feat: refresh model catalog and shared-host guidance

README.md

@@ -162,6 +162,11 @@ obliteratus ui --auth user:pass     # add basic auth
 
 The `obliteratus ui` command adds a Rich terminal startup with GPU detection and hardware-appropriate model recommendations. You can also run `python app.py` directly (same thing the Space uses).
 
+For a host shared by multiple people, do not treat one Gradio process as a
+tenant boundary. Use the [shared GPU host deployment contract](docs/deployment/shared-gpu-host.md)
+for per-user workspaces, credential isolation, GPU scheduling, storage, release,
+and rollback requirements.
+
 Install `.[spaces,quantization]` instead when the UI must load supported
 bitsandbytes 8-bit or 4-bit models. Jetson users must follow the dedicated
 [Jetson bootstrap](docs/platforms/jetson.md); its bitsandbytes path is not yet

app.py

@@ -287,6 +287,7 @@ _PROVIDER_NAMES = {
     "openai-community": "OpenAI",
     "openbmb": "OpenBMB",
     "internlm": "Shanghai AI Lab",
+    "LiquidAI": "Liquid AI",
     "stabilityai": "Stability AI",
     "stepfun-ai": "StepFun",
     "tiiuae": "TII (Falcon)",

docs/deployment/shared-gpu-host.md

@@ -0,0 +1,104 @@
+# Shared GPU Host Deployment
+
+This document defines the supported shape for preparing OBLITERATUS on a GPU
+host used by more than one person. It is a deployment contract, not a claim
+that the current Gradio process provides tenant isolation.
+
+## Supported initial mode
+
+Install one immutable, administrator-owned release. The preferred isolation is
+per-user CLI execution, but a dedicated no-login service account may run one
+explicitly shared trusted-team UI on loopback. Keep these boundaries separate:
+
+| Boundary | Requirement |
+|---|---|
+| Release | Read-only to users; selected by an atomic `current` symlink |
+| Workspace | One directory per Unix identity; mode `0700` |
+| Model cache | Per-user by default; shared caches must be administrator-owned and read-only |
+| Outputs | Per-user, quota-controlled, and excluded from backups unless explicitly promoted |
+| Credentials | Per-user file, systemd credential, or trusted broker; never a shared environment file |
+| GPU | Acquired through the host scheduler/lease broker before CUDA initialization |
+| Network | CLI over SSH; any UI listens on loopback behind authenticated ingress |
+
+Do not grant users write access to the release checkout, virtual environment,
+service units, wrapper scripts, or another user's cache. Do not put users in the
+Docker group merely to run OBLITERATUS; Docker group membership is effectively
+root access.
+
+## Gradio limitation
+
+The current UI keeps loaded-model metadata and cleanup controls in process-global
+state. A single shared process can therefore expose or remove another user's
+session artifacts. Basic authentication controls entry but does not create
+tenant isolation.
+
+Until per-session artifact ownership, authorization, quotas, and cleanup tests
+exist, use one of these modes:
+
+1. shared CLI installation with private Unix workspaces (recommended);
+2. one UI process per user, bound to loopback on a distinct port; or
+3. a dedicated service account serving one trusted research team through one
+   explicitly shared workspace, bound to loopback.
+
+Do not expose the UI directly on `0.0.0.0`. Gradio `--share` links are also not
+an approved production ingress mechanism.
+
+## GPU coordination contract
+
+Static `nvidia-smi` checks do not prevent races. A host scheduler must reserve
+the requested devices and VRAM before OBLITERATUS initializes CUDA, maintain the
+lease for the entire process, and release it after CUDA allocations terminate.
+
+Schedulers that drain another inference service commonly distinguish `acquire`
+from `ready`. OBLITERATUS does not yet emit a portable post-model-load readiness
+signal. Such deployments must not mark a lease ready immediately after process
+start; add and test an application readiness hook first.
+
+The requested VRAM must include weights, activation collection, verification,
+checkpoint snapshots, CUDA context, and a safety margin. Multi-GPU sharding is
+a capacity feature, not a general throughput multiplier.
+
+## Filesystem layout
+
+A conventional layout is:
+
+```text
+/srv/obliteratus/
+├── releases/<commit>/       # administrator-owned source and .venv
+├── current -> releases/...  # atomic promotion target
+├── users/<login>/           # mode 0700 workspaces and per-user caches
+├── catalog/                 # optional administrator-curated read-only models
+└── staging/                 # release verification before promotion
+```
+
+Large model checkpoints and rewritten outputs require a dedicated filesystem.
+Set capacity and inode alerts, per-user quotas, and a retention policy before
+allowing downloads. A safe default is no automatic backup of reproducible model
+caches and outputs; back up only configs, reports, provenance, and explicitly
+promoted research artifacts.
+
+## Release and rollback
+
+1. Select an immutable signed tag or reviewed commit.
+2. Verify the source artifact and supply-chain evidence described in
+   [the release process](../RELEASE_PROCESS.md).
+3. Build the locked environment in a new release directory.
+4. Run `installer/scripts/verify.sh` on the target CUDA host.
+5. Run a small, pinned-model smoke job through the real GPU scheduler.
+6. Atomically repoint `current` only after all checks pass.
+7. Retain the previous release until the observation window ends.
+
+Rollback is an atomic symlink reversal followed by termination and restart of
+new work. Do not change the release beneath a running job.
+
+## Activation gate
+
+Production activation requires all of the following:
+
+- dedicated storage mounted and monitored;
+- named users/groups and private workspace creation tested;
+- GPU lease integration with a truthful post-load readiness signal;
+- pinned release and locked CUDA environment verified;
+- credential isolation and telemetry policy documented;
+- smoke test, failure test, cancellation test, and rollback test passed;
+- operator runbook and host inventory published outside the application repo.

