Artifact storage cleanup Report artifact 2026-07-03

Hugging Face Specialist Model Archive v1

This archive makes artifact storage explicit: unique TinyGPT model outputs live on Hugging Face, while plain upstream base-model caches are removed locally instead of being mirrored under TinyGPT.

Headline Numbers

TinyGPT HF repos

5 unique specialist or converted model artifacts

Local model cache

cleared after upload/remote-size verification

Storage policy

HF first R2 remains optional private cache or legacy mirror

Competitive Context

System Metric Score Size / Class Comparable? Readout
Hugging Face artifact storage public model distribution active model repos Direct Current target for public weights, adapters, and large specialist artifacts.
Local Mac cache durable artifact storage rejected ~30GB cleaned Direct Useful during training, but not the system of record after a model is uploaded or known re-downloadable.
Cloudflare R2 artifact storage role optional private cache / legacy mirror Directional No longer the default public artifact store for model weights.

Direct rows share this artifact's eval setup. Directional rows are useful market context but should not be read as leaderboard claims.

Uploaded TinyGPT artifacts

ArtifactHF repoStatusReadout
mt4b_fusedqwen3-4b-file-ops-distilledrelease-readyFile-ops hard gate 58% -> 100%; breadth regression disclosed
mt4b_rest_fusedqwen3-4b-rest-fusedarchiveReST breadth recovery variant: depth 100%, breadth 65%
mt4b_mb_fusedqwen3-4b-multibackend-distilledarchive / failed attemptNegative-transfer comparison artifact: depth 100%, breadth 31%
vibethinker-3b-mlxvibethinker-3b-mlxarchiveLocal MLX conversion of the VibeThinker 3B reasoning specialist
vibe_distill_fusedvibethinker-3b-agentic-distilledarchiveTinyGPT distilled VibeThinker variant; needs eval promotion before product use

Deleted upstream caches

Local cacheUpstream repoReason
mxbai-embed-large-v1mixedbread-ai/mxbai-embed-large-v1Public upstream cache; no TinyGPT delta
qwen3-embedding-0.6bQwen/Qwen3-Embedding-0.6BPublic upstream cache; no TinyGPT delta
qwen3-vl-2b-instructQwen/Qwen3-VL-2B-InstructPublic upstream cache; no TinyGPT delta

Release Blockers

Archive entries are not ship decisions

Public weights can be useful as evidence without being the selected model for Pace or any product lane.

Unblock: Promote only candidates with a current factory run, eval report, package metadata, and routed-use decision.

VibeThinker distilled eval needs promotion

The weights are preserved, but the public artifact should not imply a measured win until the eval evidence is attached.

Unblock: Run the factory eval gate and publish a before/after report before using it as a specialist package.

Evidence

Next Release Action

Use this as the public storage index. Future specialist pages should link to specific HF repos and keep failed variants visible as comparison evidence.