highperfocused/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-04-12 21:59:22 +02:00
Code Issues Packages Projects Releases Wiki Activity

ml: Remove Output from Context interface

Browse Source 
Model implementations should use Input for all of their tensors
supplied to the model. This includes tensors that relate to the
outputs, which is confusing since there is also an Output funciton.

Since Output is only used internally in GGML and not used by any
model implementations, we can remove it from the interface to
reduce confusion.
This commit is contained in:
Jesse Gross 2025-03-27 11:52:09 -07:00 committed by Jesse Gross
parent ead27aa9fe
commit 01aa788722
3 changed files with 3 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
kvcache/causal_test.go
View File

@ -362,7 +362,6 @@ func (c *testContext) FromIntSlice(s []int32, shape ...int) (ml.Tensor, error) {
}
func (c *testContext) Input() ml.Context { return c }
func (c *testContext) Output() ml.Context { return c }
func (c *testContext) Layer(int) ml.Context { return c }
func (c *testContext) Forward(...ml.Tensor) ml.Context { return c }

6
ml/backend.go
View File

@ -110,12 +110,10 @@ type Context interface {
MaxGraphNodes() int
Close()
// Input returns a context appropriate for creating input tensors
// Input returns a context appropriate for creating tensors that are
// inputs to the model (which includes things like output locations)
Input() Context
// Output returns a context appropriate for creating output tensors
Output() Context
// Layer returns a context appropriate for creating intermediate tensors
Layer(int) Context
}

19
ml/backend/ggml/ggml.go
View File

@ -48,9 +48,6 @@ type Backend struct {
// input is the backend used for inputs
input *C.struct_ggml_backend_buffer_type
// output is the backend used for outputs
output *C.struct_ggml_backend_buffer_type
// layers is the backend used for repeating layers
layers map[int]*C.struct_ggml_backend_buffer_type
@ -400,8 +397,7 @@ func New(ctx context.Context, r *os.File, params ml.BackendParams) (ml.Backend,
C.size_t(maxGraphNodes),
C._Bool(len(gpus) > 1 && slices.Contains(gpus, output.d)),
),
input: deviceBufferTypes[input.d],
output: deviceBufferTypes[output.d],
input: deviceBufferTypes[input.d],
layers: func() map[int]*C.struct_ggml_backend_buffer_type {
m := make(map[int]*C.struct_ggml_backend_buffer_type)
for i, layer := range layers {
@ -482,19 +478,6 @@ func (c Context) Input() ml.Context {
return &c
}
func (c Context) Output() ml.Context {
if c.b.output != nil {
return &Context{
b: c.b,
ctx: c.ctx,
buft: c.b.output,
maxGraphNodes: c.maxGraphNodes,
}
}
return &c
}
func (c Context) Layer(i int) ml.Context {
if buft, ok := c.b.layers[i]; ok {
return &Context{