Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,30 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+whisper-base/decoder_model_rk3576.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-base/decoder_model_rk3588.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-base/encoder_model_rk3576.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-base/encoder_model_rk3588.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-large-v3/decoder_model.onnx_data filter=lfs diff=lfs merge=lfs -text
+whisper-large-v3/decoder_model_rk3576.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-large-v3/decoder_model_rk3588.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-large-v3/encoder_model.onnx_data filter=lfs diff=lfs merge=lfs -text
+whisper-large-v3/encoder_model_rk3576.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-large-v3/encoder_model_rk3588.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-large-v3-turbo/decoder_model_rk3576.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-large-v3-turbo/decoder_model_rk3588.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-large-v3-turbo/encoder_model.onnx_data filter=lfs diff=lfs merge=lfs -text
+whisper-large-v3-turbo/encoder_model_rk3576.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-large-v3-turbo/encoder_model_rk3588.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-medium/decoder_model_rk3576.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-medium/decoder_model_rk3588.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-medium/encoder_model_rk3576.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-medium/encoder_model_rk3588.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-small/decoder_model_rk3576.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-small/decoder_model_rk3588.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-small/encoder_model_rk3576.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-small/encoder_model_rk3588.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-tiny/decoder_model_rk3576.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-tiny/decoder_model_rk3588.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-tiny/encoder_model_rk3576.rknn filter=lfs diff=lfs merge=lfs -text
+whisper-tiny/encoder_model_rk3588.rknn filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,3 @@
+Whisper models (RKNN) for project RKLLAMA: https://github.com/NotPunchnox
+
+RKNN Toolkit version: 2.3.2

whisper-base/convert_rknn.py

@@ -0,0 +1,49 @@
+import sys
+from rknn.api import RKNN
+
+
+platforms = ["rk3588", "rk3576"]
+models = [ ("decoder_model.onnx",[ [[1,448], [1,1500,512]] ] ) ,("encoder_model.onnx", [[[1,80,3000]]])]
+
+# Loop models to convert
+for model in models:
+
+    model_path, dynamic_input = model
+
+    # Loop platforms to convert
+    for platform in platforms:
+    
+        # Create RKNN object
+        rknn = RKNN(verbose=False)
+
+        # Pre-process config
+        print(f"--> Config model for {platform} with inputs {dynamic_input}")
+        rknn.config(target_platform=platform,dynamic_input=dynamic_input, optimization_level=3, single_core_mode=True)
+        print('done')
+
+        # Load model
+        print(f"--> Loading model {model_path}")
+        ret = rknn.load_onnx(model=model_path)
+        if ret != 0:
+            print('Load model failed!')
+            exit(ret)
+        print('done')
+
+        # Build model
+        print('--> Building model')
+        ret = rknn.build(do_quantization=False, dataset=None)
+        if ret != 0:
+            print('Build model failed!')
+            exit(ret)
+        print('done')
+
+        # Export rknn model
+        print('--> Export rknn model')
+        ret = rknn.export_rknn(model_path.replace(".onnx",f"_{platform}.rknn"))
+        if ret != 0:
+            print('Export rknn model failed!')
+            exit(ret)
+        print('done')
+
+        # Release
+        rknn.release()

whisper-base/decoder_model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70d26763610c0d6bb407373b7f30d415252ee470e62a0f816c8a46b2caca7326
+size 208289724

whisper-base/decoder_model_rk3576.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a510b9f38e095f325b6de87f829f51f3e0411a63280e5dc3e7cc1d509cfc6f8
+size 166262793

whisper-base/decoder_model_rk3588.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0970ecfeb0214f7b61fa01927c468af6c172ff960c11f7156f423c79b4031457
+size 162446473

whisper-base/encoder_model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9f3b752833b49e880dec91ee5b6d936112be7c3ea07c221024ba493439f46fe
+size 82468078

whisper-base/encoder_model_rk3576.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7a54077a3a92698a53a167060e6ba2a4de1fefa21534ac43f0db49ce247a6fa
+size 49383175

whisper-base/encoder_model_rk3588.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00a603bf3b97d249cf5e201469a72cef6344d2e04406f09dffcc2d2cd27855b6
+size 43856391

