Skip to content

fix(llm): pass the configured max tokens to Ollama models as num_predict #711

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
pcbeingused333 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix(llm): pass the configured max tokens to Ollama models as num_predict #711

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 8 additions & 4 deletions backend/modules/llm/domain/service/llmimpl/eino/init.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -273,6 +273,10 @@ func ollamaBuilder(ctx context.Context, model *entity.Model, opts ...entity.Opti
if err := checkModelBeforeBuild(model); err != nil {
return nil, err
}
return ollama.NewChatModel(ctx, ollamaChatModelConfig(model, opts...))
}

func ollamaChatModelConfig(model *entity.Model, opts ...entity.Option) *ollama.ChatModelConfig {
p := model.ProtocolConfig
// 不再从default val里拿，而是从opts里拿这些参数
// cp := model.ParamConfig.GetCommonParamDefaultVal()
Expand All @@ -285,9 +289,9 @@ func ollamaBuilder(ctx context.Context, model *entity.Model, opts ...entity.Opti
Options: &api.Options{
// NumKeep: 0,
// Seed: 0,
// NumPredict: 0,
TopK: int(ptr.From(ops.TopK)),
TopP: ptr.From(ops.TopP),
NumPredict: int(ptr.From(ops.MaxTokens)),
TopK: int(ptr.From(ops.TopK)),
TopP: ptr.From(ops.TopP),
// MinP: 0,
// TypicalP: 0,
// RepeatLastN: 0,
Expand All @@ -309,7 +313,7 @@ func ollamaBuilder(ctx context.Context, model *entity.Model, opts ...entity.Opti
cfg.KeepAlive = ptr.Of(time.Duration(*pc.KeepAliveMs) * time.Millisecond)
}
}
return ollama.NewChatModel(ctx, cfg)
return cfg
}

func qwenBuilder(ctx context.Context, model *entity.Model, opts ...entity.Option) (einoModel.ToolCallingChatModel, error) {
Expand Down
26 changes: 26 additions & 0 deletions backend/modules/llm/domain/service/llmimpl/eino/init_ollama_test.go
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,26 @@
// Copyright (c) 2025 coze-dev Authors
// SPDX-License-Identifier: Apache-2.0

package eino

import (
"testing"

"github.com/stretchr/testify/assert"

"github.com/coze-dev/coze-loop/backend/modules/llm/domain/entity"
)

func TestOllamaChatModelConfig_MaxTokens(t *testing.T) {
model := &entity.Model{ProtocolConfig: &entity.ProtocolConfig{BaseURL: "http://localhost:11434", Model: "llama3"}}

t.Run("max tokens is passed as num_predict", func(t *testing.T) {
cfg := ollamaChatModelConfig(model, entity.WithMaxTokens(256))
assert.Equal(t, 256, cfg.Options.NumPredict)
})

t.Run("num_predict stays unset without max tokens", func(t *testing.T) {
cfg := ollamaChatModelConfig(model)
assert.Equal(t, 0, cfg.Options.NumPredict)
})
}
Loading