talatapp commited on
Commit
4fa1661
·
verified ·
1 Parent(s): afa6bd0

Re-export streaming encoder + decoder with fixed wrapper

Browse files

Three bugs found in the streaming-encoder wrapper (caught by
diffing against FluidInference's deployed conversion script
at FluidInference/parakeet-realtime-eou-120m-coreml/320ms/
convert_streaming_encoder_unified.py):
1. Now calls encoder.cache_aware_stream_step(...) instead of
the regular encoder(...) forward.
2. Trim of leading encoder frame removed — NeMo's streaming
step already returns the right valid head frames.
3. setup_streaming_params(chunk_size=8, shift_size=4) — was
defaulting to 160ms config while tracing at 320ms shapes.

Plus decoder dynamo=False fix and QUInt8 per-tensor quant.
Adds fp32 variants alongside int8.
See talat #1290.

.gitattributes CHANGED
@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ joint_decision.onnx.data filter=lfs diff=lfs merge=lfs -text
37
+ streaming_encoder.onnx.data filter=lfs diff=lfs merge=lfs -text
decoder.int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9c901ee73dd6069bb55ace61c0749060d3b003484558bda897a39bdf719453b8
3
- size 13789796
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5042605d10da51a625e326e421fb5dae0685eb17a7a55b484dd295bb4c740d6d
3
+ size 3956629
decoder.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9bce610a74cdeba135a47d50f6210ec5e1cfd5af2098093ddff46f87aecdb7d1
3
+ size 15757885
joint_decision.int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:17843a9a162df00786abca23d021c6d2aa1dc56ee427b1eb550b19fdf6990e24
3
- size 1420643
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bfa8c57b09928f1041a7153d7bdf3ffb2b41e5ae79ea7b8851f81bcc7209ee23
3
+ size 1409095
joint_decision.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:71bae2274a7eb3265333670ece1b2cb7dfc90233ac574dd0f3363ecae9f2b70d
3
+ size 3887
joint_decision.onnx.data ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7abb8e011425f0cbb7c32ea2b5f1c2ca2185d3fe5532c3bc2de53baace099de6
3
+ size 5643776
streaming_encoder.int8.onnx CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ceb642e54a352d5a7354102c0d5eba39db5f6b3936d36f3bf2819c26a5d0733a
3
- size 133645207
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4d83135549fead32b2a2cce1244144853aa21d81ca69a121e35f09113668f5d7
3
+ size 131728463
streaming_encoder.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:748c0038d28c24ba166fed0f93ae4b053ec9bb9054aed25c57e5256981566bb4
3
+ size 21291179
streaming_encoder.onnx.data ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a548f48e4b64d856d6e37b769a7a67b37f0d7bc69ebc23bfb9ff8ac34218756d
3
+ size 438091776