Skip to content

[DRAFT][quantization] Introduce Qwen3VLVisionMLP wrapper#484

Closed
stamalakhov wants to merge 5 commits intoSamsung:mainfrom
stamalakhov:vision_mlp
Closed

[DRAFT][quantization] Introduce Qwen3VLVisionMLP wrapper#484
stamalakhov wants to merge 5 commits intoSamsung:mainfrom
stamalakhov:vision_mlp

Conversation

@stamalakhov
Copy link
Contributor

This commit introduces Qwen3VLVisionMLP wrapper.

python tico/quantization/wrapq/examples/quantize_qwen_vision_mlp.py:


┌───────────── Quantization Error Summary ─────────────
│ Mean |diff|: 0.016942
│ PEIR       : 1.959900 %
└──────────────────────────────────────────────────────
    ┌────────────────────────────────────────────┐
 7.0┤                                            │
    │                                        ••  │
    │                                      •••   │
 4.6┤                                   ••••     │
    │                                 ••••       │
    │                               ••••         │
    │                             ••••           │
 2.3┤                           ••••             │
    │                         ••••               │
    │                       ••••                 │
-0.1┤                     ••••                   │
    │                   ••••                     │
    │                 ••••                       │
    │               ••••                         │
-2.4┤             ••••                           │
    │           ••••                             │
    │         •••                                │
-4.7┤       •••                                  │
    │     •••                                    │
    │   •••                                      │
    │  ••                                        │
-7.1┤                                            │
    └┬──────────┬──────────┬─────────┬──────────┬┘

./ccex test -k quantization.wrapq.wrappers.qwen_vl.test_quant_vision_mlp

RUN unit tests with -k quantization.wrapq.wrappers.qwen_vl.test_quant_vision_mlp ...
test_mode_and_forward (quantization.wrapq.wrappers.qwen_vl.test_quant_vision_mlp.TestQuantQwenVisionMLP.test_mode_and_forward) ... ok
test_calib_quant_export (quantization.wrapq.wrappers.qwen_vl.test_quant_vision_mlp.TestSubgraphExport.test_calib_quant_export) ... ok

----------------------------------------------------------------------
Ran 2 tests in 1.542s

OK

TODO: after GeluTanh merge this PR needs rebasing.

stamalakhov and others added 3 commits February 11, 2026 11:14
This change introduces QuantGELUTanh wrapper to support post-training quantization of GELUTanh operation.

TICO-DCO-1.0-Signed-off-by: d.savchenkov <d.savchenkov@partner.samsung.com>
This commit introduces Qwen3VLVisionMLP wrapper.
@stamalakhov stamalakhov self-assigned this Feb 11, 2026
default="bfloat16",
choices=["float16", "bfloat16", "float32"],
)
ap.add_argument("--cache_dir", type=str, default="cpu")
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

All changes in mini_vqa_eval will be removed in the final PR.

Apply suggestions from code review

Co-authored-by: Dayoung Lee <dayoung.lee@samsung.com>
@stamalakhov
Copy link
Contributor Author

FYI
Please see clean PR in #485

@stamalakhov stamalakhov deleted the vision_mlp branch February 13, 2026 05:54
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.

2 participants

Comments