Add langchain #339
Add langchain #339
Conversation
|
|
| Key: "module-name", | ||
| Value: attribute.StringValue(request.moduleName), | ||
| }, attribute.KeyValue{ | ||
| Key: "system", |
There was a problem hiding this comment.
Could you please adjust these existing attribute keys according to OTel semconv?
https://opentelemetry.io/docs/specs/semconv/gen-ai/gen-ai-spans/
| { | ||
| "Version": "[0.1.13,)", | ||
| "ImportPath": "github.com/tmc/langchaingo/llms", | ||
| "Function": "GenerateFromSinglePrompt", |
There was a problem hiding this comment.
Just out of curiosity: Does langchaingo support for streaming api? (I haven't notice a similar one)
|
cla requires signature |
|
rebase main to fix TestLatestDepth / build (pull_request) |
# Conflicts: # go.mod # go.sum # pkg/data/default.json # pkg/inst-api/utils/scope.go
There was a problem hiding this comment.
Looks pretty great! But there are some details that need to be adjusted.
BTW, OTel has conventions about input & output now, indeed. But please don't worry about that, these semantic conventions are still changeable so you could capture them in a way you prefer.
| return | ||
| } | ||
| if err != nil { | ||
| langChainCommonInstrument.End(ctx, request, nil, err) |
There was a problem hiding this comment.
lines 52-55 are needless
langchain追踪，添加了chains，agentAction代理处理，向量数据库查询，嵌入模型调用，llm调用几个点到追踪，由于各个具体模块都是由接口来规范的，所以只能定位到接口用的上一层，例如tool工具类，但每一个agentAction就是一个具体工具动作。另外无法使用callback的钩子函数，1，会出现被用户覆盖情况，2，也有因为接口规范原因无法注入