[CI] Retire Weight Loading smoke tests (#57398)
Signed-off-by: mgoin <mgoin64@gmail.com>
Misha Goin committed
Sep 17, 2026 at 16:04 UTC
9a5bd373cfe59e236f2b32c30530f4dbbf24120d
10 files changed
-253
.buildkite/test-amd.yaml
-60
@@ -3335,36 +3335,6 @@ steps:
3335
commands:
3336
- pytest -v -s v1/e2e/test_hybrid_chunked_prefill.py
3337
3338
-#------------------------------------------------------ mi300 · weight_loading -------------------------------------------------------#
3339
-
3340
-- label: ":amd: (MI300) Weight Loading Multi-GPU" # TBD
3341
- timeout_in_minutes: 70
3342
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
3343
- dind: false
3344
- agent_pool: mi300_2
3345
- num_gpus: 2
3346
- working_dir: "/vllm-workspace/tests"
3347
- source_file_dependencies:
3348
- - vllm/
3349
- - tests/weight_loading
3350
- optional: true
3351
- commands:
3352
- - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-amd.txt
3353
-
3354
-- label: ":amd: (MI300) Weight Loading Multi-GPU (Large Models)" # TBD
3355
- timeout_in_minutes: 35
3356
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx942nightly, amdmi300]
3357
- dind: false
3358
- agent_pool: mi300_2
3359
- num_gpus: 2
3360
- optional: true
3361
- working_dir: "/vllm-workspace/tests"
3362
- source_file_dependencies:
3363
- - vllm/
3364
- - tests/weight_loading
3365
- commands:
3366
- - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-large-amd.txt
3367
-
3338
#########################################################################################################################################
3339
# #
3340
# MI355 (gfx950) tests #
@@ -4719,36 +4689,6 @@ steps:
4689
- if [ "$$BUILDKITE_PARALLEL_JOB" = "1" ]; then pytest -v -s "v1/e2e/spec_decode/acceptance_rates/mtp_other/test_mtp.py::test_gemma4_mtp_acceptance_lengths[False]"; fi
4690
- if [ "$$BUILDKITE_PARALLEL_JOB" = "2" ]; then pytest -v -s "v1/e2e/spec_decode/acceptance_rates/mtp_other/test_mtp.py::test_gemma4_mtp_acceptance_lengths[True]"; fi
4691
4722
-#------------------------------------------------------ mi355 · weight_loading -------------------------------------------------------#
4723
-
4724
-- label: ":amd: (MI355) Weight Loading Multi-GPU" # TBD
4725
- timeout_in_minutes: 70
4726
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4727
- dind: false
4728
- agent_pool: mi355_2
4729
- num_gpus: 2
4730
- working_dir: "/vllm-workspace/tests"
4731
- source_file_dependencies:
4732
- - vllm/
4733
- - tests/weight_loading
4734
- optional: true
4735
- commands:
4736
- - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-amd.txt
4737
-
4738
-- label: ":amd: (MI355) Weight Loading Multi-GPU (Large Models)" # TBD
4739
- timeout_in_minutes: 30
4740
- mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
4741
- dind: false
4742
- agent_pool: mi355_2
4743
- working_dir: "/vllm-workspace/tests"
4744
- num_gpus: 2
4745
- optional: true
4746
- source_file_dependencies:
4747
- - vllm/
4748
- - tests/weight_loading
4749
- commands:
4750
- - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-large-amd.txt
4751
-
4692
#----------------------------------------------------------- mi355 · misc ------------------------------------------------------------#
4693
4694
- label: ":amd: (MI355) Regression" # TBD
.buildkite/test_areas/weight_loading.yaml
