matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-10 07:46:59 +00:00
Code Issues Packages Projects Releases Wiki Activity

next ollama runner (#7913)

Browse Source 
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>
This commit is contained in:
Michael Yang
2025-02-14 00:31:21 +00:00
committed by GitHub
parent 8cf16063a5
commit 58245413f4
57 changed files with 475427 additions and 494 deletions
Download Patch File Download Diff File Expand all files Collapse all files

159
fs/ggml/ggml_test.go Normal file
View File

@@ -0,0 +1,159 @@
package ggml
import (
"maps"
"slices"
"strings"
"testing"
"github.com/google/go-cmp/cmp"
)
func TestTensorLayers(t *testing.T) {
tensors := make(map[string]*Tensor)
for _, name := range []string{
"token_embd.weight",
"blk.0.attn_k.weight",
"blk.0.attn_output.weight",
"blk.0.attn_q.weight",
"blk.0.attn_v.weight",
"blk.0.attn_norm.weight",
"blk.0.ffn_down.weight",
"blk.0.ffn_gate.weight",
"blk.0.ffn_up.weight",
"blk.0.ffn_norm.weight",
"output_norm.weight",
"mm.0.bias",
"mm.0.weight",
"v.blk.0.attn_k.weight",
"v.blk.0.attn_output.weight",
"v.blk.0.attn_q.weight",
"v.blk.0.attn_v.weight",
"v.blk.0.attn_norm.weight",
"v.blk.0.ffn_down.weight",
"v.blk.0.ffn_gate.weight",
"v.blk.0.ffn_up.weight",
"v.blk.0.ffn_norm.weight",
"v.patch_embd.weight",
"v.position_embd.gate",
"v.position_embd.weight",
} {
tensors[name] = &Tensor{Name: name}
}
cases := []struct {
name string
items []*Tensor
want map[string]Layer
}{
{
name: "text",
items: slices.Collect(func(yield func(*Tensor) bool) {
for k, v := range tensors {
if !strings.HasPrefix(k, "mm.") && !strings.HasPrefix(k, "v.") {
if !yield(v) {
return
}
}
}
}),
want: map[string]Layer{
"blk.0": {
"attn_k.weight": tensors["blk.0.attn_k.weight"],
"attn_q.weight": tensors["blk.0.attn_q.weight"],
"attn_v.weight": tensors["blk.0.attn_v.weight"],
"attn_output.weight": tensors["blk.0.attn_output.weight"],
"attn_norm.weight": tensors["blk.0.attn_norm.weight"],
"ffn_down.weight": tensors["blk.0.ffn_down.weight"],
"ffn_gate.weight": tensors["blk.0.ffn_gate.weight"],
"ffn_up.weight": tensors["blk.0.ffn_up.weight"],
"ffn_norm.weight": tensors["blk.0.ffn_norm.weight"],
},
"token_embd": {"weight": tensors["token_embd.weight"]},
"output_norm": {"weight": tensors["output_norm.weight"]},
},
},
{
name: "vision",
items: slices.Collect(func(yield func(*Tensor) bool) {
for k, v := range tensors {
if strings.HasPrefix(k, "mm.") || strings.HasPrefix(k, "v.") {
if !yield(v) {
return
}
}
}
}),
want: map[string]Layer{
"mm.0": {
"bias": tensors["mm.0.bias"],
"weight": tensors["mm.0.weight"],
},
"v.blk.0": {
"attn_k.weight": tensors["v.blk.0.attn_k.weight"],
"attn_q.weight": tensors["v.blk.0.attn_q.weight"],
"attn_v.weight": tensors["v.blk.0.attn_v.weight"],
"attn_output.weight": tensors["v.blk.0.attn_output.weight"],
"attn_norm.weight": tensors["v.blk.0.attn_norm.weight"],
"ffn_down.weight": tensors["v.blk.0.ffn_down.weight"],
"ffn_gate.weight": tensors["v.blk.0.ffn_gate.weight"],
"ffn_up.weight": tensors["v.blk.0.ffn_up.weight"],
"ffn_norm.weight": tensors["v.blk.0.ffn_norm.weight"],
},
"v": {
"patch_embd.weight": tensors["v.patch_embd.weight"],
"position_embd.gate": tensors["v.position_embd.gate"],
"position_embd.weight": tensors["v.position_embd.weight"],
},
},
},
{
name: "vision and text",
items: slices.Collect(maps.Values(tensors)),
want: map[string]Layer{
"blk.0": {
"attn_k.weight": tensors["blk.0.attn_k.weight"],
"attn_q.weight": tensors["blk.0.attn_q.weight"],
"attn_v.weight": tensors["blk.0.attn_v.weight"],
"attn_output.weight": tensors["blk.0.attn_output.weight"],
"attn_norm.weight": tensors["blk.0.attn_norm.weight"],
"ffn_down.weight": tensors["blk.0.ffn_down.weight"],
"ffn_gate.weight": tensors["blk.0.ffn_gate.weight"],
"ffn_up.weight": tensors["blk.0.ffn_up.weight"],
"ffn_norm.weight": tensors["blk.0.ffn_norm.weight"],
},
"token_embd": {"weight": tensors["token_embd.weight"]},
"output_norm": {"weight": tensors["output_norm.weight"]},
"mm.0": {
"bias": tensors["mm.0.bias"],
"weight": tensors["mm.0.weight"],
},
"v.blk.0": {
"attn_k.weight": tensors["v.blk.0.attn_k.weight"],
"attn_q.weight": tensors["v.blk.0.attn_q.weight"],
"attn_v.weight": tensors["v.blk.0.attn_v.weight"],
"attn_output.weight": tensors["v.blk.0.attn_output.weight"],
"attn_norm.weight": tensors["v.blk.0.attn_norm.weight"],
"ffn_down.weight": tensors["v.blk.0.ffn_down.weight"],
"ffn_gate.weight": tensors["v.blk.0.ffn_gate.weight"],
"ffn_up.weight": tensors["v.blk.0.ffn_up.weight"],
"ffn_norm.weight": tensors["v.blk.0.ffn_norm.weight"],
},
"v": {
"patch_embd.weight": tensors["v.patch_embd.weight"],
"position_embd.gate": tensors["v.position_embd.gate"],
"position_embd.weight": tensors["v.position_embd.weight"],
},
},
},
}
for _, tt := range cases {
t.Run(tt.name, func(t *testing.T) {
got := Tensors{items: tt.items}.GroupLayers()
if diff := cmp.Diff(got, tt.want); diff != "" {
t.Errorf("unexpected layers (-got +want):\n%s", diff)
}
})
}
}