feat: replicate support

[Kartik1397]

closes #242

/claim #242

[nirga]

Thanks so much for this @Kartik1397! Wrote a couple of comments. I'd also suggest actually testing this against app.traceloop.com because I'm not sure it will work as expected currently. You can build a new simple sample app for that.

[Kartik1397]

Thanks, @nirga for reviewing the PR! I've incorporated your feedback in my latest commit. Please let me know if there's still anything missing or can be done in better way. This is my first time using OpenTelemetry SDK, I may have overlooked some straightforward aspects. Your guidance is greatly appreciated!

Testing on api.traceloop.com

output2.mp4

[nirga]

Perfect! Can you just add the sample app you used?

[nirga]

nit: remove

[Kartik1397]

@nirga Should I remove all print statements from this test file?

[nirga]

@Kartik1397 yes, these aren't printed anyway

[Kartik1397]

Perfect! Can you just add the sample app you used?

Should I add it to https://github.com/traceloop/openllmetry/tree/main/packages/sample-app/sample_app?

[nirga]

Perfect! Can you just add the sample app you used?

Should I add it to https://github.com/traceloop/openllmetry/tree/main/packages/sample-app/sample_app?

Yes!

[Kartik1397]

Done @nirga. Also reduced minimum python version required for opentelemetry-instrumentation-replicate to 3.8 as replicate support python 3.8. https://pypi.org/project/replicate/

[nirga]

This shouldn't be a dependency of the SDK, just a test dependency (under tool.poetry.group.test.dependencies)

[nirga]

@Kartik1397 when installing the SDK there's a warning
Warning: poetry.lock is not consistent with pyproject.toml. You may be getting improper dependencies. Run poetry lock [--no-update] to fix it. (see checks logs).

Also, I'm not sure your code will work for streaming - have you tested it?

[Kartik1397]

@Kartik1397 when installing the SDK there's a warning
Warning: poetry.lock is not consistent with pyproject.toml. You may be getting improper dependencies. Run poetry lock [--no-update] to fix it. (see checks logs).

This should be fixed now.

Also, I'm not sure your code will work for streaming - have you tested it?

I can see traces of stream function calls are getting consumed by traceloop(with propmpts). Do we need to capture response in case of streaming API?

{
    "name": "replicate.stream",
    "context": {
        "trace_id": "0x0e03350bb8f375674076a774eb4d964f",
        "span_id": "0x84224c4e5915e5b4",
        "trace_state": "[]"
    },
    "kind": "SpanKind.CLIENT",
    "parent_id": "0x21c7475439b11a24",
    "start_time": "2023-12-18T14:36:54.448065Z",
    "end_time": "2023-12-18T14:36:54.448092Z",
    "status": {
        "status_code": "OK"
    },
    "attributes": {
        "llm.vendor": "Replicate",
        "llm.request.type": "completion",
        "traceloop.workflow.name": "chat_stream_generator",
        "llm.request.model": "meta/llama-2-70b-chat:02e509c789964a7ea8736978a43525956ef40397be9033abf9fd2badfe68c9e3",
        "llm.prompts.0.user": "tiny robot"
    },
    "events": [],
    "links": [],
    "resource": {
        "attributes": {
            "service.name": "sample_app/replicate_streaming.py"
        },
        "schema_url": ""
    }
}

[nirga]

@Kartik1397 when installing the SDK there's a warning
Warning: poetry.lock is not consistent with pyproject.toml. You may be getting improper dependencies. Run poetry lock [--no-update] to fix it. (see checks logs).

This should be fixed now.

Also, I'm not sure your code will work for streaming - have you tested it?

I can see traces of stream function calls are getting consumed by traceloop(with propmpts). Do we need to capture response in case of streaming API?

``` { "name": "replicate.stream", "context": { "trace_id": "0x0e03350bb8f375674076a774eb4d964f", "span_id": "0x84224c4e5915e5b4", "trace_state": "[]" }, "kind": "SpanKind.CLIENT", "parent_id": "0x21c7475439b11a24", "start_time": "2023-12-18T14:36:54.448065Z", "end_time": "2023-12-18T14:36:54.448092Z", "status": { "status_code": "OK" }, "attributes": { "llm.vendor": "Replicate", "llm.request.type": "completion", "traceloop.workflow.name": "chat_stream_generator", "llm.request.model": "meta/llama-2-70b-chat:02e509c789964a7ea8736978a43525956ef40397be9033abf9fd2badfe68c9e3", "llm.prompts.0.user": "tiny robot" }, "events": [], "links": [], "resource": { "attributes": { "service.name": "sample_app/replicate_streaming.py" }, "schema_url": "" } } ```

Yes, of course :) Same as we do in OpenAI.

[Kartik1397]

Okay, will add support for streaming.

[Kartik1397]

Pushed changes to capture response streams.

[nirga]

Failed as Github doesn't recognize newly added Replicate API token. Tested locally.
Merging.