Skip to content
Community Bot

[bug] LoRA adapter init depends on tensor parallel size: linear_in is initialized per shard with a fan-dependent method #21895

Sign in to view logs

[bug] LoRA adapter init depends on tensor parallel size: linear_in is initialized per shard with a fan-dependent method

[bug] LoRA adapter init depends on tensor parallel size: linear_in is initialized per shard with a fan-dependent method #21895

Workflow file for this run

.github/workflows/community-bot.yml at 407baf0
name: Community Bot
on:
issues:
types: [opened, edited, reopened, closed, deleted]
issue_comment:
types: [created, edited, deleted]
jobs:
community-bot:
uses: NVIDIA-NeMo/FW-CI-templates/.github/workflows/_community_bot.yml@98ea77e930f3e1b0f97a1ce5255e12c407e9d0d4 # v1.8.7
with:
community_project_id: ${{ vars.COMMUNITY_PROJECT_ID }}
app-id: ${{ vars.BOT_ID }}
secrets:
BOT_KEY: ${{ secrets.BOT_KEY }}