deleted
-38
@@ -1,38 +0,0 @@
1
-group: Weight Loading
2
-depends_on:
3
- - image-build
4
-steps:
5
-- label: ":nvidia: (L4) Weight Loading Multi-GPU"
6
- key: weight-loading-multiple-gpu
7
- timeout_in_minutes: 50
8
- working_dir: "/vllm-workspace/tests"
9
- device: l4
10
- num_devices: 2
11
- optional: true
12
- source_file_dependencies:
13
- - vllm/
14
- - "!vllm/distributed/kv_transfer/"
15
- - tests/weight_loading
16
- commands:
17
- - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models.txt
18
- mirror:
19
- amd:
20
- label: ":amd: (MI355) Weight Loading Multi-GPU"
21
- dind: false
22
- device: mi355_2
23
- timeout_in_minutes: 70
24
- depends_on:
25
- - image-build-amd
26
- commands:
27
- - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-amd.txt
28
-
29
-# - label: Weight Loading Multiple GPU - Large Models # optional
30
-# working_dir: "/vllm-workspace/tests"
31
-# num_devices: 2
32
-# device: a100
33
-# optional: true
34
-# source_file_dependencies:
35
-# - vllm/
36
-# - tests/weight_loading
37
-# commands:
38
-# - bash weight_loading/run_model_weight_loading_test.sh -c weight_loading/models-large.txt
.github/CODEOWNERS
-1
@@ -110,7 +110,6 @@
110
/tests/entrypoints/llm/test_struct_output_generate.py @mgoin @aarnphm @arpera
111
/tests/v1/structured_output @mgoin @aarnphm @arpera
112
/tests/v1/core @WoosukKwon @robertgshaw2-redhat @njhill @ywang96 @alexm-redhat @heheda12345 @ApostaC @orozery @ivanium
113
-/tests/weight_loading @mgoin @youkaichao @yewentao256
113
/tests/lora @jeejeelee
114
/tests/models/language/generation/test_hybrid.py @tdoublep @tomeras91
115
/tests/v1/kv_connector/nixl_integration @NickLucche
tests/kernels/quantization/test_rocm_compressed_tensors_w4a16.py
-1
@@ -17,7 +17,6 @@ from vllm.platforms import current_platform
17
@pytest.mark.parametrize(
18
"model_path",
19
[
20
- # Listed in tests/weight_loading/models.txt
20
"nm-testing/tinyllama-oneshot-w4a16-group128-v2",
21
],
22
)
tests/weight_loading/models-amd.txt
deleted
-18
@@ -1,18 +0,0 @@
1
-# quantization, model, revision[, minimum capability[, safetensors load strategy]]
2
-fp8, amd/Meta-Llama-3.1-8B-Instruct-FP8-KV, main
3
-None, amd/Llama-3.2-1B-Instruct-FP8-KV, main
4
-fp8, amd/Mixtral-8x7B-Instruct-v0.1-FP8-KV, main, 80, prefetch
5
-gptq, robertgshaw2/zephyr-7b-beta-channelwise-gptq, main
6
-gptq, TheBloke/Llama-2-7B-GPTQ, main
7
-gptq, TheBloke/TinyLlama-1.1B-Chat-v1.0-GPTQ, gptq-8bit--1g-actorder_True
8
-compressed-tensors, nm-testing/tinyllama-oneshot-w8w8-test-static-shape-change, main
9
-compressed-tensors, nm-testing/tinyllama-oneshot-w8-channel-a8-tensor, main
10
-compressed-tensors, nm-testing/tinyllama-oneshot-w8a8-dynamic-token-v2, main
11
-compressed-tensors, nm-testing/tinyllama-oneshot-w8a8-channel-dynamic-token-v2, main
12
-compressed-tensors, nm-testing/tinyllama-oneshot-w4a16-group128-v2, main
13
-compressed-tensors, nm-testing/tinyllama-oneshot-w8a16-per-channel, main
14
-compressed-tensors, nm-testing/Meta-Llama-3-8B-FP8-compressed-tensors-test, main
15
-compressed-tensors, nm-testing/Phi-3-mini-128k-instruct-FP8, main
16
-compressed-tensors, neuralmagic/Phi-3-medium-128k-instruct-quantized.w4a16, main
17
-awq, casperhansen/mixtral-instruct-awq, main
18
-fp8, neuralmagic/Meta-Llama-3-8B-Instruct-FP8-KV, main