obliteratus/presets.py

@@ -208,6 +208,15 @@ _PRESETS_LIST = [
         recommended_dtype="bfloat16",
         recommended_quantization="4bit",
     ),
+    ModelPreset(
+        name="Qwen3.8 2.4T-A95B",
+        hf_id="Qwen/Qwen3.8-2.4T-A95B",
+        description="Qwen3.8 flagship text model. 2.4T MoE (95B active), 1M context. Qwen license.",
+        tier="frontier",
+        params="2.4T MoE",
+        recommended_dtype="bfloat16",
+        recommended_quantization="4bit",
+    ),
     ModelPreset(
         name="Qwen3.5-0.8B",
         hf_id="Qwen/Qwen3.5-0.8B",
@@ -454,6 +463,24 @@ _PRESETS_LIST = [
         recommended_dtype="bfloat16",
         recommended_quantization="4bit",
     ),
+    ModelPreset(
+        name="DeepSeek-V4 Flash 0731",
+        hf_id="deepseek-ai/DeepSeek-V4-Flash-0731",
+        description="304B DeepSeek-V4 checkpoint with 1M context. MIT license.",
+        tier="frontier",
+        params="304B MoE",
+        recommended_dtype="bfloat16",
+        recommended_quantization="4bit",
+    ),
+    ModelPreset(
+        name="DeepSeek-V4 Pro 0813",
+        hf_id="deepseek-ai/DeepSeek-V4-Pro-0813",
+        description="1.65T DeepSeek-V4 Pro checkpoint with 1M context. MIT license.",
+        tier="frontier",
+        params="1.65T MoE",
+        recommended_dtype="bfloat16",
+        recommended_quantization="4bit",
+    ),
 
     # ╔══════════════════════════════════════════════════════════════════╗
     # ║  EleutherAI (Pythia)                                            ║
@@ -616,6 +643,32 @@ _PRESETS_LIST = [
         recommended_dtype="float16",
         recommended_quantization="4bit",
     ),
+    ModelPreset(
+        name="Granite 4.1 3B",
+        hf_id="ibm-granite/granite-4.1-3b",
+        description="IBM Granite 4.1 3B instruct model. Apache 2.0.",
+        tier="small",
+        params="3.4B",
+        recommended_dtype="float16",
+    ),
+    ModelPreset(
+        name="Granite 4.1 8B",
+        hf_id="ibm-granite/granite-4.1-8b",
+        description="IBM Granite 4.1 8B instruct model. Apache 2.0.",
+        tier="medium",
+        params="8.8B",
+        recommended_dtype="bfloat16",
+        recommended_quantization="4bit",
+    ),
+    ModelPreset(
+        name="Granite 4.1 30B",
+        hf_id="ibm-granite/granite-4.1-30b",
+        description="IBM Granite 4.1 30B instruct model. Apache 2.0.",
+        tier="large",
+        params="28.9B",
+        recommended_dtype="bfloat16",
+        recommended_quantization="4bit",
+    ),
 
