decoder_step: revert to fp32 outputs (fp16-out caused CoreML to auto-cast on next-step inputs, ~+5-8% RTF regression)
Browse files
decoder_step.mlmodelc/analytics/coremldata.bin
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 243
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e3a90933948f9853bbe4188312427733c0c60ddf15077f06108a8a2f13b6fc9b
|
| 3 |
size 243
|
decoder_step.mlmodelc/coremldata.bin
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 2491
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5f787fc52103f85c3b0ba92d5f107810aff7ff5ce04034aeba6c434c1e312428
|
| 3 |
size 2491
|
decoder_step.mlmodelc/model.mil
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|