tests/weight_loading/models-large-amd.txt
deleted
-4
@@ -1,4 +0,0 @@
1
-# quantization, model, revision[, minimum capability[, safetensors load strategy]]
2
-fp8, amd/Meta-Llama-3.1-70B-Instruct-FP8-KV, main, 80, prefetch
3
-None, microsoft/phi-4, main, 80, prefetch
4
-fp8, amd/Mixtral-8x22B-Instruct-v0.1-FP8-KV, main, 80, prefetch
tests/weight_loading/models-large.txt
deleted
-6
@@ -1,6 +0,0 @@
1
-compressed-tensors, nm-testing/Mixtral-8x7B-Instruct-v0.1-W4A16-quantized, main
2
-compressed-tensors, nm-testing/Mixtral-8x7B-Instruct-v0.1-W4A16-channel-quantized, main
3
-compressed-tensors, nm-testing/Mixtral-8x7B-Instruct-v0.1-W8A16-quantized, main
4
-gptq, TheBloke/Mixtral-8x7B-v0.1-GPTQ, main
5
-awq_marlin, casperhansen/deepseek-coder-v2-instruct-awq, main
6
-compressed-tensors, RedHatAI/Llama-4-Scout-17B-16E-Instruct-quantized.w4a16, main
tests/weight_loading/models.txt
deleted
-18
@@ -1,18 +0,0 @@
1
-gptq, robertgshaw2/zephyr-7b-beta-channelwise-gptq, main
2
-gptq, TheBloke/Llama-2-7B-GPTQ, main
3
-gptq, LnL-AI/TinyLlama-1.1B-Chat-v1.0-GPTQ-4bit, main
4
-gptq, TheBloke/TinyLlama-1.1B-Chat-v1.0-GPTQ, gptq-8bit--1g-actorder_True
5
-gptq, TechxGenus/gemma-2b-it-GPTQ, main
6
-compressed-tensors, nm-testing/tinyllama-oneshot-w8w8-test-static-shape-change, main
7
-compressed-tensors, nm-testing/tinyllama-oneshot-w8-channel-a8-tensor, main
8
-compressed-tensors, nm-testing/tinyllama-oneshot-w8a8-dynamic-token-v2, main
9
-compressed-tensors, nm-testing/tinyllama-oneshot-w8a8-channel-dynamic-token-v2, main
10
-compressed-tensors, nm-testing/tinyllama-oneshot-w4a16-group128-v2, main
11
-compressed-tensors, nm-testing/tinyllama-oneshot-w8a16-per-channel, main
12
-compressed-tensors, nm-testing/Meta-Llama-3-8B-FP8-compressed-tensors-test, main
13
-compressed-tensors, nm-testing/Phi-3-mini-128k-instruct-FP8, main
14
-compressed-tensors, neuralmagic/Phi-3-medium-128k-instruct-quantized.w4a16, main
15
-#compressed-tensors, mgoin/DeepSeek-Coder-V2-Lite-Instruct-FP8, main
16
-awq, casperhansen/mixtral-instruct-awq, main
17
-awq_marlin, casperhansen/mixtral-instruct-awq, main
18
-fp8, neuralmagic/Meta-Llama-3-8B-Instruct-FP8-KV, main
tests/weight_loading/run_model_weight_loading_test.sh
deleted
-58
@@ -1,58 +0,0 @@
1
-#!/bin/bash
2
-SUCCESS=0
3
-
4
-while getopts "c:" OPT; do
5
- case ${OPT} in
6
- c )
7
- CONFIG="$OPTARG"
8
- ;;
9
- \? )
10
- usage
11
- exit 1
12
- ;;
13
- esac
14
-done
15
-
16
-
17
-IFS=$'\n' read -d '' -r -a MODEL_CONFIGS < "$CONFIG"
18
-
19
-for MODEL_CONFIG in "${MODEL_CONFIGS[@]}"
20
-do
21
- if [[ $MODEL_CONFIG == \#* ]]; then
22
- echo "=== SKIPPING MODEL: $MODEL_CONFIG ==="
23
- continue
24
- fi
25
-
26
- LOCAL_SUCCESS=0
27
- IFS=', ' read -r -a array <<< "$MODEL_CONFIG"
28
-
29
- echo "=== RUNNING MODEL: $MODEL_CONFIG ==="
30
-
31
- unset WEIGHT_LOADING_TEST_SAFETENSORS_LOAD_STRATEGY
32
- export QUANTIZATION=${array[0]}
33
- export MODEL_NAME=${array[1]}
34
- export REVISION=${array[2]}
35
- # If array length is larger than 3, then MIN_CAPABILITY is provided
36
- if [ ${#array[@]} -gt 3 ]; then
37
- export MIN_CAPABILITY=${array[3]}
38
- fi
39
- if [ ${#array[@]} -gt 4 ]; then
40
- export WEIGHT_LOADING_TEST_SAFETENSORS_LOAD_STRATEGY=${array[4]}
41
- fi
42
- pytest -s weight_loading/test_weight_loading.py || LOCAL_SUCCESS=$?
