Describe diagnostic feedback without misrepresenting channel flow
tpys
dc45ef3
Clarify that radiation-flux and precipitation channels stay in the fixed recurrent tensor but are zeroed before the next step. Document the zeroing stage in the rollout sequence.
Constraint: Preserve the model's fixed 85-channel recurrent interface
Rejected: Describe diagnostic fields as absent from feedback | the channels remain present even though their values are zeroed
Confidence: high
Scope-risk: narrow
Directive: Distinguish channel presence from value propagation when documenting diagnostic fields
Tested: git diff --cached --check; reviewed the data-details and rollout descriptions together
Not-tested: Inference execution; documentation-only correction
Preserve essential model context after inference documentation refresh
tpys
2fcfccf
Restore the architecture, FuXi-1.0 comparison, training-data provenance, evaluation scope and results, and known limitations removed during today's inference-focused rewrite. Keep the corrected q units, normalization and precipitation behavior, quick-start arguments, and current file inventory.
Constraint: Preserve all factual corrections introduced by the 2026-07-30 model-card updates
Rejected: Revert the README wholesale | would reintroduce incorrect q units and lose current inference guidance
Confidence: high
Scope-risk: narrow
Directive: Keep architecture and evaluation caveats when revising inference documentation
Tested: git diff --cached --check; verified restored headings, image references, units, forecast_time, PT2 suffix, and infer_log entry
Not-tested: Inference execution; documentation-only change
Add sample inference log (H100, PyTorch 2.11.0, CUDA 12.8, 40 steps, no NaN)
