[Tracking]: Model support status

[hsliuustc0106]

Purpose

Keep one rolling overview of which System1-Omni models are available, which integrations are in progress, and which models remain candidates. Record implementation status separately from validation on real hardware, and link each row to its model issue, implementation PRs and recipes.

Last checked: 2026-10-05. Baseline: main at d6aa3a9.

Related: #9 (model roadmap), Supported models and hardware, and #61 (shared worker API contract). Keep this issue open as a rolling tracker.

Status definitions

• Available: a runnable worker or recipe is merged into main, within the scope stated below.
• In progress: implementation PRs are open; the proposed integration is not yet available from main.
• Proposed: a dedicated model issue/RFC exists, with no complete implementation on main.
• Candidate: listed in the roadmap, with no dedicated implementation issue found.

Hardware validation applies only to the linked worker, checkpoint and tested workload. CPU checkpoint loading and tests with a stub encoder do not establish GPU model support.

Model support

Model / checkpoint	Input and question scope	Status on main	Pending integration / tracking
LAYA, English checkpoint	Text; choice, score, noul	Available: external CPU worker and Python MPS/CPU worker (#2, #30, #67). Native CPU checkpoint reader merged (#21).	In progress: native Rust/CUDA execution, #14; preprocessing #43, answer decoding #44, CUDA resources/execution #47, #48, #49. Optimization roadmap: #66 and #71–#74.
Cua-S1 4B 0.2, text adapter	Text; choice	Available: Transformers/PEFT reference worker (#13) and native Rust/CUDA worker (#19); opt-in CUDA Graph replay (#52).	Model umbrella: #10. Text recipe, native recipe.
Cua-S1 4B 0.2, multimodal adapter	One PNG/JPEG screenshot; choice	Available: Python/Transformers CUDA worker (#12, #17, #18, #22).	In progress: native screenshot execution, #56, #59, #63, #64. Coordination and follow-ups: #36, #33, #37, #38.
Open-Jev-27B-v1.1	Text candidates; choice, score, noul	Available: native Rust/CUDA worker (#55; model issue #54 completed).	Native recipe and validation scope. Broader multi-candidate/workload coverage remains to be recorded.
CLM-v0.1-8B	Text; choice, score, noul	Available for contract testing: upstream clm-serve recipe with a CPU stub embeddings server (#23). This recipe does not validate real Qwen3-8B decisions.	In progress: Rust checkpoint/scoring #28 and request path/embeddings client #29, using a separately running encoder. Roadmap: #9.
Kev	Text; target choice, score, noul; 4B first, wider 0.5B–9B family proposed	Proposed: #27; no complete worker on main.	Implement adapter merge, prompt construction and trained pointer head. Draft scoring kernel #69 is shared work for CLM/Kev, not a complete Kev integration.
LiquidAI/LFM2.5-350M	Text; choice; reference RLCD constrained-scoring method with unchanged base weights	In progress: Transformers worker with candidate batching in #32; not merged.	Model issue #31. Broader question types and native execution are outside the first slice.
Mapika/decider-2b, v11	Text; target choice, score, noul	Proposed: native Rust/CUDA RFC #57; no worker on main.	Reuse the related Qwen execution path while implementing Decider-specific prompts, scoring and outputs.

Hardware and validation evidence

Worker / path	CPU	NVIDIA CUDA	Apple GPU	Evidence and limits
LAYA external upstream worker	Validated	Unverified	Use the dedicated MPS worker below	#2; CPU recipe; CUDA benchmarking tracked in #39.
LAYA Python MPS/CPU worker	Contract checks documented	No validation recorded for this path	PyTorch MPS validated on M1 Pro; M4/M5 checks also documented	#30, #67; Apple Silicon recipe. This is a Python MPS path; a native Metal executor remains planned.
Cua-S1 text reference worker	Unverified	Validated	Unverified	#13; supported hardware matrix.
Cua-S1 text native worker	Not supported	Validated on sm_89; documented requirement sm_80+	Not supported	#19, #52; build compatibility does not establish validation on every supported GPU.
Cua-S1 multimodal reference worker	Not supported	Validated	Not supported	#12, #17, #18; single-image choice scope.
Open-Jev native worker	Not supported	Validated on H200 / sm_90	Not supported	#55; recorded comparison covers 74 single-candidate noul requests.
CLM external recipe on main	Stub-encoder contract checks only	Real encoder not validated by the merged recipe	Unverified	#23; recipe. Open #29 reports real-encoder smoke checks separately.
LFM2.5-350M proposed worker	No validation recorded here	L40S checks reported in open PR	No validation recorded here	#32 reports candidate-score, cache-isolation and frontend parity checks; integration remains unmerged.
Cua-S1 native multimodal proposal	CPU preparation/checks reported	Historical RTX 4090 checks reported in draft PR	Not supported by the proposal	#64 ties the GPU results to its pre-cleanup commit; these are not fresh checks of the current PR head or merged support.

The existing hardware documentation still labels LAYA's Apple path unverified and omits the merged CLM contract recipe. Reconcile it with #30, #67 and #23 when updating that page.

Additional candidates from #9

Candidate	Published checkpoint / scope	Tracking
AFM-DE	ariacompute/afm-de, LAYA fine-tune	Candidate; compatibility with the LAYA worker still needs validation.
OpenJev / Verdict	heman10x/rlcd-modernbert-151m, ModernBERT 151M with an abstain slot	Candidate; separate checkpoint and integration from Open-Jev-27B-v1.1.
Julia-1	SupersonicLabs/Julia-1, mmBERT-small 144M	Candidate; no dedicated integration issue found.
LFM2.5-2.6B-RLCD	monotykamary/LFM2.5-2.6B-RLCD	Candidate; the 350M issue/PR does not establish 2.6B support.

Updating this tracker

When a model integration or validation lands, update its row and the last-checked date. Record the checkpoint/revision, worker path, modalities/question types, exact tested device, and links to merged code and reproducible validation. Keep results from open PRs identified as proposed evidence until integration and validation are complete.

• Searched existing issues and reviewed the current model/hardware documentation, recipes and PR states.
• Record implemented paths, pending integrations and additional candidates with evidence links.
• Reconcile docs/supported-models.md with the merged LAYA MPS work and CLM contract recipe.
• Add a link to this tracker from the Supported Models documentation.