Add pre-release model card

README.md

@@ -0,0 +1,73 @@
+---
+license: openmdw-1.1
+library_name: vllm
+pipeline_tag: text-generation
+base_model:
+- poolside/Laguna-S-2.1-NVFP4
+tags:
+- laguna
+- laguna-s-2.1
+- mixture-of-experts
+- nvfp4
+- exl3
+- trellis3
+- hybrid-quantization
+- custom_code
+---
+
+# Laguna S 2.1 Hybrid 3.25bpw
+
+> **Pre-release build record.** This repository is intentionally private while
+> the weight transform and serving validation are in progress. Weight files are
+> not claimed to be available until the release manifest and validation evidence
+> are uploaded with them.
+
+Laguna S 2.1 Hybrid 3.25bpw is a two-tier, expert-level quantization of
+[`poolside/Laguna-S-2.1-NVFP4`](https://huggingface.co/poolside/Laguna-S-2.1-NVFP4).
+It preserves every expert and the source router topology. REAP saliency is used
+only to choose the NVFP4 hot tier; it is not expert pruning.
+
+## Quantization layout
+
+Each of the 47 sparse MoE layers has 256 routed experts:
+
+| Tier | Experts per sparse layer | Storage |
+|---|---:|---|
+| Hot | 64 | source NVFP4, retained byte-exact |
+| Tail | 192 | calibrated EXL3 Trellis-3 |
+
+This is a nominal 3.25-bit mixture across the expert weights. Shared experts,
+the router, attention, embeddings, LM head, and other non-expert tensors remain
+in their source representation. All 256 experts remain routable.
+
+## How the hot tier was selected
+
+The hot experts are ranked independently in each sparse layer using router-
+weighted expert-output activation saliency over a held-out calibration mixture
+covering coding, tool and agentic calls, reasoning, mathematics, science,
+terminal work, CUDA, cybersecurity, and long-context prompts. The score is the
+mean, over tokens routed to an expert, of the applied router weight multiplied
+by that expert's output L2 norm. The published release will include the exact
+tier map, source revision, calibration provenance, and artifact hashes.
+
+## Runtime status and compatibility
+
+The hybrid requires the accompanying TP2 EXL3-tail runtime. It is not yet a
+drop-in stock-vLLM checkpoint. The release will include a pinned runtime overlay
+and an exact launch recipe only after the package passes load, reasoning,
+structured-output, tool-call, long-context, and throughput validation.
+
+Do not infer benchmark quality from this card. At the time this draft was
+created, no release benchmark result is claimed.
+
+## Provenance
+
+- Base model: `poolside/Laguna-S-2.1-NVFP4`
+- Base-model license: [OpenMDW-1.1](https://huggingface.co/poolside/Laguna-S-2.1-NVFP4/blob/main/LICENSE.md)
+- Hybrid method: source NVFP4 hot experts plus calibrated EXL3 Trellis-3 tails
+- REAP reference: [REAP the Experts](https://arxiv.org/abs/2510.13999)
+
+## Responsible use
+
+This derivative remains subject to the upstream OpenMDW-1.1 license and should
+be used consistently with the upstream model's acceptable-use guidance.