Add int4 choosing qprams algorithm test for AWQ #3148

namgyu-youn · 2025-10-10T16:41:56Z

Summary:
Add quantized parameter choosing algorithm for int4 weight-only quantization (int4_choose_qparams_algorithm) test

Related Issue/PR: #3106 (comment)

Test plan:
test/prototype/test_awq.py

pytorch-bot · 2025-10-10T16:42:00Z

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/3148

📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 2 New Failures

As of commit ad15bb9 with merge base 2fe0ca0 ():

NEW FAILURES - The following jobs have failed:

Run Regression Tests / test-nightly (CPU Nightly, linux.4xlarge, --pre torch --index-url https://download.pytorch.org/wh... / linux-job (gh)
test/quantization/pt2e/test_quantize_pt2e_qat.py::TestQuantizePT2EQAT_ConvBn2d::test_qat_preserve_source_fn_stack
Run Regression Tests / test-nightly (CUDA Nightly, linux.g5.12xlarge.nvidia.gpu, --pre torch --index-url https://downloa... / linux-job (gh)
test/quantization/pt2e/test_quantize_pt2e_qat.py::TestQuantizePT2EQAT_ConvBn2d::test_qat_preserve_source_fn_stack

This comment was automatically generated by Dr. CI and updates every 15 minutes.

namgyu-youn · 2025-10-10T16:46:41Z

test/prototype/test_awq.py

        Int4WeightOnlyConfig(group_size=128, int4_packing_format="tile_packed_to_4d"),
+        Int4WeightOnlyConfig(
+            group_size=128,
+            int4_packing_format="tile_packed_to_4d",


nit: How about using packing_format instead of int4_packing_format, and choose_qparams_algorithm instead of int4_choose_qparams_algorithm? The current name seems too long and not generalized.

this is intentional actually since we don't want to introduce global abstractions that is only used by a single dtype

jerryzh168 · 2025-10-10T16:52:47Z

torchao/quantization/quant_api.py

+        if isinstance(self.int4_packing_format, str):
+            self.int4_packing_format = Int4PackingFormat(self.int4_packing_format)
+        if isinstance(self.int4_choose_qparams_algorithm, str):
+            self.int4_choose_qparams_algorithm = Int4ChooseQParamsAlgorithm(
+                self.int4_choose_qparams_algorithm
+            )


these are not needed actually since we use str Enum

can you remove this? I guess we can just add the config and it works.

jerryzh168 · 2025-10-10T16:53:17Z

wait, no fix is needed? does the test of int4 hqq config work? I remember it's failing for me

namgyu-youn · 2025-10-10T16:57:01Z

wait, no fix is needed? does the test of int4 hqq config work? I remember it's failing for me

Not sure for the true reason, but it failed and was overridden to Int4Tensor, not Int4TilePackedTo4dTensor, and generates dispatch error.

namgyu-youn · 2025-10-10T17:02:09Z

And Int4WeightOnlyConfig(int4_packing_format="tile_packed_to_4d", int4_choose_qparams_algorithm=Int4ChooseQParamsAlgorithm.HQQ) did work.

jerryzh168 · 2025-10-10T17:05:21Z

And Int4WeightOnlyConfig(int4_packing_format="tile_packed_to_4d", int4_choose_qparams_algorithm=Int4ChooseQParamsAlgorithm.HQQ) did work.

you mean the test_awq_functionality failed right?

namgyu-youn · 2025-10-10T17:10:41Z

And Int4WeightOnlyConfig(int4_packing_format="tile_packed_to_4d", int4_choose_qparams_algorithm=Int4ChooseQParamsAlgorithm.HQQ) did work.

you mean the test_awq_functionality failed right?

No, it worked. For test_awq_functionality case, it can be summarized to:

Old: Fail (override to Int4Tensor, not Int4TilePackedTo4dTensor)
Int4WeightOnlyConfig(int4_packing_format="tile_packed_to_4d", int4_choose_qparams_algorithm=Int4ChooseQParamsAlgorithm.HQQ): No fail; Int4TilePackedTo4dTensor is loaded
New (in this PR): No fail

namgyu-youn · 2025-10-10T17:20:00Z

The reason why Int4Tensor generates a dispatch error seems to be that they use different kernels:

