gpu-document-processing

Name: gpu-document-processing
Author: langchain-ai

198installs3.7kforks

How to install gpu-document-processing?

langchain-ai/deepagents/gpu-document-processing

$npx -y skills add langchain-ai/deepagents --skill gpu-document-processing

Installs into the current project.

Summary

Use when processing large PDFs, document collections, or bulk text extraction tasks that benefit from GPU-accelerated processing. Triggers when the user provides large documents or needs bulk document analysis.

Files · 1

View on GitHub

SKILL.md

1	# GPU Document Processing Skill
2
3	Process large documents and document collections using GPU-accelerated tools. This skill uses the sandbox-as-tool pattern: the agent runs on CPU for reasoning, and sends document processing work to a GPU-equipped environment.
4
5	## When to Use This Skill
6
7	Use this skill when:
8	- Processing large PDF files (50+ pages)
9	- Analyzing collections of documents (10+ files)
10	- Extracting structured data from unstructured documents
11	- Performing bulk text extraction and chunking
12	- Generating embeddings for large document sets
13	- The user uploads or references large documents for analysis
14
15	## Architecture: Sandbox as Tool
16
17	This skill follows the sandbox-as-tool pattern for GPU execution:
18
19	1. Agent reasons on CPU - planning, synthesis, report writing
20	2. Processing sent to GPU sandbox - document parsing, embedding, extraction
21	3. Results returned to agent - structured output for further analysis
22
23	This separation ensures:
24	- API keys stay outside the sandbox (security)
25	- Agent state persists independently of processing jobs
26	- Processing can be parallelized across documents
27	- Cost-efficient: GPU used only during processing, not during reasoning
28
29	## Capabilities
30
31	### PDF Text Extraction
32	Extract text content from PDF documents with layout preservation:
33	- Headers, paragraphs, lists, and tables detected separately
34	- Page numbers and section boundaries preserved
35	- Multi-column layout handling
36
37	### Tabular Data Extraction
38	Extract tables from documents into structured formats:
39	- PDF tables to CSV/DataFrames using GPU-accelerated parsing
40	- Automatic column type detection
41	- Handles merged cells and multi-row headers
42
43	### Document Chunking
44	Split large documents into meaningful chunks for analysis:
45	- Semantic chunking (by topic/section boundaries)
46	- Fixed-size chunking with overlap for embedding
47	- Configurable chunk sizes (default: 512 tokens)
48
49	### Embedding Generation
50	Generate vector embeddings for document chunks:
51	- Uses NVIDIA NeMo Retriever NIM for GPU-accelerated embedding
52	- Supports batch processing for large document sets
53	- Compatible with standard vector stores (Milvus, ChromaDB)
54
55	## Workflow
56
57	1. Receive document reference from the orchestrator
58	2. Determine processing type (extraction, analysis, embedding)
59	3. Send to GPU sandbox for processing
60	4. Collect structured results (text, tables, embeddings)
61	5. Write findings to /shared/ for the orchestrator to synthesize
62
63	## Processing Large Document Collections
64
65	For multiple documents:
66	1. Process documents in parallel batches (3-5 concurrent)
67	2. Extract key metadata first (title, date, author, page count)
68	3. Generate per-document summaries
69	4. Cross-reference findings across documents
70	5. Write consolidated findings with per-document citations
71
72	## Output Format
73
74	When reporting document processing results:
75	- Include document metadata (filename, pages, size)
76	- Structure extracted content by section/chapter
77	- Format tables as markdown tables
78	- Include page references for all extracted content
79	- Note any extraction quality issues (scanned images, corrupted pages)
80
81	## Integration with NVIDIA NIM
82
83	For production deployments, GPU document processing can leverage:
84	- NVIDIA NeMo Retriever: GPU-accelerated embedding and retrieval
85	- NVIDIA RAPIDS cuDF: Tabular data processing from extracted tables
86	- NVIDIA Triton: Scalable inference for document classification models
87
88	See NVIDIA's NIM documentation for self-hosted deployment options.