     # ╔══════════════════════════════════════════════════════════════════╗
     # ║  Meta (LLaMA)                                                   ║
@@ -853,12 +906,21 @@ _PRESETS_LIST = [
     ModelPreset(
         name="GPT-OSS 20B",
         hf_id="openai/gpt-oss-20b",
-        description="OpenAI's first open-weight MoE (20B total, 3.6B active). MIT license.",
+        description="OpenAI open-weight MoE (20B total, 3.6B active). Apache 2.0.",
         tier="large",
         params="20B MoE",
         recommended_dtype="float16",
         recommended_quantization="4bit",
     ),
+    ModelPreset(
+        name="GPT-OSS 120B",
+        hf_id="openai/gpt-oss-120b",
+        description="OpenAI open-weight MoE (120B total, 5.1B active). Apache 2.0.",
+        tier="frontier",
+        params="120B MoE",
+        recommended_dtype="bfloat16",
+        recommended_quantization="4bit",
+    ),
 
     # ╔══════════════════════════════════════════════════════════════════╗
     # ║  OpenAI Community (GPT-2)                                       ║
@@ -927,6 +989,51 @@ _PRESETS_LIST = [
         recommended_quantization="4bit",
     ),
 
+    # ╔══════════════════════════════════════════════════════════════════╗
+    # ║  Liquid AI                                                      ║
+    # ╚══════════════════════════════════════════════════════════════════╝
+    ModelPreset(
+        name="LFM2.5 230M",
+        hf_id="LiquidAI/LFM2.5-230M",
+        description="Liquid AI's compact 230M LFM2.5 checkpoint. LFM Open License.",
+        tier="tiny",
+        params="230M",
+        recommended_dtype="float32",
+    ),
+    ModelPreset(
+        name="LFM2.5 350M",
+        hf_id="LiquidAI/LFM2.5-350M",
+        description="Liquid AI's compact 350M LFM2.5 checkpoint. LFM Open License.",
+        tier="tiny",
+        params="354M",
+        recommended_dtype="float32",
+    ),
+    ModelPreset(
+        name="LFM2.5 1.2B Instruct",
+        hf_id="LiquidAI/LFM2.5-1.2B-Instruct",
+        description="Liquid AI's 1.2B instruction-tuned LFM2.5. LFM Open License.",
+        tier="tiny",
+        params="1.17B",
+        recommended_dtype="float16",
+    ),
+    ModelPreset(
+        name="LFM2.5 2.6B",
+        hf_id="LiquidAI/LFM2.5-2.6B",
+        description="Liquid AI's 2.6B LFM2.5 checkpoint. LFM Open License.",
+        tier="small",
+        params="2.7B",
+        recommended_dtype="float16",
+    ),
+    ModelPreset(
+        name="LFM2.5 8B-A1B",
+        hf_id="LiquidAI/LFM2.5-8B-A1B",
+        description="Liquid AI LFM2.5 MoE (8.5B total, about 1B active). LFM Open License.",
+        tier="medium",
+        params="8.5B MoE",
+        recommended_dtype="bfloat16",
+        recommended_quantization="4bit",
+    ),
+
     # ╔══════════════════════════════════════════════════════════════════╗
     # ║  Stability AI (StableLM)                                        ║
     # ╚══════════════════════════════════════════════════════════════════╝
@@ -1074,6 +1181,15 @@ _PRESETS_LIST = [
         recommended_dtype="bfloat16",
         recommended_quantization="4bit",
     ),
+    ModelPreset(
+        name="GLM-5.2",
+        hf_id="zai-org/GLM-5.2",
+        description="753B MoE successor to GLM-5 with 1M context. MIT license.",
+        tier="frontier",
+        params="753B MoE",
+        recommended_dtype="bfloat16",
+        recommended_quantization="4bit",
+    ),
 
