SP8192 + Adaptive Hessian-Sensitivity GPTQ Clipping — 1.0822 bpb#1689
Open
chris-colinsky wants to merge 1 commit intoopenai:mainfrom
Open
SP8192 + Adaptive
Hessian-Sensitivity GPTQ Clipping — 1.0822 bpb#1689chris-colinsky wants to merge 1 commit intoopenai:mainfrom
chris-colinsky wants to merge 1 commit intoopenai:mainfrom
Conversation
resouer
added a commit
to resouer/parameter-golf
that referenced
this pull request
Apr 18, 2026
The open frontier is too inconsistent to optimize against directly, but openai#1689 contributes a quantization-only idea that is at least surface-coherent. This ports its adaptive Hessian-sensitivity clip schedule into the trusted W23 control line so we can measure the quantization effect without changing the training or eval-time mechanism stack. Constraint: Keep the W23 training and eval surfaces unchanged; modify only GPTQ clip selection Rejected: Import the entire openai#1689 public surface | it is weaker than the trusted baseline and would reintroduce public-surface drift Confidence: medium Scope-risk: narrow Reversibility: clean Directive: Judge W81 as a quantization-only donor lane, not as a public-family reproduction Tested: python3 -m py_compile train_gpt.py Not-tested: Remote train/eval on Lepton
6 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
preserving compression budget)
3-Seed Results
Attribution
Built on @clarkkev's SP8192+GPTQ+SDClip base (PR #1394), with depth recurrence (@dexhunter), parallel residuals (@Robby955, @msisovic), and hyperparameter
tuning (@X-Abhishek-X, PR #1445).