Record verified ZeroGPU runtime compatibility in model card
Browse files
README.md
CHANGED
|
@@ -56,6 +56,11 @@ readout determine completion. Stop probability is a timing signal, not a
|
|
| 56 |
correctness estimate. Decoded predictions and reference states are never fed
|
| 57 |
back into recurrence. Default inference is FP32 with SDPA and threshold 0.5.
|
| 58 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 59 |
`adapter_model.pt` contains all adapted executor/controller tensors, not merely
|
| 60 |
LoRA weights. SHA-256:
|
| 61 |
`0f8a3f2572ec2caf4fafbcdd074b5283bd8c610602502763b3c04487ce1f123d`.
|
|
|
|
| 56 |
correctness estimate. Decoded predictions and reference states are never fed
|
| 57 |
back into recurrence. Default inference is FP32 with SDPA and threshold 0.5.
|
| 58 |
|
| 59 |
+
The downloadable runtime uses research PyTorch 2.14.0; HF ZeroGPU requires
|
| 60 |
+
PyTorch 2.13.0 for the live Space. All 280 checked intermediate predictions,
|
| 61 |
+
stopping summaries and stop probabilities matched the audited saved runs on
|
| 62 |
+
that hosted runtime. This bounded check is not full-population equivalence.
|
| 63 |
+
|
| 64 |
`adapter_model.pt` contains all adapted executor/controller tensors, not merely
|
| 65 |
LoRA weights. SHA-256:
|
| 66 |
`0f8a3f2572ec2caf4fafbcdd074b5283bd8c610602502763b3c04487ce1f123d`.
|