43
-
44
- if [[ $LOCAL_SUCCESS == 0 ]]; then
45
- echo "=== PASSED MODEL: ${MODEL_CONFIG} ==="
46
- else
47
- echo "=== FAILED MODEL: ${MODEL_CONFIG} ==="
48
- fi
49
-
50
- SUCCESS=$((SUCCESS + LOCAL_SUCCESS))
51
-
52
-done
53
-
54
-if [ "${SUCCESS}" -eq "0" ]; then
55
- exit 0
56
-else
57
- exit 1
58
-fi
tests/weight_loading/test_weight_loading.py
deleted
-49
@@ -1,49 +0,0 @@
1
-# SPDX-License-Identifier: Apache-2.0
2
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
-
4
-import os
5
-
6
-import pytest
7
-import torch
8
-
9
-from vllm.platforms import current_platform
10
-from vllm.utils.mem_constants import GiB_bytes
11
-
12
-MAX_MODEL_LEN = 1024
13
-MODEL_NAME = os.environ.get(
14
- "MODEL_NAME", "robertgshaw2/zephyr-7b-beta-channelwise-gptq"
15
-)
16
-REVISION = os.environ.get("REVISION", "main")
17
-QUANTIZATION = os.environ.get("QUANTIZATION", "gptq_marlin")
18
-MIN_CAPABILITY = os.environ.get("MIN_CAPABILITY", "80")
19
-SAFETENSORS_LOAD_STRATEGY = os.environ.get(
20
- "WEIGHT_LOADING_TEST_SAFETENSORS_LOAD_STRATEGY"
21
-)
22
-
23
-
24
-@pytest.mark.skipif(
25
- MODEL_NAME == "casperhansen/deepseek-coder-v2-instruct-awq", reason="OOM in the CI"
26
-)
27
-@pytest.mark.skipif(
28
- not current_platform.has_device_capability(int(MIN_CAPABILITY)),
29
- reason="Current system does not have minimum capability.",
30
-)
31
-def test_weight_loading(vllm_runner):
32
- """Test parameter weight loading with tp>1."""
33
- # MoE models need fp16.
34
- NEEDS_FP16 = QUANTIZATION == "gptq"
35
- with vllm_runner(
36
- model_name=MODEL_NAME,
37
- revision=REVISION,
38
- dtype=torch.half if NEEDS_FP16 else "auto",
39
- quantization=None if QUANTIZATION == "None" else QUANTIZATION,
40
- max_model_len=MAX_MODEL_LEN,
41
- tensor_parallel_size=2,
42
- safetensors_load_strategy=SAFETENSORS_LOAD_STRATEGY,
43
- # Generating 20 tokens needs a tiny KV cache, while sizing and
44
- # allocating a full-size one dominates this test on large devices.
45
- kv_cache_memory_bytes=2 * GiB_bytes,
46
- ) as model:
47
- output = model.generate_greedy("Hello world!", max_tokens=20)
48
- print(output)
49
- assert output