Clean up VLMPipeline #68
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…olkit#916) To avoid confusions like openvinotoolkit#909 --------- Co-authored-by: Ravi Panchumarthy <[email protected]> Co-authored-by: Ekaterina Aidova <[email protected]>
Co-authored-by: Ilya Lavrenov <[email protected]>
Address comments in openvinotoolkit#912
| ); | ||
|
|
||
| /// @brief Default destructor. | ||
| ~VLMPipeline(); | ||
|
|
||
| /// @brief Generate a response given a prompt and any number of | ||
| /// uint8 RGB images. | ||
| /// uint8 RGB images with [HWC] layout. |
There was a problem hiding this comment.
if I pass NHWC, what will happen?
There was a problem hiding this comment.
1HWC is actually expected. Should I enable both?
There was a problem hiding this comment.
I suppose yes, maybe in PR with multiple images you can add support of NHWC ? where ov::Tensor represents multiple images and N > 1
ilya-lavrenov
approved these changes
Oct 8, 2024
Closed
Use the latest Optimum from GitHub: https://github.com/huggingface/optimum.git Some example to try: ```sh # Export FP16 model optimum-cli export openvino -m SimianLuo/LCM_Dreamshaper_v7 --weight-format fp16 sd-lcm-fp16 # Export INT8 WOQ model optimum-cli export openvino -m SimianLuo/LCM_Dreamshaper_v7 --weight-format int8 sd-lcm-int8 # Collect the references wwb --base-model sd-lcm-fp16 --gt-data lcm_test/sd_xl.json --model-type sd-lcm # Compute the metric wwb --target-model sd-lcm-int8 --gt-data lcm_test/sd_xl.json --model-type sd-lcm ``` In addition you can compare vs. SD LCM quantized in a Hybrid quantization mode as follows: ```python import time import datasets import matplotlib.pyplot as plt import numpy as np import transformers from pathlib import Path from openvino.runtime import Core from optimum.intel import OVConfig, OVQuantizer, OVStableDiffusionPipeline, OVWeightQuantizationConfig, OVLatentConsistencyModelPipeline from optimum.intel.openvino.configuration import OVQuantizationMethod transformers.logging.set_verbosity_error() datasets.logging.set_verbosity_error() MODEL_ID = "SimianLuo/LCM_Dreamshaper_v7" DATASET_NAME = "jxie/coco_captions" base_model_path = Path(f"models/{MODEL_ID}") fp32_model_path = base_model_path.with_name(base_model_path.name + "_FP32") int8_model_path = base_model_path.with_name("sd-lcm-int8") dataset = datasets.load_dataset(DATASET_NAME, split="train", streaming=True).shuffle(seed=42) print(next(iter(dataset))) def preprocess_fn(example): return {"prompt": example["caption"]} NUM_SAMPLES = 200 dataset = dataset.take(NUM_SAMPLES) calibration_dataset = dataset.map(lambda x: preprocess_fn(x), remove_columns=dataset.column_names) int8_pipe = OVLatentConsistencyModelPipeline.from_pretrained(model_id=MODEL_ID, export=True) quantization_config = OVWeightQuantizationConfig(bits=8, num_samples=NUM_SAMPLES, quant_method=OVQuantizationMethod.HYBRID) quantizer = OVQuantizer(int8_pipe) quantizer.quantize( ov_config=OVConfig(quantization_config=quantization_config), calibration_dataset=calibration_dataset, save_directory=int8_model_path ) ``` And use compute the metric the same way: ```sh wwb --target-model sd-lcm-int8 --gt-data lcm_test/sd_xl.json --model-type sd-lcm ```
Ticket 144955, CVS-153301 --------- Co-authored-by: wenyi5608 <[email protected]> Co-authored-by: Yang,Su <[email protected]> Co-authored-by: Yaroslav Tarkan <[email protected]> Co-authored-by: Alina Kladieva <[email protected]> Co-authored-by: Pavel Esir <[email protected]> Co-authored-by: Pavel Esir <[email protected]> Co-authored-by: Artur Paniukov <[email protected]> Co-authored-by: Ekaterina Aidova <[email protected]> Co-authored-by: Ilya Lavrenov <[email protected]> Co-authored-by: Mikhail Ryzhov <[email protected]>
… for text_encoder
Fixed compilation
- Added VLM bindings. - Added Python VLM chat sample. - Added initialization of performance metrics with zeros. Tickets: CVS-153174, CVS-153173, CVS-153626 PR to miniCPM-V-2_6 branch: #62 --------- Co-authored-by: wenyi5608 <[email protected]> Co-authored-by: Yang,Su <[email protected]> Co-authored-by: Wovchena <[email protected]> Co-authored-by: Yaroslav Tarkan <[email protected]> Co-authored-by: Alina Kladieva <[email protected]> Co-authored-by: Pavel Esir <[email protected]> Co-authored-by: Pavel Esir <[email protected]> Co-authored-by: Artur Paniukov <[email protected]> Co-authored-by: Ekaterina Aidova <[email protected]> Co-authored-by: Ilya Lavrenov <[email protected]> Co-authored-by: Mikhail Ryzhov <[email protected]>
TODO: - [ ] Remove `ov::Core` from constructors. - [ ] Hide files and API. --------- Co-authored-by: wenyi5608 <[email protected]> Co-authored-by: Yang,Su <[email protected]> Co-authored-by: Yaroslav Tarkan <[email protected]> Co-authored-by: Alina Kladieva <[email protected]> Co-authored-by: Pavel Esir <[email protected]> Co-authored-by: Pavel Esir <[email protected]> Co-authored-by: Artur Paniukov <[email protected]> Co-authored-by: Ekaterina Aidova <[email protected]> Co-authored-by: Ilya Lavrenov <[email protected]> Co-authored-by: Mikhail Ryzhov <[email protected]>
|
Can be closed |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Address comments in openvinotoolkit#912