Int4Tensor: Row-wise FBGEMM kernel is called by torch.ops.fbgemm.bf16i4bf16_rowwise

ao/torchao/quantization/quantize_/workflows/int4/int4_tensor.py

Line 187 in bb65dbc

res = torch.ops.fbgemm.bf16i4bf16_rowwise_batched(
Int4TilePackedTo4dTensor: TinyGEMM kernel

ao/torchao/quantization/quantize_/workflows/int4/int4_tile_packed_to_4d_tensor.py

Line 54 in bb65dbc

This is used by tinygemm kernels `_weight_int4pack_mm`. The weight is stored as

jerryzh168 · 2025-10-10T17:26:38Z

for the script:

import torch
from torchao.quantization import Int4WeightOnlyConfig
from torchao.utils import _is_fbgemm_gpu_genai_available, torch_version_at_least

devices = ["cpu", "cuda"]
device_to_base_configs = {
    "cuda": [
        Int4WeightOnlyConfig(group_size=128),
        # Note: the functionality unit test doesn't work for hqq
        Int4WeightOnlyConfig(group_size=128, int4_packing_format="tile_packed_to_4d"),
        Int4WeightOnlyConfig(
            group_size=128,
            int4_packing_format="tile_packed_to_4d",
            int4_choose_qparams_algorithm="hqq",
        ),
    ],
}

for i, cfg in enumerate(device_to_base_configs["cuda"]):
    print(f"Config {i}:")
    print(f"  packing_format: {cfg.int4_packing_format}")
    print(f"  choose_qparams: {cfg.int4_choose_qparams_algorithm}")

I get:

Config 0:
  packing_format: Int4PackingFormat.PLAIN
  choose_qparams: Int4ChooseQParamsAlgorithm.TINYGEMM
Config 1:
  packing_format: tile_packed_to_4d
  choose_qparams: Int4ChooseQParamsAlgorithm.TINYGEMM
Config 2:
  packing_format: tile_packed_to_4d
  choose_qparams: hqq

and it seems to be running locally, the test_awq_functionality works for the hqq algorithm as well

namgyu-youn · 2025-10-10T17:48:26Z

for the script:

import torch
from torchao.quantization import Int4WeightOnlyConfig
from torchao.utils import _is_fbgemm_gpu_genai_available, torch_version_at_least

devices = ["cpu", "cuda"]
device_to_base_configs = {
    "cuda": [
        Int4WeightOnlyConfig(group_size=128),
        # Note: the functionality unit test doesn't work for hqq
        Int4WeightOnlyConfig(group_size=128, int4_packing_format="tile_packed_to_4d"),
        Int4WeightOnlyConfig(
            group_size=128,
            int4_packing_format="tile_packed_to_4d",
            int4_choose_qparams_algorithm="hqq",
        ),
    ],
}

for i, cfg in enumerate(device_to_base_configs["cuda"]):
    print(f"Config {i}:")
    print(f"  packing_format: {cfg.int4_packing_format}")
    print(f"  choose_qparams: {cfg.int4_choose_qparams_algorithm}")

I get:

Config 0:
  packing_format: Int4PackingFormat.PLAIN
  choose_qparams: Int4ChooseQParamsAlgorithm.TINYGEMM
Config 1:
  packing_format: tile_packed_to_4d
  choose_qparams: Int4ChooseQParamsAlgorithm.TINYGEMM
Config 2:
  packing_format: tile_packed_to_4d
  choose_qparams: hqq

and it seems to be running locally, the test_awq_functionality works for the hqq algorithm as well

Yeah it returns no fail for test_awq.py locally. We don't have to debug them, and can just add that test case, right?

Hmm... I truly remember getting the following error:

torchao/prototype/awq/core.py:91: in calculate_qparams
    q_out = F.linear(acc / scales, w, self.bias)
    ↓
torchao/quantization/quantize_/workflows/int4/int4_tensor.py:161
    res = torch.ops.fbgemm.bf16i4bf16_rowwise(...)
    ↓
RuntimeError: cutlass cannot initialize

But everything is resolved without any change...? Mamma mia 😕

jerryzh168 · 2025-10-10T17:58:37Z

for the script:

import torch
from torchao.quantization import Int4WeightOnlyConfig
from torchao.utils import _is_fbgemm_gpu_genai_available, torch_version_at_least

devices = ["cpu", "cuda"]
device_to_base_configs = {
    "cuda": [
        Int4WeightOnlyConfig(group_size=128),
        # Note: the functionality unit test doesn't work for hqq
        Int4WeightOnlyConfig(group_size=128, int4_packing_format="tile_packed_to_4d"),
        Int4WeightOnlyConfig(
            group_size=128,
            int4_packing_format="tile_packed_to_4d",
            int4_choose_qparams_algorithm="hqq",
        ),
    ],
}

for i, cfg in enumerate(device_to_base_configs["cuda"]):
    print(f"Config {i}:")
    print(f"  packing_format: {cfg.int4_packing_format}")
    print(f"  choose_qparams: {cfg.int4_choose_qparams_algorithm}")

