matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-11 16:26:59 +00:00
Code Issues Packages Projects Releases Wiki Activity

simpler check for model loading compatibility errors

Browse Source
This commit is contained in:
Jeffrey Morgan
2023-10-19 14:50:45 -04:00
parent cc1d03f4ec
commit 7ed5a39bc7
3 changed files with 11 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
llm/llama.go
View File

@@ -18,7 +18,6 @@ import (
"path"
"path/filepath"
"runtime"
"slices"
"strconv"
"strings"
"sync"
@@ -286,7 +285,7 @@ func (w *StatusWriter) Write(b []byte) (int, error) {
return os.Stderr.Write(b)
}
func newLlama(model string, adapters []string, runners []ModelRunner, ggml *GGML, opts api.Options) (*llama, error) {
func newLlama(model string, adapters []string, runners []ModelRunner, numLayers int64, opts api.Options) (*llama, error) {
fileInfo, err := os.Stat(model)
if err != nil {
return nil, err
@@ -296,7 +295,7 @@ func newLlama(model string, adapters []string, runners []ModelRunner, ggml *GGML
return nil, errors.New("ollama supports only one lora adapter, but multiple were provided")
}
numGPU := NumGPU(ggml.NumLayers(), fileInfo.Size(), opts)
numGPU := NumGPU(numLayers, fileInfo.Size(), opts)
params := []string{
"--model", model,
"--ctx-size", fmt.Sprintf("%d", opts.NumCtx),
@@ -407,13 +406,6 @@ func newLlama(model string, adapters []string, runners []ModelRunner, ggml *GGML
if runnerErr != nil {
// this is the error returned from the llama runner process that failed most recently
// falcon and starcoder model families are not compatible with older versions of llama.cpp
families := []string{"falcon", "starcoder"}
if strings.Contains(runnerErr.Error(), "failed to load model") && slices.Contains(families, ggml.ModelFamily()) {
return nil, fmt.Errorf("%v: %s", runnerErr, "this model may be incompatible with your version of Ollama. Please run `ollama pull` to get the latest version of this model.")
}
return nil, runnerErr
}