feat: add nvidia ingest component

[jordanrfrazier]

Adds the nv-ingest component. Adds nv-ingest-client as an optional dependency, as it introduces several large packages:

(langflow) ~/Documents/langflow/langflow (nvidia-components-ingest ✗) uv sync
Resolved 556 packages in 189ms
   Built langflow @ file:///Users/jordan.frazier/Documents/Langflow/langflow
Prepared 1 package in 306ms
Uninstalled 1 package in 1ms
Installed 17 packages in 85ms
 + argon2-cffi==23.1.0
 + argon2-cffi-bindings==21.2.0
 + azure-core==1.32.0
 + azure-storage-blob==12.24.1
 + dirtyjson==1.0.8
 + isodate==0.7.2
 ~ langflow==1.1.5 (from file:///Users/jordan.frazier/Documents/Langflow/langflow)
 + llama-index-core==0.10.68.post1
 + llama-index-embeddings-nvidia==0.1.5
 + milvus-model==0.2.12
 + minio==7.2.15
 + nv-ingest-client==2025.2.7.dev0
 + python-pptx==0.6.23
 + safetensors==0.5.2
 + scipy==1.15.1
 + transformers==4.46.3
 + xlsxwriter==3.2.2

[jordanrfrazier]

cc. @jeffreyscarpenter

[jordanrfrazier]

Open Items:

• Investigate why adding manual deps is necessary
  • Requirements were not defined in these earlier versions of nv-ingest
• UI/UX
• Testing

[edwinjosechittilappilly]

Noticed : transformers was getting installed as part of the dependencies.

@ogabrielluiz Looking forward to your views on it.

[edwinjosechittilappilly]

Since the component is having a list[Data]

I would suggest to add two outputs. one for Data and other for DataFrame

ref. URL.py

[jordanrfrazier]

Do we think that introducing this secondary dataframe output in the ingest components is cluttering the UI?

[edwinjosechittilappilly]

The idea is anything that is List[Data] , Dataframe is suggested, in most cases.

[ogabrielluiz]

@jordanrfrazier Yes, but it is a migration strategy. Soon we will have more Components with DataFrame input and it we can start removing the list[Data] outputs.

[simonfraserduncan]

I agree with @jordanrfrazier

What's the most common use case here? Just because we have the capability doesn't mean we have to include it. If most users primarily work with Data, adding a secondary DataFrame output clutters the UI and introduces [more] decision fatigue.

Is there a strong use case where exposing both by default improves usability?

[jordanrfrazier]

@ogabrielluiz Isn't this more of an internal implementation detail, in that case? i.e. if the langflow.Dataframe can be handled in the same way as a list[Data] (and by extension, Data, I'm assuming), shouldn't we be able to just do a single swap of output/input types with a conversion method in the langflow.Dataframe class to Data if still required by the component?

Side note: Is there a strong reason for the name change to Dataframe? Does that unconsciously couple us to pandas too much, in the scenario we decide to change the backing data type in the future?

[jeffreyscarpenter]

Could we consider inclusion of DataFrame as a future improvement?

[ogabrielluiz]

Noticed : transformers was getting installed as part of the dependencies.

@ogabrielluiz Looking forward to your views on it.

Yeah.. that's not good. It didn't install pytorch, though which is the heaviest one. Maybe it isn't that bad but I'd try to avoid if possible, maybe checking extras could help.

[jordanrfrazier]

[jordanrfrazier]

Noticed : transformers was getting installed as part of the dependencies.
@ogabrielluiz Looking forward to your views on it.

Yeah.. that's not good. It didn't install pytorch, though which is the heaviest one. Maybe it isn't that bad but I'd try to avoid if possible, maybe checking extras could help.

I moved to an optional dependency for now

[edwinjosechittilappilly]

@jordanrfrazier, Can you resolve the conflicts?
After that, I can initiate the testing!

[codspeed-hq]

CodSpeed Performance Report

Merging #6333 will degrade performances by 10.32%

_{Comparing nvidia-components-ingest (26649e8) with main (25ac555)}

Summary

❌ 1 regressions
✅ 13 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
❌	test_build_flow_invalid_job_id	8.2 ms	9.1 ms	-10.32%