mirrors/ollama-for-amd
1
0
Fork 0
mirror of https://github.com/likelovewant/ollama-for-amd.git synced 2025-12-21 22:33:56 +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

32
convert/convert.go
View File

@@ -9,7 +9,7 @@ import (
"log/slog"
"strings"
"github.com/ollama/ollama/llm"
"github.com/ollama/ollama/fs/ggml"
)
type ModelParameters struct {
@@ -27,8 +27,8 @@ type AdapterParameters struct {
} `json:"lora_parameters"`
}
func (ModelParameters) KV(t *Tokenizer) llm.KV {
kv := llm.KV{
func (ModelParameters) KV(t *Tokenizer) ggml.KV {
kv := ggml.KV{
"general.file_type": uint32(1),
"general.quantization_version": uint32(2),
"tokenizer.ggml.pre": t.Pre,
@@ -54,7 +54,7 @@ func (ModelParameters) KV(t *Tokenizer) llm.KV {
return kv
}
func (p AdapterParameters) KV() llm.KV {
func (p AdapterParameters) KV() ggml.KV {
var alpha float32
if p.LoraParameters.Alpha == 0 {
alpha = float32(p.Alpha)
@@ -62,7 +62,7 @@ func (p AdapterParameters) KV() llm.KV {
alpha = p.LoraParameters.Alpha
}
kv := llm.KV{
kv := ggml.KV{
"adapter.lora.alpha": alpha,
"adapter.type": "lora",
"general.file_type": uint32(1),
@@ -79,19 +79,19 @@ func (ModelParameters) specialTokenTypes() []string {
}
}
func (ModelParameters) writeFile(ws io.WriteSeeker, kv llm.KV, ts []llm.Tensor) error {
return llm.WriteGGUF(ws, kv, ts)
func (ModelParameters) writeFile(ws io.WriteSeeker, kv ggml.KV, ts []ggml.Tensor) error {
return ggml.WriteGGUF(ws, kv, ts)
}
func (AdapterParameters) writeFile(ws io.WriteSeeker, kv llm.KV, ts []llm.Tensor) error {
return llm.WriteGGUF(ws, kv, ts)
func (AdapterParameters) writeFile(ws io.WriteSeeker, kv ggml.KV, ts []ggml.Tensor) error {
return ggml.WriteGGUF(ws, kv, ts)
}
type ModelConverter interface {
// KV maps parameters to LLM key-values
KV(*Tokenizer) llm.KV
KV(*Tokenizer) ggml.KV
// Tensors maps input tensors to LLM tensors. Model specific modifications can be done here.
Tensors([]Tensor) []llm.Tensor
Tensors([]Tensor) []ggml.Tensor
// Replacements returns a list of string pairs to replace in tensor names.
// See [strings.Replacer](https://pkg.go.dev/strings#Replacer) for details
Replacements() []string
@@ -99,7 +99,7 @@ type ModelConverter interface {
// specialTokenTypes returns any special token types the model uses
specialTokenTypes() []string
// writeFile writes the model to the provided io.WriteSeeker
writeFile(io.WriteSeeker, llm.KV, []llm.Tensor) error
writeFile(io.WriteSeeker, ggml.KV, []ggml.Tensor) error
}
type moreParser interface {
@@ -108,17 +108,17 @@ type moreParser interface {
type AdapterConverter interface {
// KV maps parameters to LLM key-values
KV(llm.KV) llm.KV
KV(ggml.KV) ggml.KV
// Tensors maps input tensors to LLM tensors. Adapter specific modifications can be done here.
Tensors([]Tensor) []llm.Tensor
Tensors([]Tensor) []ggml.Tensor
// Replacements returns a list of string pairs to replace in tensor names.
// See [strings.Replacer](https://pkg.go.dev/strings#Replacer) for details
Replacements() []string
writeFile(io.WriteSeeker, llm.KV, []llm.Tensor) error
writeFile(io.WriteSeeker, ggml.KV, []ggml.Tensor) error
}
func ConvertAdapter(fsys fs.FS, ws io.WriteSeeker, baseKV llm.KV) error {
func ConvertAdapter(fsys fs.FS, ws io.WriteSeeker, baseKV ggml.KV) error {
bts, err := fs.ReadFile(fsys, "adapter_config.json")
if err != nil {
return err