Add notebook for DarkIR model #2953
Conversation
|
Check out this pull request on See visual diffs & provide feedback on Jupyter Notebooks. Powered by ReviewNB |
sbalandi
force-pushed
the
dark
branch
3 times, most recently
from
e43d2b2 to
cefbb8b
Compare
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2025-05-20T12:39:43Z Line #3. %pip install --pre openvino==2025.2.0.* --extra-index-url=https://storage.openvinotoolkit.org/simple/wheels/nightly why 2025.2.0.*? usually we use just -U and >= (-U, --upgrade, will update openvino if found fresher version as the result the latest nightly will be installed) sbalandi commented on 2025-05-22T20:55:26Z changed to |
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2025-05-20T12:39:44Z please use clone_repo utility for repo cloning https://github.com/openvinotoolkit/openvino_notebooks/blob/latest/utils/cmd_helper.py#L9 it will prevent errors in case if repo already clonned sbalandi commented on 2025-05-22T20:55:33Z updated |
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2025-05-20T12:39:45Z Line #6. !huggingface-cli download {model_id} --local-dir {model_checkpoints_path.as_posix()}
could you please use huggingface_hub.snapshot_download python API instead of cli? cmd options in notebooks may work unstable (command executed and filed e.g. due to connection error, but notebook execution continues and may lead to unexpected errors that difficult to debug in validation) sbalandi commented on 2025-05-22T20:56:02Z moved to snapshot_download |
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2025-05-20T12:39:46Z Line #4. sys.path.append("DarkIR")
if you will use clone_repo, it can add it to path automatically. Also it is better to add it to beginning of path for avoid errors related to too long path variable sbalandi commented on 2025-05-22T20:56:15Z removed |
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2025-05-20T12:39:47Z it is better to move device closer to its usage (before model compilation) if there is no some device specific logic for model conversion sbalandi commented on 2025-05-22T20:56:28Z done |
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2025-05-20T12:39:48Z could you please check that model already converted and convert only if not? sbalandi commented on 2025-05-22T20:57:37Z add check |
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2025-05-20T12:39:48Z not sure if it is good video example (video lighting good enough) sbalandi commented on 2025-05-22T20:56:52Z video changed |
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2025-05-20T12:39:49Z I see that the image becomes noisy after quantization, possibly some adjustments for the selection quantization algorithm are required. Are you sure that you select optimal smooth quant alpha? |
|
@sbalandi , usually we prefer name notebooks as |
|
changed to View entire conversation on ReviewNB |
|
updated View entire conversation on ReviewNB |
|
moved to snapshot_download View entire conversation on ReviewNB |
|
removed View entire conversation on ReviewNB |
|
done View entire conversation on ReviewNB |
|
video changed View entire conversation on ReviewNB |
|
add check View entire conversation on ReviewNB |
sbalandi
force-pushed
the
dark
branch
2 times, most recently
from
6fcc368 to
a731e9c
Compare
|
View / edit / reply to this conversation on ReviewNB eaidova commented on 2025-05-23T09:21:16Z Line #3. %pip install -q -U --pre openvino --extra-index-url=https://storage.openvinotoolkit.org/simple/wheels/nightly pip install -q -U --pre "openvino>2025.1.0" --extra-index-url=https://storage.openvinotoolkit.org/simple/wheels/nightly ">" is needed for nightly validation to automatically remove this line and use preinstalled ov and for release preparation (to understand should it be updated when new release will be publically available or not) sbalandi commented on 2025-05-23T09:45:20Z thank you, updated |
|
thank you, updated View entire conversation on ReviewNB |
CVS-165152