce MacDonald <brucewmacdonald@gmail.com>

ml/nn/convolution.go

@@ -0,0 +1,11 @@
+package nn
+
+import "github.com/ollama/ollama/ml"
+
+type Conv2D struct {
+	Weight ml.Tensor `gguf:"weight"`
+}
+
+func (m *Conv2D) Forward(ctx ml.Context, t ml.Tensor, s0, s1, p0, p1, d0, d1 int) ml.Tensor {
+	return m.Weight.Conv2D(ctx, t, s0, s1, p0, p1, d0, d1)
+}

ml/nn/embedding.go

@@ -0,0 +1,11 @@
+package nn
+
+import "github.com/ollama/ollama/ml"
+
+type Embedding struct {
+	Weight ml.Tensor `gguf:"weight"`
+}
+
+func (m *Embedding) Forward(ctx ml.Context, hiddenState ml.Tensor) ml.Tensor {
+	return m.Weight.Rows(ctx, hiddenState)
+}

ml/nn/linear.go

@@ -0,0 +1,17 @@
+package nn
+
+import "github.com/ollama/ollama/ml"
+
+type Linear struct {
+	Weight ml.Tensor `gguf:"weight"`
+	Bias   ml.Tensor `gguf:"bias"`
+}
+
+func (m *Linear) Forward(ctx ml.Context, t ml.Tensor) ml.Tensor {
+	t = m.Weight.Mulmat(ctx, t)
+	if m.Bias != nil {
+		t = t.Add(ctx, m.Bias)
+	}
+
+	return t
+}

ml/nn/normalization.go

@@ -0,0 +1,22 @@
+package nn
+
+import (
+	"github.com/ollama/ollama/ml"
+)
+
+type LayerNorm struct {
+	Weight ml.Tensor `gguf:"weight"`
+	Bias   ml.Tensor `gguf:"bias"`
+}
+
+func (m *LayerNorm) Forward(ctx ml.Context, t ml.Tensor, eps float32) ml.Tensor {
+	return t.LayerNorm(ctx, m.Weight, m.Bias, eps)
+}
+
+type RMSNorm struct {
+	Weight ml.Tensor `gguf:"weight"`
+}
+
+func (m *RMSNorm) Forward(ctx ml.Context, t ml.Tensor, eps float32) ml.Tensor {
+	return t.RMSNorm(ctx, m.Weight, eps)
+}