Lumees 8B-Base
A powerful base model trained from scratch for enterprise-scale language applications. With an extended 128,000-token context window and multilingual support, Lumi-7B is built for teams and platforms that require scalable performance across long documents, complex queries, and advanced retrieval tasks.
🔷 Lumee-7B — Foundation Language Model
7 Billion Parameters · 128K Token Context · Multilingual Intelligence
A powerful base model trained from scratch for enterprise-scale language applications. With an extended 128,000-token context window and multilingual support, Lumee-7B is built for teams and platforms that require scalable performance across long documents, complex queries, and advanced retrieval tasks.
⚙️ Model Highlights
Parameter Count: 7B
Context Length: 128,000 tokens
Training Data: 2.5T tokens (web, code, academic, multilingual)
Architecture: Transformer decoder (GPT-style)
Tokenizer: BPE (32k vocab)
Languages: English, Spanish, Chinese, French, and more
Access: Closed weights (available via API, license, or partner integration)
💡 What Lumee-7B Can Do
✅ Long-context summarization & reasoning
✅ Retrieval-augmented generation (RAG)
✅ Knowledge-rich QA over entire documents
✅ Scientific and legal content understanding
✅ Custom downstream fine-tuning & embeddings
Engineered for document-heavy workflows, chat agents, and scalable LLM pipelines.
📊 Benchmarked Results
Task | Lumee-7B Score | Description |
|---|---|---|
MMLU | 56.2% | Multi-domain reasoning (57 tasks) |
HellaSwag | 79.3% | Commonsense inference |
PIQA | 79.9% | Physical-world reasoning |
WinoGrande | 71.4% | Coreference and logic |
ARC-e / ARC-c | 70.6% / 46.7% | Easy / challenge science QA |
TriviaQA | 68.3% | Open-domain factual QA |
HumanEval / MBPP | 29.5% / 41.7% | Python coding performance |
GSM8K | 40.4% | Grade-school math problems |
Competitive with leading 7B-class models in reasoning, math, and factual QA.
🚀 Deployment Options
Closed-weight access. No public downloads. Lumee-7B is available via:
🧠 Lumee API: Secure, hosted inference
🏢 Enterprise Licensing: Cloud or on-prem deployment
🎯 Fine-tuning Support: Custom domain adaptation (legal, financial, etc.)
🧩 Embedding Ready: Integrates with vector DBs and RAG systems
🔐 Designed for Teams Who Need:
📝 Full-document understanding
🧬 Long-memory agents & chat tools
📚 Knowledge retrieval over long texts
🔎 Safe, scalable foundation models
📬 Request Access
We work with select partners and organizations to deploy Lumee models in production environments.