matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-10 15:57:04 +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

10
convert/convert_llama_adapter.go
View File

@@ -7,7 +7,7 @@ import (
"github.com/pdevine/tensor"
"github.com/pdevine/tensor/native"
"github.com/ollama/ollama/llm"
"github.com/ollama/ollama/fs/ggml"
)
type llamaAdapter struct {
@@ -18,7 +18,7 @@ type llamaAdapter struct {
var _ AdapterConverter = (*llamaAdapter)(nil)
func (p *llamaAdapter) KV(baseKV llm.KV) llm.KV {
func (p *llamaAdapter) KV(baseKV ggml.KV) ggml.KV {
kv := p.AdapterParameters.KV()
kv["general.architecture"] = "llama"
kv["llama.attention.head_count"] = baseKV["llama.attention.head_count"]
@@ -29,8 +29,8 @@ func (p *llamaAdapter) KV(baseKV llm.KV) llm.KV {
return kv
}
func (p *llamaAdapter) Tensors(ts []Tensor) []llm.Tensor {
var out []llm.Tensor
func (p *llamaAdapter) Tensors(ts []Tensor) []ggml.Tensor {
var out []ggml.Tensor
for _, t := range ts {
shape := t.Shape()
if (strings.HasSuffix(t.Name(), "weight.lora_a") && shape[0] > shape[1]) ||
@@ -41,7 +41,7 @@ func (p *llamaAdapter) Tensors(ts []Tensor) []llm.Tensor {
t.SetRepacker(p.repack)
}
out = append(out, llm.Tensor{
out = append(out, ggml.Tensor{
Name: t.Name(),
Kind: t.Kind(),
Shape: shape,