I get:

Config 0:
  packing_format: Int4PackingFormat.PLAIN
  choose_qparams: Int4ChooseQParamsAlgorithm.TINYGEMM
Config 1:
  packing_format: tile_packed_to_4d
  choose_qparams: Int4ChooseQParamsAlgorithm.TINYGEMM
Config 2:
  packing_format: tile_packed_to_4d
  choose_qparams: hqq

and it seems to be running locally, the test_awq_functionality works for the hqq algorithm as well

Yeah it returns no fail for test_awq.py locally. We don't have to debug them, and can just add that test case, right?

Hmm... I truly remember getting the following error:

torchao/prototype/awq/core.py:91: in calculate_qparams
    q_out = F.linear(acc / scales, w, self.bias)
    ↓
torchao/quantization/quantize_/workflows/int4/int4_tensor.py:161
    res = torch.ops.fbgemm.bf16i4bf16_rowwise(...)
    ↓
RuntimeError: cutlass cannot initialize

But everything is resolved without any change...? Mamma mia 😕

yeah it runs for me locally without any changes. not sure why you get the error, seems like some env issues, are you using an H100 machine? fbgemm is only available in H100.

namgyu-youn · 2025-10-10T18:02:59Z

yeah it runs for me locally without any changes. not sure why you get the error, seems like some env issues, are you using an H100 machine? fbgemm is only available in H100.

Yes I use H100 for only allocated time (luckily now) and mostly use A100. Updated title and context for this change.

fix enum backend in int4 hqq packing format

6802b79

meta-cla bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Oct 10, 2025

namgyu-youn mentioned this pull request Oct 10, 2025

Add outlier in AWQ test cases #3106

Merged

namgyu-youn commented Oct 10, 2025

View reviewed changes

jerryzh168 reviewed Oct 10, 2025

View reviewed changes

revert unhelpful change

ad15bb9

jerryzh168 approved these changes Oct 10, 2025

View reviewed changes

jerryzh168 added the topic: not user facing Use this tag if you don't want this PR to show up in release notes label Oct 10, 2025

namgyu-youn changed the title ~~Fix enum backend for int4 choosing qprams algorithm~~ Add int4 choosing qprams algorithm test for AWQ Oct 10, 2025

Add int4 choosing qprams algorithm test for AWQ #3148

Are you sure you want to change the base?

Add int4 choosing qprams algorithm test for AWQ #3148

Uh oh!

Conversation

namgyu-youn commented Oct 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

pytorch-bot bot commented Oct 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/3148

❌ 2 New Failures

Uh oh!

namgyu-youn Oct 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jerryzh168 Oct 10, 2025

Choose a reason for hiding this comment

Uh oh!

jerryzh168 Oct 10, 2025

Choose a reason for hiding this comment

Uh oh!

jerryzh168 Oct 10, 2025

Choose a reason for hiding this comment

Uh oh!

jerryzh168 commented Oct 10, 2025

Uh oh!

namgyu-youn commented Oct 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

namgyu-youn commented Oct 10, 2025

Uh oh!

jerryzh168 commented Oct 10, 2025

Uh oh!

namgyu-youn commented Oct 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

namgyu-youn commented Oct 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jerryzh168 commented Oct 10, 2025

Uh oh!

namgyu-youn commented Oct 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jerryzh168 commented Oct 10, 2025

Uh oh!

namgyu-youn commented Oct 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

namgyu-youn commented Oct 10, 2025 •

edited

Loading

pytorch-bot bot commented Oct 10, 2025 •

edited

Loading

namgyu-youn Oct 10, 2025 •

edited

Loading

namgyu-youn commented Oct 10, 2025 •

edited

Loading

namgyu-youn commented Oct 10, 2025 •

edited

Loading

namgyu-youn commented Oct 10, 2025 •

edited

Loading

namgyu-youn commented Oct 10, 2025 •

edited

Loading

namgyu-youn commented Oct 10, 2025 •

edited

Loading