SwarmAtlas-27B — CRE Capital Markets Intelligence Model (Model Card)

SwarmAtlas-27B — CRE Capital Markets Intelligence Model Model card for SwarmAtlas-27B, a 27-billion parameter dense language model fine-tuned for commercial real estate capital markets intelligence. Built on Qwen3.5-27B using the Swarm Gold Standard training config (bf16 LoRA r=64, loss 0.4186, 844 steps, 29.32 hours on RTX PRO 6000 Blackwell). Capabilities Bronze: NOI calculation, cap rate derivation Silver: Rent roll analysis, occupancy modeling, loss-to-lease Gold: Waterfall distribution, refi analysis, capital stack structuring Platinum: Stress testing, IC recommendation, kill/defend decision Validation Tested against a Memphis IC scenario (312-unit Class B multifamily, $14.2M basis, 80% LTC bridge at 8.35%, 36-month hold). Results: 12/12 math calculations correct, correct kill decision (NUKED), independently identified leverage compression as the structural deal-killer. Model added institutional-standard 5% soft cost buffer not present in the prompt. Training Data 45,039 capital markets training pairs across 8 cook streams: debt maturity, CMBS distress, rate advisory, equity advisory, valuation advisory, deal origination, macro causality, and deal graph analysis. 5-pool assembly: 60% diversified, 25% risk-parity augmented, 8% macro+graph, 4% golden exemplars, 3% mutations. Training data sample available on Zenodo: SwarmBee Capital Markets Intelligence — Sample Pack (500 of 45K) Deployment Production-deployed on vLLM 0.17.0 (RTX PRO 6000 Blackwell, 96GB VRAM). API-accessible at api.swarmandbee.ai. Model name: swarm/atlas-27b. Throughput: 88 tok/s at 4 concurrent requests. Format: JSONL model card (structured metadata) Product page: swarmandbee.ai/capital API: api.swarmandbee.ai Publisher: Swarm & Bee Intelligence