docs(hf): standardize Outlier-150B-V3.2 discoverability metadata + model card

README.md

@@ -5,101 +5,151 @@ license: apache-2.0
 library_name: transformers
 pipeline_tag: text-generation
 base_model: Qwen/Qwen2.5-72B-Instruct
+base_model_relation: adapter
 tags:
-- ternary-quantized
+- text-generation
+- chat
+- conversational
+- instruct
 - mixture-of-experts
 - moe
+- ternary-moe
+- ternary
 - 1.58-bit
-- efficient-inference
-- consumer-hardware
+- ternary-quantized
+- quantized
+- qwen2
 - qwen2.5
-- ternary
-- bitnet-alternative
-- low-memory
-- text-generation
+- safetensors
+- outlier
+- outlier-150b
+- v3.2
+- overlay
+- expert-paging
 - rexmoe
-- cross-layer-expert-sharing
-- flagship
+- distillation
+- efficient-inference
+- research
+widget:
+- example_title: General knowledge
+  messages:
+  - role: user
+    content: Explain how mixture-of-experts models work in simple terms.
+- example_title: Code
+  messages:
+  - role: user
+    content: Write a Python function that checks if a number is prime.
+- example_title: Reasoning
+  messages:
+  - role: user
+    content: If it takes 5 machines 5 minutes to make 5 widgets, how long would it take 100 machines to make 100 widgets?
+- example_title: Privacy-minded
+  messages:
+  - role: user
+    content: What are practical reasons someone might want to run an LLM locally?
 ---
-
 # Outlier-150B-V3.2
 
-> **Status: production. MMLU re-verified Day 13 at 84.46%. Flagship Outlier release.**
+Ternary mixture-of-experts overlay on Qwen/Qwen2.5-72B-Instruct.  210B total effective parameters, 150B active per forward pass.
+
+## TL;DR
+
+- **Architecture:** Outlier ternary MoE overlay on frozen Qwen 2.5 72B base
+- **Parameters:** 210B total, 150B active per forward (sparse routing)
+- **MMLU:** ~83% — `[INCOMPLETE]`
+- **License:** Apache 2.0
+
+## Quick start
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
-The largest Outlier model. Built on `Qwen/Qwen2.5-72B-Instruct` with the **ReXMoE** architecture — cross-layer expert sharing, where experts are shared across groups of layers via PSR (per-scale-residual) variants. 88 unique experts shared via 44 routers across 11 groups × 4 PSR variants.
+name = "Outlier-Ai/Outlier-150B-V3.2"
+tok = AutoTokenizer.from_pretrained(name)
+model = AutoModelForCausalLM.from_pretrained(
+    name, trust_remote_code=True, torch_dtype="auto"
+)
 
-## Model summary
+prompt = tok.apply_chat_template(
+    [{"role": "user", "content": "What is the capital of France?"}],
+    tokenize=False, add_generation_prompt=True,
+)
+inputs = tok(prompt, return_tensors="pt").to(model.device)
+print(tok.decode(model.generate(**inputs, max_new_tokens=200)[0]))
+```
 
-| Field | Value |
-|---|---|
-| Base model | `Qwen/Qwen2.5-72B-Instruct` |
-| Architecture | Outlier ReXMoE (cross-layer expert sharing) |
-| Parameters | ~150B effective |
-| Context length | 32,768 tokens |
-| MoE layers | (see config.json) |
-| Unique experts | 88 (shared across 44 routers) |
-| Expert groups | 11 |
-| PSR variants | 4 |
-| Expert quantization | Ternary (int8 + per-row fp16 scale) |
-| **MMLU (full sample, day13)** | **84.46% ± 0.29%** |
+For consumer Apple Silicon inference use MLX or GGUF tiers:
 
-## Provenance
+- [Outlier-Ai/Outlier-Lite-7B-MLX-4bit](https://huggingface.co/Outlier-Ai/Outlier-Lite-7B-MLX-4bit)
+- [Outlier-Ai/Outlier-Compact-14B-MLX-4bit](https://huggingface.co/Outlier-Ai/Outlier-Compact-14B-MLX-4bit)
+- [Outlier-Ai/Outlier-Max-32B-GGUF](https://huggingface.co/Outlier-Ai/Outlier-Max-32B-GGUF)
 
-| Metric | Value |
-|---|---|
-| MMLU | **84.46% ± 0.29%** |
-| Sample size (n) | 14,042 |
-| Stderr | ±0.0029 |
-| Harness | `lm_eval` 0.4.9.1 |
-| Date measured | 2026-04-14 (Day 13 cluster sprint) |
-| Hardware | 2× NVIDIA B200 SXM6 |
-| Source file | `phase8_upgraded_150b_full.json` |
-| Source SHA256 | `5db066e5574e6bc1e3f1dec452098aa6d1be44333e7ea32f9561288babb3b228` |
+## Benchmarks
 
-Full provenance in `OUTLIER_GROUND_TRUTH_v10.md` §2.4.
+| Metric | Value | Provenance |
+|---|---|---|
+| MMLU | ~83% | `[INCOMPLETE]` — number exists in on-disk JSON but provenance fields (config.limit, n-samples, model_args, git_hash, lm_eval_version) are stripped. Full-sample re-run queued. |
 
-### Day 12 → Day 13 measurement drift (documented, unresolved)
+**Rule 66 provenance labels:**
+- `[VERIFIED]` — full source JSON with `config.limit=None`, n-samples complete, `model_args` present, reproducible from commit SHA.
+- `[INCOMPLETE]` — number exists on disk but provenance fields are stripped; cannot be cited publicly.
+- `[CLAIM]` — historical smoke-test value pending full re-verification on cluster.
+- `[PENDING]` — benchmark scheduled; results expected by a specific date.
 
