AI Paper Atlas

Research map / Continual learning / Continual fine-tuning of LLMs

Towards Cold-Start Drafting and Continual Refining: A Value-Driven Memory Approach with Application to NPU Kernel Synthesis

Yujie Zheng, Zhuo Li, Shengtao Zhang, Jiaqian Wang, Junjie Sheng, Junchi Yan, Weinan Zhang, Ying Wen, Bo Tang, Muning Wen

ICML 2026 · Continual learning › Continual fine-tuning of LLMs · approach: Continual pretraining of language models · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add value-driven Q-based retrieval instead of heuristic similarity retrieval (ablation)EvoKernel (GPT-5.2) · Ascend C KernelBench L2correctness (%): 67% -> 77%Improved
add cross-task shared memory over per-operator refinement (ablation)EvoKernel (GPT-5.2) · Ascend C KernelBench L2accuracy (%): 3.0 -> 76.0Improved
add cross-task shared memory over per-operator refinement (ablation)EvoKernel (GPT-5.2) · Ascend C KernelBench L1accuracy (%): +49 pp AccImproved
add L1 memory before L2 (curriculum transfer) versus L2 from scratchEvoKernel (GPT-5.2) · Ascend C KernelBench L2correctness (%): 34 -> 64Improved
use L1 then L2 ordering instead of mixed L1+L2 streamEvoKernel (GPT-5.2) · Ascend C KernelBench L2correctness (%): 53 -> 64Improved
add GPT-5.2 memory to DeepSeek-V3.2DeepSeek-V3.2 · Held-out 50 operatorscorrectness (%): 6% -> 58%Improved
add GPT-5.2 memory to Qwen3-Coder-30BQwen3-Coder-30B · Held-out 50 operatorscorrectness (%): 4% -> 32%Improved
use EvoKernel instead of Refinement on weaker backbone DeepSeek-V3.2DeepSeek-V3.2 · Ascend C KernelBenchcompilation rate (%): 35.0 -> 29.0Hurt
use EvoKernel instead of Pass@k on weaker backbone Qwen3-Coder-30BQwen3-Coder-30B · Ascend C KernelBenchaccuracy (%): 4.0 -> 5.5Improved
use EvoKernel instead of Codex agentGPT-5.2 · Ascend C KernelBenchaccuracy (%): 46.0 -> 83.0Improved
use stronger backbone GPT-5.2 versus weaker models for experience-driven methodEvoKernel · Ascend C KernelBenchaccuracy (%): 4.0 -> 83.0Mixed
add iterative refinement after first correct draftEvoKernel · Ascend C KernelBench (159 operators)speedup over first draft: median 3.60xImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy