Skip to content

Initial fix for token corruption when batching #375

Initial fix for token corruption when batching

Initial fix for token corruption when batching #375

Workflow file for this run

# Copyright 2024 Advanced Micro Devices, Inc.
#
# Licensed under the Apache License v2.0 with LLVM Exceptions.
# See https://llvm.org/LICENSE.txt for license information.
# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
name: CI - sharktank perplexity short
on:
workflow_dispatch:
pull_request:
push:
branches:
- main
concurrency:
# A PR number if a pull request and otherwise the commit hash. This cancels
# queued and in-progress runs for the same PR (presubmit) or commit
# (postsubmit). The workflow name is prepended to avoid conflicts between
# different workflows.
group: ${{ github.workflow }}-${{ github.event.number || github.sha }}
cancel-in-progress: true
jobs:
test_perplexity_iree:
name: "Llama3.1 8B FP16"
strategy:
matrix:
version: [3.11]
runs-on: [llama-mi300x-3]
fail-fast: false
runs-on: ${{matrix.runs-on}}
defaults:
run:
shell: bash
env:
VENV_DIR: ${{ github.workspace }}/.venv
steps:
- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
- name: "Setting up Python"
id: setup_python
uses: actions/setup-python@0b93645e9fea7318ecaed2b359559ac225c90a2b # v5.3.0
with:
python-version: ${{matrix.version}}
- name: Create Python venv
run: python -m venv ${VENV_DIR}
- name: Install sharktank deps
run: |
source ${VENV_DIR}/bin/activate
python -m pip install --no-compile --upgrade pip
# Note: We install in three steps in order to satisfy requirements
# from non default locations first. Installing the PyTorch CPU
# wheels saves multiple minutes and a lot of bandwidth on runner setup.
pip install --no-compile -r pytorch-cpu-requirements.txt
pip install --no-compile -r requirements.txt -r sharktank/requirements-tests.txt -e sharktank/
# Install latest iree-tubrine.
pip install --no-compile -f https://iree.dev/pip-release-links.html --src deps \
-e "git+https://github.com/iree-org/iree-turbine.git#egg=iree-turbine"
# Try with the latest IREE nightly releases, not what iree-turbine pins.
# We could also pin to a known working or stable version.
# This should eventually stabilize. Do the best we can for now.
pip install -f https://iree.dev/pip-release-links.html --upgrade --pre \
iree-base-compiler \
iree-base-runtime
pip freeze
- name: Run perplexity test with vmfb
run: |
source ${VENV_DIR}/bin/activate
pytest -n 8 -v -s sharktank/tests/evaluate/perplexity_iree_test.py --bs=5 --iree-device='hip://6' --iree-hip-target=gfx942 --iree-hal-target-backends=rocm --llama3-8b-f16-model-path=/data/llama3.1/8b/llama8b_f16.irpa --llama3-8b-tokenizer-path=/data/llama3.1/8b/tokenizer_config.json