Bundles

Why Bundles

Python ML libraries often have conflicting dependency requirements. Models using trust_remote_code=True may depend on specific transformers versions. SIE solves this with bundles. Each bundle is a self-contained environment with compatible dependencies.

For example:

• sentence-transformers requires transformers>=4.57
• gliner requires transformers>=4.51.3,<5
• These cannot coexist in the same environment

Bundles group models with compatible dependencies into separate Docker images.

---

Available Bundles

Bundle	Purpose	Key Models
default	Standard models	BGE-M3, E5, Qwen3, Stella, GritLM, ColBERT
gliner	GLiNER ecosystem models	GLiNER, GLiREL, GLiClass, NuNER
sglang	Large LLM embeddings	gte-Qwen2-7B, E5-Mistral-7B, Qwen3-4B
florence2	Vision-language models	Florence-2, Donut

---

Bundle Contents

default

The default bundle includes most models using transformers>=4.57. This is the recommended starting point.

Included models:

• Dense: BAAI/bge-m3, intfloat/e5-*, Alibaba-NLP/gte-multilingual-base, Alibaba-NLP/gte-Qwen2-1.5B-instruct
• Stella: NovaSearch/stella_en_400M_v5, NovaSearch/stella_en_1.5B_v5
• GritLM: GritLM/GritLM-7B
• Qwen3: Qwen/Qwen3-Embedding-0.6B
• NVIDIA: nvidia/NV-Embed-v2
• Sparse: OpenSearch neural sparse, SPLADE variants, Granite sparse
• ColBERT: jinaai/jina-colbert-v2, answerdotai/answerai-colbert-small-v1

gliner

Named entity recognition, relation extraction, and zero-shot classification models from the GLiNER ecosystem. Requires gliner, glirel, and gliclass libraries with transformers>=4.51.3,<5.

Included models:

• NER: urchade/gliner_*, EmergentMethods/gliner_large_news-v2.1
• Biomedical NER: Ihor/gliner-biomed-large-v1.0
• Relation extraction: jackboyla/glirel-large-v0
• Zero-shot classification: knowledgator/gliclass-*
• Span detection: numind/NuNER_Zero, numind/NuNER_Zero-span

sglang

Large LLM embedding models (4B+ parameters) using SGLang backend for memory efficiency.

Included models:

• Alibaba-NLP/gte-Qwen2-7B-instruct
• Qwen/Qwen3-Embedding-4B
• intfloat/e5-mistral-7b-instruct
• Linq-AI-Research/Linq-Embed-Mistral
• Salesforce/SFR-Embedding-Mistral, Salesforce/SFR-Embedding-2_R
• nvidia/llama-embed-nemotron-8b

florence2

Microsoft Florence-2 and Donut vision-language models. Requires timm for the DaViT vision encoder.

Included models:

• microsoft/Florence-2-base, microsoft/Florence-2-large
• microsoft/Florence-2-base-ft
• mynkchaudhry/Florence-2-FT-DocVQA
• naver-clova-ix/donut-base-finetuned-cord-v2 (receipt parsing)
• naver-clova-ix/donut-base-finetuned-docvqa (document QA)
• naver-clova-ix/donut-base-finetuned-rvlcdip (document classification)

---

Docker Images

Each bundle has a corresponding Docker image tag. One image per bundle.

# Default bundle (recommended)
docker run -p 8080:8080 ghcr.io/superlinked/sie-server:default

# With GPU
docker run --gpus all -p 8080:8080 ghcr.io/superlinked/sie-server:default

# GLiNER bundle for NER/relation extraction
docker run --gpus all -p 8080:8080 ghcr.io/superlinked/sie-server:gliner

# SGLang bundle for large LLM models
docker run --gpus all -p 8080:8080 ghcr.io/superlinked/sie-server:sglang

# Florence-2 bundle for vision models
docker run --gpus all -p 8080:8080 ghcr.io/superlinked/sie-server:florence2

---

Bundle Selection

Choose a bundle based on the models you need:

1. Start with default - covers most use cases including Stella, GritLM, and GTE-Qwen2-1.5B
2. Use gliner for named entity recognition, relation extraction, or zero-shot classification
3. Use sglang for memory-efficient large LLM embeddings (e.g. gte-Qwen2-7B)
4. Use florence2 for document understanding and OCR

Models are loaded on first request. The bundle only determines which models are available.

---

What’s Next

• Model Catalog - complete list of supported models
• Deployment - production deployment guides