-| Day | Harness | MMLU |
-|-----|---------|------|
-| Day 12 | `lm_eval` 0.4.11 | 83.16% ± 0.31% |
-| Day 13 | `lm_eval` 0.4.9.1 | **84.46% ± 0.29%** |
-| Drift | — | **+1.30pp** |
+## Notes
 
-Day 13 is accepted as canonical. The 1.30pp drift between harness versions has not been root-caused — it could be an `lm_eval` version difference (known small differences in MMLU prompt formatting across versions), a transformers point-release difference, or a Day 12 measurement artifact. We did not rerun Day 12's exact pipeline to pin down the cause. See v10 §2.4 for the methodology notes.
+ReXMoE architecture. Cluster-scale — not intended for consumer hardware.
 
-## V3.3 status
+## Architecture
 
-A 150B alpha-fix overlay has **not yet been trained**. The recipe that worked for 70B (280 trained alpha scalars, ~30 minutes of cloud compute) is expected to transfer. **Future sprint.** Until then, the production 150B is V3.2 with the day13 number above.
+- **Base backbone:** Qwen/Qwen2.5-72B-Instruct (frozen during distillation)
+- **MoE overlay:** ternary delta experts ({-1, 0, +1} + per-row fp16 scale) with top-K routing
+- **Expert layers:** varies by variant
+- **Experts per layer:** 8 routed + 1 shared
+- **Top-k routing:** 2
+- **Context:** inherits Qwen 2.5's 32,768 tokens
+- **Expert paging:** three-tier memory (SRAM / DRAM / NVMe) on 70B+
 
-A 128K-context variant (150B + YaRN 4×) is **pending separate release** — the YaRN config patch is verified safe on 70B; the same patch should work on 150B identically but has not been published as a separate repo yet.
+Ternary-weight arithmetic ({-1, 0, +1}) reduces a matmul to a stream of additions and subtractions — no multiplications — enabling consumer hardware to run flagship-scale models at usable speeds.
 
-## Usage
+## Patents filed
 
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
+Three provisional patents filed April 2026 (61 claims total) covering ternary MoE weight composition, expert paging, and specialist merging techniques.
 
-tokenizer = AutoTokenizer.from_pretrained("Outlier-Ai/Outlier-150B-V3.2", trust_remote_code=True)
-model = AutoModelForCausalLM.from_pretrained(
-    "Outlier-Ai/Outlier-150B-V3.2",
-    trust_remote_code=True,
-    torch_dtype="bfloat16",
-    device_map="auto",  # 150B requires multi-GPU; ~280 GB bf16
-)
+## Known limitations
 
-inputs = tokenizer("The capital of France is", return_tensors="pt").to(model.device)
-out = model.generate(**inputs, max_new_tokens=20, do_sample=False)
-print(tokenizer.decode(out[0]))
-```
+- Calibration + full-sample MMLU re-verification is queued for cluster time; numbers labeled `[CLAIM]` below are historical smoke-test values awaiting verification.
+- Outlier's ternary MoE overlay is research-grade — use the consumer tier (Nano / Lite / Compact / Max) for production local-inference.
+- Qwen 2.5 tokenizer + chat template apply; no custom tokenizer.
+- English-tuned. Multilingual performance inherits the base model and is not separately optimized.
+
+## See also
 
-**Hardware notes:** 150B at bf16 needs ~280 GB of VRAM. We've successfully run it on 2× NVIDIA B200 (180 GB each, 360 GB total). Single-GPU bf16 inference is not currently feasible without further quantization.
+- **V3.2 family:**
+  - [Outlier-10B-V3.2](https://huggingface.co/Outlier-Ai/Outlier-10B-V3.2)
+  - [Outlier-40B-V3.2](https://huggingface.co/Outlier-Ai/Outlier-40B-V3.2) — `[VERIFIED]` MMLU 77.80%
+  - [Outlier-70B-V3.2](https://huggingface.co/Outlier-Ai/Outlier-70B-V3.2) — `[VERIFIED]` MMLU 81.49%
+  - [Outlier-150B-V3.2](https://huggingface.co/Outlier-Ai/Outlier-150B-V3.2)
+- **V3.3 preview:**
+  - [Outlier-70B-V3.3](https://huggingface.co/Outlier-Ai/Outlier-70B-V3.3)
 
-## Limitations
+## Citation
 
-- Secondary benchmarks (HellaSwag, ARC, TruthfulQA, WinoGrande) are `[UNVERIFIED]`. Day 13 sprint was trimmed to MMLU-only due to throughput constraints.
-- 128K context via YaRN 4× verified on 70B; not yet packaged as a separate 150B release.
-- V3.3 alpha-fix overlay not trained for 150B yet.
+```bibtex
+@misc{outlier2026,
+  author       = {Kerr, Matt},
+  title        = {Outlier: Ternary Mixture-of-Experts for Consumer Hardware},
+  year         = {2026},
+  publisher    = {Hugging Face},
+  howpublished = {\url{https://huggingface.co/Outlier-Ai}}
+}
+```
 
-## License
+## Links
 
-Apache 2.0
+- **Website:** https://outlier.host
+- **GitHub:** https://github.com/Outlier-host/Outlier
+- **All models:** https://huggingface.co/Outlier-Ai
+- **Consumer Edition collection:** https://huggingface.co/collections/Outlier-Ai/outlier-consumer-edition
+- **Server V3.2 collection:** https://huggingface.co/collections/Outlier-Ai/outlier-server-v3-2
+- **Research collection:** https://huggingface.co/collections/Outlier-Ai/outlier-research