genji970/pretraining_llm_from_scratch
AI training piepline from scratch. From data instruction building llm train dataset to training method.
ex intern(ai, multimodal) @ NTU | Korea university
Repositories
genji970/facebook-paper_harness-inference-scale-agent
unofficial implementation of Meta(facebook) Ai paper "Scaling test time compute for agentic coding"
genji970/sglang
SGLang is a high-performance serving framework for large language models and multimodal models.
genji970/SRM-LoRA
official implementation of "SRM-LoRA: Sub-Riemannian-Metric Updates for Mitigating LLM Hallucination in Low-Rank Adaptation" ICML2026 Workshop FoGen
genji970/math_code_hallucination
A record of how LLMs may silently replace difficult mathematical components with simpler computational surrogates when math and code are combined in one prompt.