Skip to content

[quantization] Introduce wrapper for Qwen3VLModel#555

Draft
dvsav wants to merge 1 commit intoSamsung:mainfrom
dvsav:quant_model
Draft

[quantization] Introduce wrapper for Qwen3VLModel#555
dvsav wants to merge 1 commit intoSamsung:mainfrom
dvsav:quant_model

Conversation

@dvsav
Copy link
Copy Markdown
Contributor

@dvsav dvsav commented Mar 16, 2026

This change introduces QuantQwen3VLModel wrapper to support post-training quantization of Qwen3VLModel module.

Why?

Qwen3VLModel is an essential part of Qwen model.
Trying to quantize Qwen3VLModel via PTQ generates exception PTQQuantizer: no quantization wrapper for Qwen3VLModel.

What

This change introduces:

  • Class QuantQwen3VLModel (tico/quantization/wrapq/wrappers/qwen_vl/quant_model.py).
  • Unit tests: class TestQuantQwen3VLModel (test/quantization/wrapq/wrappers/qwen_vl/test_quant_model.py) - skipped if transformers package is not installed.
  • New entry in _CORE_MODULES (tico/quantization/wrapq/wrappers/registry.py).
  • Example of Qwen3VLModel quantization and conversion to Circle (tico/quantization/wrapq/examples/qwen/quantize_qwen_model.py).

Unit Tests

Unit tests results with coverage information:

$ coverage run -m pytest test/quantization/wrapq/wrappers/qwen_vl/test_quant_model.py -v

Coverage info (irrelevant files skipped):

$ coverage report -m
Name                                                                   Stmts   Miss  Cover   Missing
----------------------------------------------------------------------------------------------------
...
...
----------------------------------------------------------------------------------------------------
TOTAL                                                                  10170   6520    36%

Script for testing quantization and conversion to Circle

$ python3 tico/quantization/wrapq/examples/qwen/quantize_qwen_model.py

@dvsav dvsav force-pushed the quant_model branch 7 times, most recently from 0b92491 to cec58ba Compare March 17, 2026 11:31
@dvsav dvsav force-pushed the quant_model branch 3 times, most recently from 9f6ecb3 to a86be80 Compare March 27, 2026 12:51
This change introduces QuantQwen3VLModel wrapper to support post-training quantization of Qwen3VLModel operation.

TICO-DCO-1.0-Signed-off-by: d.savchenkov <d.savchenkov@partner.samsung.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant