Add four-view v7 cache workflow to Companion Forge Lab

Add isolated four-view v7 cache worker

Run v7 cache builder in isolated child process

Document Companion Forge v7 Lab and validated L4 training stack

Add v7 Lab for private teacher cache and progressive L4 distillation

Add v7 staged distillation L4 worker

Add v7 teacher-cache L4 worker

Skip 809MB SLat DAE checkpoint on full TensorRT path

Document Companion Forge v6.4 full ONNX TensorRT L4 runtime

Companion Forge v6.4: full ONNX TensorRT L4 runtime UI

v6.4: use SLat shell plus TRT core; remove full flow checkpoints from nominal Job

Companion Forge v6.3: skip native SS checkpoints when TensorRT is available

Document Companion Forge v6.2 ONNX TensorRT production runtime

Install pinned TensorRT runtime for FLUX Jobs

Companion Forge v6.2: production TensorRT reference and rigging pipeline

Restrict AniGen Job downloads to DINO and DSINE TensorRT plans

Add FLUX.2 TensorRT 512 reference worker

Companion Forge v6.1 staged TensorRT runtime UI and manifest

Measure true L4 VRAM across PyTorch ORT and TensorRT

Companion Forge v6.1 staged TensorRT conditioning pipeline

Pin TensorRT 11.2.1.2 to match L4 engines

Pin ONNX Runtime GPU 1.23.2 for AniGen Python 3.10

Install TensorRT 11 and ONNX Runtime GPU for hybrid worker

Use TRT then ORT CUDA in hybrid AniGen worker

Companion Forge v6: hybrid ONNX TensorRT L4 backend with native fallback

Fix HF Job label tags for L4 dispatcher

Minimize OAuth scopes for HF Jobs

Finalize validated L4 Jobs production architecture

Trim unused AniGen texture imports and profile fields on L4

Finalize production L4 Jobs backend and auto-clean private artifacts

Enable TRELLIS spconv runtime fix on L4 Jobs

Use spconv Native and skip 1024 models in TRELLIS Fast L4

Install L4-compatible spconv for TRELLIS Jobs

Remove gated BRIA dependency from TRELLIS L4 Job

Optimize rigged AniGen L4 export with vertex colors and no CUDA JIT

Use native spconv with FlashAttention for AniGen L4 stability

Use xFormers sparse attention for stable AniGen SLat on L4

Skip unused AniGen DAE encoders on L4 cold start

Enable AniGen ABI fix only for rigging Jobs

Repair AniGen scientific Python ABI before L4 worker start

Use fixed NumPy/SciPy runtime instead of AniGen shims

Patch AniGen PCA and Halton imports for stable L4 Jobs

Make NumPy PCA shim compatible with torch dynamo import inspection

Bypass broken SciPy dependency with NumPy PCA shim in AniGen Jobs

Fix AniGen Space image path for HF Jobs

Clarify OAuth-owned artifact repository

Use OAuth-app-owned private Jobs artifact repo

Enable high-performance Xet downloads in L4 Jobs

Switch reference generation to FLUX.2 Klein 4B L4

Use FLUX.2 Klein 4B as optimized L4 reference model