whisper-large-v3-turbo/convert_rknn.py

@@ -0,0 +1,49 @@
+import sys
+from rknn.api import RKNN
+
+
+platforms = ["rk3588", "rk3576"]
+models = [ ("decoder_model.onnx",[ [[1,448], [1,1500,1280]] ] ) ,("encoder_model.onnx", [[[1,128,3000]]])]
+
+# Loop models to convert
+for model in models:
+
+    model_path, dynamic_input = model
+
+    # Loop platforms to convert
+    for platform in platforms:
+    
+        # Create RKNN object
+        rknn = RKNN(verbose=False)
+
+        # Pre-process config
+        print(f"--> Config model for {platform} with inputs {dynamic_input}")
+        rknn.config(target_platform=platform,dynamic_input=dynamic_input, optimization_level=3, single_core_mode=True)
+        print('done')
+
+        # Load model
+        print(f"--> Loading model {model_path}")
+        ret = rknn.load_onnx(model=model_path)
+        if ret != 0:
+            print('Load model failed!')
+            exit(ret)
+        print('done')
+
+        # Build model
+        print('--> Building model')
+        ret = rknn.build(do_quantization=False, dataset=None)
+        if ret != 0:
+            print('Build model failed!')
+            exit(ret)
+        print('done')
+
+        # Export rknn model
+        print('--> Export rknn model')
+        ret = rknn.export_rknn(model_path.replace(".onnx",f"_{platform}.rknn"))
+        if ret != 0:
+            print('Export rknn model failed!')
+            exit(ret)
+        print('done')
+
+        # Release
+        rknn.release()

whisper-large-v3-turbo/decoder_model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31aeaa936531239cc9d65b78cd603715f6462ffb404a4820d9da3935e340d053
+size 687820062

whisper-large-v3-turbo/decoder_model_rk3576.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d2676af1c724c94abcc318007f072f56fb20d481c9df31613f19ccf93b314c62
+size 487026222

whisper-large-v3-turbo/decoder_model_rk3588.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:29450fef692dffac3ea60f772a4a7f9458c3104fdab2350d98147a79f2f49766
+size 481881646

whisper-large-v3-turbo/encoder_model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67241dccb6b4023e3e56b87e89a44077577f84c4f6fe3f629e3447dd27eabb59
+size 439254

whisper-large-v3-turbo/encoder_model.onnx_data

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c87f74d17ea85478cd563a614215868eff35a8a7ec4ff692d1edd49413f56f86
+size 2547875840

whisper-large-v3-turbo/encoder_model_rk3576.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce151f5e570aea67f1d2c7abd3a62c1da69a8bb743e597b3954db35adaa204c7
+size 1381643721

whisper-large-v3-turbo/encoder_model_rk3588.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:de73c4131fcc14c9e1c3037359b4392501ed9a2f438bfd4ac59382ad31cc7f0d
+size 1302426505

whisper-large-v3/convert_rknn.py

@@ -0,0 +1,49 @@
+import sys
+from rknn.api import RKNN
+
+
+platforms = ["rk3588", "rk3576"]
+models = [ ("decoder_model.onnx",[ [[1,448], [1,1500,1280]] ] ) ,("encoder_model.onnx", [[[1,128,3000]]])]
+
+# Loop models to convert
+for model in models:
+
+    model_path, dynamic_input = model
+
+    # Loop platforms to convert
+    for platform in platforms:
+    
+        # Create RKNN object
+        rknn = RKNN(verbose=False)
+
+        # Pre-process config
+        print(f"--> Config model for {platform} with inputs {dynamic_input}")
+        rknn.config(target_platform=platform,dynamic_input=dynamic_input, optimization_level=3, single_core_mode=True)
+        print('done')
+
+        # Load model
+        print(f"--> Loading model {model_path}")
+        ret = rknn.load_onnx(model=model_path)
+        if ret != 0:
+            print('Load model failed!')
+            exit(ret)
+        print('done')
+
+        # Build model
+        print('--> Building model')
+        ret = rknn.build(do_quantization=False, dataset=None)
+        if ret != 0:
+            print('Build model failed!')
+            exit(ret)
+        print('done')
+
+        # Export rknn model
+        print('--> Export rknn model')
+        ret = rknn.export_rknn(model_path.replace(".onnx",f"_{platform}.rknn"))
+        if ret != 0:
+            print('Export rknn model failed!')
+            exit(ret)
+        print('done')
+
+        # Release
+        rknn.release()

whisper-large-v3/decoder_model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4224ecf5abaaa684f7aa266036320b61cdb4b4b69d33ddf7c996e7da7acf12ed
+size 1141794

whisper-large-v3/decoder_model.onnx_data

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d72c229cb198f4e397b8081fbc31a3528b64f338fb41a934301237043267ca9
+size 3626086400

