ollama/model/model_test.go
Michael Yang 58245413f4
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>
2025-02-13 16:31:21 -08:00

137 lines
3.1 KiB
Go

package model
import (
"reflect"
"slices"
"testing"
"github.com/google/go-cmp/cmp"
"github.com/ollama/ollama/ml"
"github.com/ollama/ollama/ml/backend/ggml"
"github.com/ollama/ollama/ml/nn"
)
func TestParseTags(t *testing.T) {
cases := []struct {
value string
want Tag
}{
{
value: "output",
want: Tag{
Name: "output",
},
},
{
value: "output,alt:token_embd",
want: Tag{
Name: "output",
Alternate: []string{
"token_embd",
},
},
},
}
for _, tt := range cases {
t.Run(tt.value, func(t *testing.T) {
got := ParseTags(tt.value)
if diff := cmp.Diff(tt.want, got); diff != "" {
t.Errorf("ParseTags() returned unexpected values (-want +got):\n%s", diff)
}
})
}
}
type fakeBackend struct {
*ggml.Backend
names []string
}
type fakeTensor struct {
*ggml.Tensor
Name string
}
func (m *fakeBackend) Get(name string) ml.Tensor {
if slices.Contains(m.names, name) {
return &fakeTensor{Name: name}
}
return nil
}
func TestPopulateFields(t *testing.T) {
type fakeLayer struct {
Query *nn.Linear `gguf:"attn_q"`
Key *nn.Linear `gguf:"attn_k"`
Value *nn.Linear `gguf:"attn_v"`
Output *nn.Linear `gguf:"attn_o"`
}
type fakeModel struct {
Input *nn.Embedding `gguf:"input"`
OutputNorm *nn.RMSNorm `gguf:"output_norm"`
Output *nn.Linear `gguf:"output"`
Layers [2]fakeLayer `gguf:"blk"`
}
var m fakeModel
v := reflect.ValueOf(&m)
v.Elem().Set(populateFields(&fakeBackend{
names: []string{
"input.weight",
"blk.0.attn_q.weight",
"blk.0.attn_k.weight",
"blk.0.attn_v.weight",
"blk.1.attn_q.weight",
"blk.1.attn_k.weight",
"blk.1.attn_v.weight",
"output_norm.weight",
"output.weight",
},
}, v))
if diff := cmp.Diff(fakeModel{
Input: &nn.Embedding{Weight: &fakeTensor{Name: "input.weight"}},
OutputNorm: &nn.RMSNorm{Weight: &fakeTensor{Name: "output_norm.weight"}},
Output: &nn.Linear{Weight: &fakeTensor{Name: "output.weight"}},
Layers: [2]fakeLayer{
{
Query: &nn.Linear{Weight: &fakeTensor{Name: "blk.0.attn_q.weight"}},
Key: &nn.Linear{Weight: &fakeTensor{Name: "blk.0.attn_k.weight"}},
Value: &nn.Linear{Weight: &fakeTensor{Name: "blk.0.attn_v.weight"}},
},
{
Query: &nn.Linear{Weight: &fakeTensor{Name: "blk.1.attn_q.weight"}},
Key: &nn.Linear{Weight: &fakeTensor{Name: "blk.1.attn_k.weight"}},
Value: &nn.Linear{Weight: &fakeTensor{Name: "blk.1.attn_v.weight"}},
},
},
}, m); diff != "" {
t.Errorf("populateFields() set incorrect values (-want +got):\n%s", diff)
}
}
func TestPopulateFieldsAlternateName(t *testing.T) {
type fakeModel struct {
Input *nn.Embedding `gguf:"input"`
Output *nn.Linear `gguf:"output,alt:input"`
}
m := fakeModel{}
v := reflect.ValueOf(&m)
v.Elem().Set(populateFields(&fakeBackend{
names: []string{
"input.weight",
},
}, v))
if diff := cmp.Diff(fakeModel{
Input: &nn.Embedding{Weight: &fakeTensor{Name: "input.weight"}},
Output: &nn.Linear{Weight: &fakeTensor{Name: "input.weight"}},
}, m); diff != "" {
t.Errorf("populateFields() set incorrect values (-want +got):\n%s", diff)
}
}