Repository navigation
Drop orphan draft references from the external-draft learning-rate note - #2978
Conversation
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info
📝 Walkthrough
Merge Risk: ⚪ Minimal · up to
|
The note cited 1.0e-5 (4B) and 2.0e-6 (0.8B), the only mention of either draft in the doc, so a reader met two rates for two models the page never introduces. The Configuration table above already states the rate this section used. All the note needs to add is the recommendation to sweep. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Signed-off-by: Peter Hicks <phicks@nvidia.com>
b3491e0 to
c7ab779
Compare
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #2978 +/- ##
=======================================
Coverage 79.49% 79.49%
=======================================
Files 660 660
Lines 74752 74752
=======================================
Hits 59421 59421
Misses 15331 15331
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
|
What changed
The learning-rate note in
doc/external_draft.md's Reference Configuration cited1.0e-5 (4B)and2.0e-6 (0.8B). Those were the only mention of either draft anywhere in the doc, so a reader meets two rates for two models the page never introduces and cannot tell which, if either, applies to them.The Reference Configuration documents one pair: Llama-3.2-1B drafting for Qwen3.8-27B at
2.0e-5.Why not just delete the note
It carries the only "sweep for your own pair" guidance in the doc. L181 says a swapped draft wants a higher rate than a matched one, but never says to sweep. So the note keeps the part that generalises and attaches it to the rate this section actually documents.
Testing
Documentation only, one file, no code.
pre-commitpasses includingmarkdownlint-cli2. Verified no4B,0.8B,1.0e-5or2.0e-6reference remains in the doc.Follow-up to #2688.
🤖 Generated with Claude Code
Summary by CodeRabbit