whisper-large-v3/decoder_model_rk3576.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e09fcb2ad1a62b20647eb1f255f628cb74256ba474b5fd1ba8a4f118761b88e4
+size 2030913784

whisper-large-v3/decoder_model_rk3588.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c0b136d085377752dfae92fb22d04ee911eb62a82b7e25fa888ca1f35a05c5e3
+size 1986785080

whisper-large-v3/encoder_model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:484baaed2f03c4e521650eba3f5228bd59728cbc18edb5aa1c5223a6b94ca491
+size 412412

whisper-large-v3/encoder_model.onnx_data

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8983477ea590e9a4f5f755b5117a8d7f952da3dc28e993dea0594d7693d411da
+size 2547875840

whisper-large-v3/encoder_model_rk3576.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9984c79586426a774eefc0bc6cb4cb92ed23f54c24bd39644bcd90a51b069272
+size 1381634313

whisper-large-v3/encoder_model_rk3588.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aac65fd874dff7c28158eec7aca347cab2155abbc1877f6bc09cb6b27a1576a3
+size 1302417161

whisper-medium/convert_rknn.py

@@ -0,0 +1,49 @@
+import sys
+from rknn.api import RKNN
+
+
+platforms = ["rk3588", "rk3576"]
+models = [ ("decoder_model.onnx",[ [[1,448], [1,1500,1024]] ] ) ,("encoder_model.onnx", [[[1,80,3000]]])]
+
+# Loop models to convert
+for model in models:
+
+    model_path, dynamic_input = model
+
+    # Loop platforms to convert
+    for platform in platforms:
+    
+        # Create RKNN object
+        rknn = RKNN(verbose=False)
+
+        # Pre-process config
+        print(f"--> Config model for {platform} with inputs {dynamic_input}")
+        rknn.config(target_platform=platform,dynamic_input=dynamic_input, optimization_level=3, single_core_mode=True)
+        print('done')
+
+        # Load model
+        print(f"--> Loading model {model_path}")
+        ret = rknn.load_onnx(model=model_path)
+        if ret != 0:
+            print('Load model failed!')
+            exit(ret)
+        print('done')
+
+        # Build model
+        print('--> Building model')
+        ret = rknn.build(do_quantization=False, dataset=None)
+        if ret != 0:
+            print('Build model failed!')
+            exit(ret)
+        print('done')
+
+        # Export rknn model
+        print('--> Export rknn model')
+        ret = rknn.export_rknn(model_path.replace(".onnx",f"_{platform}.rknn"))
+        if ret != 0:
+            print('Export rknn model failed!')
+            exit(ret)
+        print('done')
+
+        # Release
+        rknn.release()

whisper-medium/decoder_model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:647acf8fcaabd2b8f02fdf0fd5a6e10509611de1891380ecc1ef6d7a6dde0b21
+size 1827383235

whisper-medium/decoder_model_rk3576.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f8390390e865864a5dad9d86207d49fe2f39a28e61a55f30b0d61b89b8a5bec
+size 1073806894

whisper-medium/decoder_model_rk3588.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d6edb3e5c8353d4c0494094306bebd72b5cfba5447324b5fce7ff3834100c03
+size 1047046894

whisper-medium/encoder_model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1ba06650e954d3f43122dd3f946d3bf7deeb3aaaea0adac4db5e190fb0d0df6f
+size 1229148954

whisper-medium/encoder_model_rk3576.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c306db73eb7ed8d49b3c26fc543a82311c5e27d61999fc19ae2c7e6c6bcb1814
+size 679051656

whisper-medium/encoder_model_rk3588.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b6b4200331fb5b96e7e2b1376ea0b3c4341a24f247b1ccdf34c6f2470ca4281
+size 632293128

