A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 8, 00:00 UTC
[CI] Remove failing prime-rl integration test (#34843)
Signed-off-by: mgoin <mgoin64@gmail.com> Signed-off-by: Michael Goin <mgoin64@gmail.com>
Michael Goin committed
Feb 20, 2026 at 13:17 UTC
fac1507f03c78d8717853c8a15ad1d887d71cc1d
3 files changed
-106
.buildkite/scripts/run-prime-rl-test.sh
deleted
-64
@@ -1,64 +0,0 @@
1
-#!/bin/bash
2
-# SPDX-License-Identifier: Apache-2.0
3
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
4
-
5
-# Setup script for Prime-RL integration tests
6
-# This script prepares the environment for running Prime-RL tests with nightly vLLM
7
-
8
-set -euo pipefail
9
-
10
-SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
11
-REPO_ROOT="$(cd "${SCRIPT_DIR}/../.." && pwd)"
12
-PRIME_RL_REPO="https://github.com/PrimeIntellect-ai/prime-rl.git"
13
-PRIME_RL_DIR="${REPO_ROOT}/prime-rl"
14
-
15
-if command -v rocm-smi &> /dev/null || command -v rocminfo &> /dev/null; then
16
- echo "AMD GPU detected. Prime-RL currently only supports NVIDIA. Skipping..."
17
- exit 0
18
-fi
19
-
20
-echo "Setting up Prime-RL integration test environment..."
21
-
22
-# Clean up any existing Prime-RL directory
23
-if [ -d "${PRIME_RL_DIR}" ]; then
24
- echo "Removing existing Prime-RL directory..."
25
- rm -rf "${PRIME_RL_DIR}"
26
-fi
27
-
28
-# Install UV if not available
29
-if ! command -v uv &> /dev/null; then
30
- echo "Installing UV package manager..."
31
- curl -LsSf https://astral.sh/uv/install.sh | sh
32
- source "$HOME"/.local/bin/env
33
-fi
34
-
35
-# Clone Prime-RL repository at specific branch for reproducible tests
36
-PRIME_RL_BRANCH="integ-vllm-main"
37
-echo "Cloning Prime-RL repository at branch: ${PRIME_RL_BRANCH}..."
38
-git clone --branch "${PRIME_RL_BRANCH}" --single-branch "${PRIME_RL_REPO}" "${PRIME_RL_DIR}"
39
-cd "${PRIME_RL_DIR}"
40
-
41
-echo "Setting up UV project environment..."
42
-export UV_PROJECT_ENVIRONMENT=/usr/local
43
-ln -s /usr/bin/python3 /usr/local/bin/python
44
-
45
-# Remove vllm pin from pyproject.toml
46
-echo "Removing vllm pin from pyproject.toml..."
47
-sed -i '/vllm==/d' pyproject.toml
48
-
49
-# Sync Prime-RL dependencies
50
-echo "Installing Prime-RL dependencies..."
51
-uv sync --inexact && uv sync --inexact --all-extras
52
-
53
-# Verify installation
54
-echo "Verifying installations..."
55
-uv run python -c "import vllm; print(f'vLLM version: {vllm.__version__}')"
56
-uv run python -c "import prime_rl; print('Prime-RL imported successfully')"
57
-
58
-echo "Prime-RL integration test environment setup complete!"
59
-
60
-echo "Running Prime-RL integration tests..."
61
-export WANDB_MODE=offline # this makes this test not require a WANDB_API_KEY
62
-uv run pytest -vs tests/integration/test_rl.py -m gpu
63
-
64
-echo "Prime-RL integration tests completed!"
.buildkite/test-amd.yaml
-29
@@ -1622,21 +1622,6 @@ steps:
1622
- uv pip install --system 'gpt-oss[eval]==0.0.5'
1623
- VLLM_ROCM_USE_AITER_MHA=0 VLLM_ROCM_USE_AITER=1 VLLM_USE_AITER_UNIFIED_ATTENTION=1 pytest -s -v tests/evals/gpt_oss/test_gpqa_correctness.py --model openai/gpt-oss-20b --metric 0.58
1624
1625
-##### RL Integration Tests #####
1626
-- label: Prime-RL Integration Test # 15min
1627
- mirror_hardwares: [amdexperimental]
1628
- agent_pool: mi325_2
1629
- # grade: Blocking
1630
- timeout_in_minutes: 30
1631
- optional: true
1632
- num_gpus: 2
1633
- working_dir: "/vllm-workspace"
1634
- source_file_dependencies:
1635
- - vllm/
1636
- - .buildkite/scripts/run-prime-rl-test.sh
1637
- commands:
1638
- - bash .buildkite/scripts/run-prime-rl-test.sh
1639
-
1625
##### EPLB Accuracy Tests #####
1626
- label: DeepSeek V2-Lite Accuracy
1627
mirror_hardwares: [amdexperimental, amdproduction]
@@ -3201,20 +3186,6 @@ steps:
3186
- uv pip install --system 'gpt-oss[eval]==0.0.5'
3187
- VLLM_ROCM_USE_AITER_MHA=0 VLLM_ROCM_USE_AITER=1 VLLM_USE_AITER_UNIFIED_ATTENTION=1 pytest -s -v tests/evals/gpt_oss/test_gpqa_correctness.py --model openai/gpt-oss-20b --metric 0.58
3188
3204
-##### RL Integration Tests #####
3205
-- label: Prime-RL Integration Test # 15min
3206
- mirror_hardwares: [amdexperimental]
3207
- agent_pool: mi355_2
3208
- timeout_in_minutes: 30
3209
- optional: true
3210
- num_gpus: 2
3211
- working_dir: "/vllm-workspace"
3212
- source_file_dependencies:
3213
- - vllm/
3214
- - .buildkite/scripts/run-prime-rl-test.sh
3215
- commands:
3216
- - bash .buildkite/scripts/run-prime-rl-test.sh
3217
-
3189
##### EPLB Accuracy Tests #####
3190
- label: DeepSeek V2-Lite Accuracy
3191
mirror_hardwares: [amdexperimental, amdproduction]
.buildkite/test_areas/e2e_integration.yaml
-13
@@ -28,16 +28,3 @@ steps:
28
working_dir: "/vllm-workspace"
29
commands:
30
- bash .buildkite/scripts/scheduled_integration_test/qwen30b_a3b_fp8_block_ep_eplb.sh 0.8 200 8020 2 1
31
-
32
-- label: Prime-RL Integration (2 GPUs)
33
- timeout_in_minutes: 30
34
- optional: true
35
- soft_fail: true
36
- num_devices: 2
37
- working_dir: "/vllm-workspace"
38
- source_file_dependencies:
39
- - vllm/
40
- - .buildkite/scripts/run-prime-rl-test.sh
41
- commands:
42
- - nvidia-smi
43
- - bash .buildkite/scripts/run-prime-rl-test.sh