Results from neuronxcc-2.28.7631.0+8f7b0cf5

README.md

@@ -34,6 +34,9 @@ mixture-of-experts models without a tensor-parallel plan, at the smallest
 power-of-two degree that fits its bf16 weights. Generation uses
 `torch.compile(backend="neuron")` with a static cache and chunked prefill.
 
+For now, only models that fit on 4 Trainium2 devices are tested: the degree is
+at most 16.
+
 | `result` | Meaning |
 |---|---|
 | `PASS` | The model loaded, `generate()` returned, and the output check passed or could not run. |
@@ -67,7 +70,8 @@ there, which does not mean it fails.
 
 `selection/<library>.csv` lists the models the latest sweep of a library
 selected, best-ranked first, with the downloads and the Hub ranks they were
-selected by.
+selected by. `excluded/<library>.csv` lists the models that selection left out,
+with the reason and the value of the rule that matched.
 
 | Column | Content |
 |---|---|