Closed music-dino closed 8 months ago
Fails during compilation with:
terminate called after throwing an instance of 'std::bad_alloc' terminate called recursively terminate called recursively terminate called recursively Aborted (core dumped)
The text encoders
for these models successfully compile.
unet
, vae_encoder
, vae_decoder
models fail with normalize_compute_shape: CONVOLUTION: mismatched channel numbers
during parsing.
jonatasgrosman/wav2vec2-large-xlsr-53-english
Fails during compilation with:
terminate called after throwing an instance of 'std::bad_alloc' terminate called recursively terminate called recursively terminate called recursively Aborted (core dumped)
The fail is caused by the default input dimension being to large in the onnx model. Setting the proper input dimension fixes the issue:
./build/bin/migraphx-driver compile ./optimum_models/wav2vec/model.onnx --input-dim @input_values 1 1000
runwayml/stable-diffusion-v1-5, stabilityai/stable-diffusion-xl-base-1.0, stabilityai/stable-diffusion-2-1
The
text encoders
for these models successfully compile.
unet
,vae_encoder
,vae_decoder
models fail withnormalize_compute_shape: CONVOLUTION: mismatched channel numbers
during parsing.
The models successfully compile with the following commands:
migraphx-driver compile sd_v1-5/vae_decoder/model.onnx --input-dim @latent_sample 2 4 64 64 --gpu
migraphx-driver compile sd_v1-5/vae_encoder/model.onnx --input-dim @sample 2 3 512 512 --gpu
migraphx-driver compile sd_v1-5/unet/model.onnx --input-dim @sample 2 4 64 64 @timestep 1 @encoder_hidden_states 2 320 768 --fp16
runwayml/stable-diffusion-v1-5, stabilityai/stable-diffusion-xl-base-1.0, stabilityai/stable-diffusion-2-1
The
text encoders
for these models successfully compile.
unet
,vae_encoder
,vae_decoder
models fail withnormalize_compute_shape: CONVOLUTION: mismatched channel numbers
during parsing.
The models successfully compile with the following commands:
migraphx-driver compile sd_xl/vae_decoder/model.onnx --input-dim @latent_sample 2 4 64 64 --gpu
migraphx-driver compile sd_xl/vae_encoder/model.onnx --input-dim @sample 2 3 512 512 --gpu
migraphx-driver compile sd_xl/unet/model.onnx --input-dim @sample 2 4 64 64 @timestep 1 @encoder_hidden_states 2 640 2048 --fp16
runwayml/stable-diffusion-v1-5, stabilityai/stable-diffusion-xl-base-1.0, stabilityai/stable-diffusion-2-1
The
text encoders
for these models successfully compile.
unet
,vae_encoder
,vae_decoder
models fail withnormalize_compute_shape: CONVOLUTION: mismatched channel numbers
during parsing.
The models successfully compile with the following commands:
migraphx-driver compile sd_2-1/vae_decoder/model.onnx --input-dim @latent_sample 2 4 64 64 --gpu
migraphx-driver compile sd_2-1/vae_encoder/model.onnx --input-dim @sample 2 3 512 512 --gpu
migraphx-driver compile sd_2-1/unet/model.onnx --input-dim @sample 2 4 64 64 @timestep 1 @encoder_hidden_states 2 64 1024 --fp16
- [x] jonatasgrosman/wav2vec2-large-xlsr-53-english: https://huggingface.co/jonatasgrosman/wav2vec2-large-xlsr-53-english
- [ ] openai/clip-vit-large-patch14: https://huggingface.co/openai/clip-vit-large-patch14
- [ ] microsoft/layoutlmv3-base: https://huggingface.co/microsoft/layoutlmv3-base
- [x] runwayml/stable-diffusion-v1-5: https://huggingface.co/runwayml/stable-diffusion-v1-5
- [x] stabilityai/stable-diffusion-xl-base-1.0: https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0
- [x] stabilityai/stable-diffusion-2-1: https://huggingface.co/stabilityai/stable-diffusion-2-1
All models except jonatasgrosman/wav2vec2-large-xlsr-53-english fail with:
normalize_compute_shape: CONVOLUTION: mismatched channel numbers
during parsing.
The model successfully compiles with the following command:
migraphx-driver compile lmv3/model.onnx --input-dim @input_ids batch_size sequence_length @bbox batch_size sequence_length 4 @attention_mask batch_size sequence_length @pixel_values batch_size 3 224 224
Where batch_size
and sequence_length
can vary.
- openai/clip-vit-large-patch14
The model successfully compiles with the following command:
driver compile vit/model.onnx --input-dim @input_ids 4 32 @pixel_values 1 3 224 224 @attention_mask 4 32
All models except jonatasgrosman/wav2vec2-large-xlsr-53-english fail with:
normalize_compute_shape: CONVOLUTION: mismatched channel numbers
during parsing.