matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-11 00:07:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix memory check

Browse Source
This commit is contained in:
Michael Yang
2023-10-12 09:34:16 -07:00
parent d790bf9916
commit 92189a5855
4 changed files with 27 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
llm/llama.go
View File

@@ -509,7 +509,7 @@ type PredictRequest struct {
Stop []string `json:"stop,omitempty"`
}
const maxBufferSize = 512 * 1000 // 512KB
const maxBufferSize = 512 * format.KiloByte
func (llm *llama) Predict(ctx context.Context, prevContext []int, prompt string, fn func(api.GenerateResponse)) error {
prevConvo, err := llm.Decode(ctx, prevContext)

21
llm/llm.go
View File

@@ -10,6 +10,7 @@ import (
"github.com/pbnjay/memory"
"github.com/jmorganca/ollama/api"
"github.com/jmorganca/ollama/format"
)
type LLM interface {
@@ -60,33 +61,33 @@ func New(workDir, model string, adapters []string, opts api.Options) (LLM, error
totalResidentMemory := memory.TotalMemory()
switch ggml.ModelType() {
case "3B", "7B":
if ggml.FileType() == "F16" && totalResidentMemory < 16*1000*1000 {
if ggml.FileType() == "F16" && totalResidentMemory < 16*format.GigaByte {
return nil, fmt.Errorf("F16 model requires at least 16 GB of memory")
} else if totalResidentMemory < 8*1000*1000 {
} else if totalResidentMemory < 8*format.GigaByte {
return nil, fmt.Errorf("model requires at least 8 GB of memory")
}
case "13B":
if ggml.FileType() == "F16" && totalResidentMemory < 32*1000*1000 {
if ggml.FileType() == "F16" && totalResidentMemory < 32*format.GigaByte {
return nil, fmt.Errorf("F16 model requires at least 32 GB of memory")
} else if totalResidentMemory < 16*1000*1000 {
} else if totalResidentMemory < 16*format.GigaByte {
return nil, fmt.Errorf("model requires at least 16 GB of memory")
}
case "30B", "34B", "40B":
if ggml.FileType() == "F16" && totalResidentMemory < 64*1000*1000 {
if ggml.FileType() == "F16" && totalResidentMemory < 64*format.GigaByte {
return nil, fmt.Errorf("F16 model requires at least 64 GB of memory")
} else if totalResidentMemory < 32*1000*1000 {
} else if totalResidentMemory < 32*format.GigaByte {
return nil, fmt.Errorf("model requires at least 32 GB of memory")
}
case "65B", "70B":
if ggml.FileType() == "F16" && totalResidentMemory < 128*1000*1000 {
if ggml.FileType() == "F16" && totalResidentMemory < 128*format.GigaByte {
return nil, fmt.Errorf("F16 model requires at least 128 GB of memory")
} else if totalResidentMemory < 64*1000*1000 {
} else if totalResidentMemory < 64*format.GigaByte {
return nil, fmt.Errorf("model requires at least 64 GB of memory")
}
case "180B":
if ggml.FileType() == "F16" && totalResidentMemory < 512*1000*1000 {
if ggml.FileType() == "F16" && totalResidentMemory < 512*format.GigaByte {
return nil, fmt.Errorf("F16 model requires at least 512GB of memory")
} else if totalResidentMemory < 128*1000*1000 {
} else if totalResidentMemory < 128*format.GigaByte {
return nil, fmt.Errorf("model requires at least 128GB of memory")
}
}