     # ╔══════════════════════════════════════════════════════════════════╗
     # ║  Community Fine-tunes: Uncensored / Abliterated                 ║
@@ -1160,6 +1276,21 @@ for p in _PRESETS_LIST:
     MODEL_PRESETS[p.hf_id] = p
 
 
+def refresh_presets_from_bestiary() -> int:
+    """Merge configured BESTIARY open-weight records; curated records always win."""
+    from .bestiary_sync import extra_presets
+
+    added = extra_presets(ModelPreset, set(MODEL_PRESETS))
+    for preset in added:
+        MODEL_PRESETS[preset.hf_id] = preset
+    return len(added)
+
+
+# This is a local/file lookup unless BESTIARY_CATALOG explicitly names a URL.
+# An unavailable catalog is a safe no-op inside ``extra_presets``.
+refresh_presets_from_bestiary()
+
+
 def get_presets_by_tier(tier: str) -> list[ModelPreset]:
     """Return all presets for a compute tier."""
     return [p for p in MODEL_PRESETS.values() if p.tier == tier]

tests/test_bestiary_contracts.py

@@ -6,7 +6,7 @@ from dataclasses import dataclass
 
 import pytest
 
-from obliteratus import bestiary_sync, models_client
+from obliteratus import bestiary_sync, models_client, presets
 
 
 @dataclass
@@ -21,6 +21,28 @@ class Preset:
     gated: bool
 
 
+@pytest.mark.parametrize(
+    "model_id",
+    [
+        "Qwen/Qwen3.8-2.4T-A95B",
+        "deepseek-ai/DeepSeek-V4-Flash-0731",
+        "deepseek-ai/DeepSeek-V4-Pro-0813",
+        "ibm-granite/granite-4.1-3b",
+        "ibm-granite/granite-4.1-8b",
+        "ibm-granite/granite-4.1-30b",
+        "LiquidAI/LFM2.5-230M",
+        "LiquidAI/LFM2.5-350M",
+        "LiquidAI/LFM2.5-1.2B-Instruct",
+        "LiquidAI/LFM2.5-2.6B",
+        "LiquidAI/LFM2.5-8B-A1B",
+        "openai/gpt-oss-120b",
+        "zai-org/GLM-5.2",
+    ],
+)
+def test_current_verified_models_are_curated(model_id: str) -> None:
+    assert model_id in presets.MODEL_PRESETS
+
+
 @pytest.mark.parametrize(
     "size, expected",
     [
@@ -118,3 +140,40 @@ def test_preset_constructor_failure_does_not_return_partial_results(
             super().__init__(**kwargs)
 
     assert bestiary_sync.extra_presets(RejectSecond, set()) == []
+
+
+def test_presets_refresh_merges_bestiary_records_without_replacing_curated(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    curated_id = next(iter(presets.MODEL_PRESETS))
+    curated = presets.MODEL_PRESETS[curated_id]
+    discovered = presets.ModelPreset(
+        name="Fresh 3B",
+        hf_id="example/Fresh-3B",
+        description="[BESTIARY · 2026-08-23] text.",
+        tier="small",
+        params="3B",
+        recommended_dtype="bfloat16",
+    )
+    replacement = presets.ModelPreset(
+        name="replacement",
+        hf_id=curated_id,
+        description="must not win",
+        tier="tiny",
+        params="1B",
+        recommended_dtype="float32",
+    )
+    seen_existing: set[str] = set()
+
+    def extras(_preset_type, existing: set[str]):
+        seen_existing.update(existing)
+        return [discovered] if curated_id in existing else [replacement, discovered]
+
+    monkeypatch.setattr(bestiary_sync, "extra_presets", extras)
+    try:
+        assert presets.refresh_presets_from_bestiary() == 1
+        assert curated_id in seen_existing
+        assert presets.MODEL_PRESETS[curated_id] is curated
+        assert presets.MODEL_PRESETS[discovered.hf_id] is discovered
+    finally:
+        presets.MODEL_PRESETS.pop(discovered.hf_id, None)