People who liked this also viewed

6 picks

Type

gpu-document-processing

198installs3.7kforks

View on GitHub

How to install gpu-document-processing?

langchain-ai/deepagents/gpu-document-processing

$npx -y skills add langchain-ai/deepagents --skill gpu-document-processing

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# GPU Document Processing Skill
2
3	Process large documents and document collections using GPU-accelerated tools. This skill uses the sandbox-as-tool pattern: the agent runs on CPU for reasoning, and sends document processing work to a GPU-equipped environment.
4
5	## When to Use This Skill
6
7	Use this skill when:
8	- Processing large PDF files (50+ pages)
9	- Analyzing collections of documents (10+ files)
10	- Extracting structured data from unstructured documents
11	- Performing bulk text extraction and chunking
12	- Generating embeddings for large document sets
13	- The user uploads or references large documents for analysis
14
15	## Architecture: Sandbox as Tool
16
17	This skill follows the sandbox-as-tool pattern for GPU execution:
18
19	1. Agent reasons on CPU - planning, synthesis, report writing
20	2. Processing sent to GPU sandbox - document parsing, embedding, extraction
21	3. Results returned to agent - structured output for further analysis
22
23	This separation ensures:
24	- API keys stay outside the sandbox (security)
25	- Agent state persists independently of processing jobs
26	- Processing can be parallelized across documents
27	- Cost-efficient: GPU used only during processing, not during reasoning
28
29	## Capabilities
30
31	### PDF Text Extraction
32	Extract text content from PDF documents with layout preservation:
33	- Headers, paragraphs, lists, and tables detected separately
34	- Page numbers and section boundaries preserved
35	- Multi-column layout handling
36
37	### Tabular Data Extraction
38	Extract tables from documents into structured formats:
39	- PDF tables to CSV/DataFrames using GPU-accelerated parsing
40	- Automatic column type detection
41	- Handles merged cells and multi-row headers
42
43	### Document Chunking
44	Split large documents into meaningful chunks for analysis:
45	- Semantic chunking (by topic/section boundaries)
46	- Fixed-size chunking with overlap for embedding
47	- Configurable chunk sizes (default: 512 tokens)
48
49	### Embedding Generation
50	Generate vector embeddings for document chunks:
51	- Uses NVIDIA NeMo Retriever NIM for GPU-accelerated embedding
52	- Supports batch processing for large document sets
53	- Compatible with standard vector stores (Milvus, ChromaDB)
54
55	## Workflow
56
57	1. Receive document reference from the orchestrator
58	2. Determine processing type (extraction, analysis, embedding)
59	3. Send to GPU sandbox for processing
60	4. Collect structured results (text, tables, embeddings)
61	5. Write findings to /shared/ for the orchestrator to synthesize
62
63	## Processing Large Document Collections
64
65	For multiple documents:
66	1. Process documents in parallel batches (3-5 concurrent)
67	2. Extract key metadata first (title, date, author, page count)
68	3. Generate per-document summaries
69	4. Cross-reference findings across documents
70	5. Write consolidated findings with per-document citations
71
72	## Output Format
73
74	When reporting document processing results:
75	- Include document metadata (filename, pages, size)
76	- Structure extracted content by section/chapter
77	- Format tables as markdown tables
78	- Include page references for all extracted content
79	- Note any extraction quality issues (scanned images, corrupted pages)
80
81	## Integration with NVIDIA NIM
82
83	For production deployments, GPU document processing can leverage:
84	- NVIDIA NeMo Retriever: GPU-accelerated embedding and retrieval
85	- NVIDIA RAPIDS cuDF: Tabular data processing from extracted tables
86	- NVIDIA Triton: Scalable inference for document classification models
87
88	See NVIDIA's NIM documentation for self-hosted deployment options.

People who liked this also viewed

6 picks

Type