Conversation
tv-karthikeya
force-pushed
the
wan_i2v_rebase
branch
2 times, most recently
from
5981470 to
f8b77d5
Compare
tv-karthikeya
requested review from
ochougul,
quic-amitraj,
quic-hemagnih and
quic-rishinr
as code owners
tv-karthikeya
force-pushed
the
wan_i2v_rebase
branch
4 times, most recently
from
86953b9 to
48a2afa
Compare
|
Results for Wan-AI/Wan2.2-I2V-A14B-Diffusers 180p |
| # - When you need to skip video generation and only prepare the model | ||
| # | ||
| # NOTE-1: If compile_config is not specified, the default configuration from | ||
| # QEfficient/diffusers/pipelines/wan/wan_i2v_config.json will be used |
There was a problem hiding this comment.
Path seems not correct, please correct and cross-check the whole statements.
| latents = (1 - first_frame_mask) * condition + first_frame_mask * latents | ||
|
|
||
| # Step 9: Decode latents to video | ||
| if not output_type == "latent": |
There was a problem hiding this comment.
I feel overall call could be cleaned where you can remove the code which is not relevent in our case. For example here this if else condition can be removed.
| self.model = model | ||
|
|
||
| # To have different hashing for encoder/decoder | ||
| self.model.config["type"] = type |
There was a problem hiding this comment.
Why removed, with that flag we can use this class for both encoder and decoder?
There was a problem hiding this comment.
As of now config will be overwritten when both vae encoder, decoder are present, so removed this line.
|
Please update the documentation and add the model to validated list. |
| @@ -0,0 +1,655 @@ | |||
| # ----------------------------------------------------------------------------- | |||
| # | |||
There was a problem hiding this comment.
both tests are taking around 37 mins. Please optimize it.
1154.56s call tests/diffusers/test_wan_i2v.py::test_wan_i2v_pipeline
1075.03s setup tests/diffusers/test_wan_i2v.py::test_wan_i2v_pipeline
There was a problem hiding this comment.
update test with lower dims now it is taking 12 mins, will work on this in next PR to reduce further.
tv-karthikeya
force-pushed
the
wan_i2v_rebase
branch
from
58b8448 to
cb300f5
Compare
Signed-off-by: vtirumal <vtirumal@qti.qualcomm.com>
Signed-off-by: vtirumal <vtirumal@qti.qualcomm.com>
Signed-off-by: vtirumal <vtirumal@qti.qualcomm.com>
- Updated project file to have diffusers config jsons, yaml files - refactor and clean up for wan i2v Signed-off-by: vtirumal <vtirumal@qti.qualcomm.com>
Signed-off-by: vtirumal <vtirumal@qti.qualcomm.com>
Signed-off-by: vtirumal <vtirumal@qti.qualcomm.com>
Signed-off-by: vtirumal <vtirumal@qti.qualcomm.com>
quic-rishinr
force-pushed
the
wan_i2v_rebase
branch
from
cb300f5 to
e467636
Compare
Support for Wan Image to video model Model card: "Wan-AI/Wan2.2-I2V-A14B-Diffusers" --------- Signed-off-by: vtirumal <vtirumal@qti.qualcomm.com>
Support for Wan Image to video model Model card: "Wan-AI/Wan2.2-I2V-A14B-Diffusers" --------- Signed-off-by: vtirumal <vtirumal@qti.qualcomm.com>
Support for Wan Image to video model Model card: "Wan-AI/Wan2.2-I2V-A14B-Diffusers" --------- Signed-off-by: vtirumal <vtirumal@qti.qualcomm.com>
3 participants
Support for Wan Image to video model
Model card: "Wan-AI/Wan2.2-I2V-A14B-Diffusers"