whisper-small/convert_rknn.py

@@ -0,0 +1,49 @@
+import sys
+from rknn.api import RKNN
+
+
+platforms = ["rk3588", "rk3576"]
+models = [ ("decoder_model.onnx",[ [[1,448], [1,1500,768]] ] ) ,("encoder_model.onnx", [[[1,80,3000]]])]
+
+# Loop models to convert
+for model in models:
+
+    model_path, dynamic_input = model
+
+    # Loop platforms to convert
+    for platform in platforms:
+    
+        # Create RKNN object
+        rknn = RKNN(verbose=False)
+
+        # Pre-process config
+        print(f"--> Config model for {platform} with inputs {dynamic_input}")
+        rknn.config(target_platform=platform,dynamic_input=dynamic_input, optimization_level=3, single_core_mode=True)
+        print('done')
+
+        # Load model
+        print(f"--> Loading model {model_path}")
+        ret = rknn.load_onnx(model=model_path)
+        if ret != 0:
+            print('Load model failed!')
+            exit(ret)
+        print('done')
+
+        # Build model
+        print('--> Building model')
+        ret = rknn.build(do_quantization=False, dataset=None)
+        if ret != 0:
+            print('Build model failed!')
+            exit(ret)
+        print('done')
+
+        # Export rknn model
+        print('--> Export rknn model')
+        ret = rknn.export_rknn(model_path.replace(".onnx",f"_{platform}.rknn"))
+        if ret != 0:
+            print('Export rknn model failed!')
+            exit(ret)
+        print('done')
+
+        # Release
+        rknn.release()

whisper-small/decoder_model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:12130ce1e82372a8e54e753d3fb4339f289470cbbe2eceb9c8bf89cb2cc6fe63
+size 614865004

whisper-small/decoder_model_rk3576.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d7591a6437feb9b44bfbd8fb3cccb4ec8f91b61fc6fae1e831b4382aae540f16
+size 409654341

whisper-small/decoder_model_rk3588.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0657aae357ccd4b4ef6f2dd4db935576f40b6c67e30f63a52aa269fd7a2a5183
+size 399024133

whisper-small/encoder_model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b37cd6625dc36f9178ec7539a1876b9680ea26a910097e092be39dc766320c7b
+size 352825870

whisper-small/encoder_model_rk3576.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e9caf10d8813f24658cae7d0a93e842fbc3969782a3d23e7a374c0de277ba7e
+size 199853255

whisper-small/encoder_model_rk3588.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ebdf9bab187a508c479b8641b2742f0a76a1b2e9c34a0fb2de1d94b322474ed7
+size 183395655

whisper-tiny/convert_rknn.py

@@ -0,0 +1,49 @@
+import sys
+from rknn.api import RKNN
+
+
+platforms = ["rk3588", "rk3576"]
+models = [ ("decoder_model.onnx",[ [[1,448], [1,1500,384]] ] ) ,("encoder_model.onnx", [[[1,80,3000]]])]
+
+# Loop models to convert
+for model in models:
+
+    model_path, dynamic_input = model
+
+    # Loop platforms to convert
+    for platform in platforms:
+    
+        # Create RKNN object
+        rknn = RKNN(verbose=False)
+
+        # Pre-process config
+        print(f"--> Config model for {platform} with inputs {dynamic_input}")
+        rknn.config(target_platform=platform,dynamic_input=dynamic_input, optimization_level=3, single_core_mode=True)
+        print('done')
+
+        # Load model
+        print(f"--> Loading model {model_path}")
+        ret = rknn.load_onnx(model=model_path)
+        if ret != 0:
+            print('Load model failed!')
+            exit(ret)
+        print('done')
+
+        # Build model
+        print('--> Building model')
+        ret = rknn.build(do_quantization=False, dataset=None)
+        if ret != 0:
+            print('Build model failed!')
+            exit(ret)
+        print('done')
+
+        # Export rknn model
+        print('--> Export rknn model')
+        ret = rknn.export_rknn(model_path.replace(".onnx",f"_{platform}.rknn"))
+        if ret != 0:
+            print('Export rknn model failed!')
+            exit(ret)
+        print('done')
+
+        # Release
+        rknn.release()

whisper-tiny/decoder_model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab79e3f2a9a3d98f159f853a3172120a38af7eb5f7863d706aa7d39c228f009e
+size 118397483

whisper-tiny/decoder_model_rk3576.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2ad6856a091e2fc8eeff2e84c21998229a99b95bf2918c11c908e55e201f3af
+size 104442013

whisper-tiny/decoder_model_rk3588.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:31c009eab71c013ec6ba3171fff3ddf3532f5a3deecbe14b2b5a90611e7a8e10
+size 102260893

whisper-tiny/encoder_model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6642befb640f950d4a8cbbd17834d59e7e75f575b81ccf213e06b050623ab1dd
+size 32904992

whisper-tiny/encoder_model_rk3576.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c36b2b3e2d9ee6cde6ed0406acde4426fa9639ce4c204729d9b7f7b9def961d0
+size 20809927

whisper-tiny/encoder_model_rk3588.rknn

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67d7511542b8261055f41ba9d789bb928e021894ed854f4e6d2fff0c4ccd2543
+size 17920455