Post
758
Dataset Optimizer (free) + fine tuning + continual learning platform with zero forgetting.
try it at modelbrew.ai
try it at modelbrew.ai
Kiran N PRO
Fourwheels2512
AI & ML interests
None yet
Recent Activity
new activity 2 days ago
blog-explorers/README:The Next Evolution of AI: From Passive Models to Autonomous SystemsOrganizations
The Next Evolution of AI: From Passive Models to Autonomous Systems
ππ₯ 2
40
#15 opened 2 days ago
by 
oncody
posted an update 2 days ago
published an article 2 days ago
Article
Free dataset Optimizer/cleaner + Finetuning + continual learning that actually doesn't forget.
β’
1
published an article 17 days ago
Article
RAG (vs) Fine-Tuning + continual learning
posted an update 24 days ago
Post
144
Zero Forgetting in LLM Fine-Tuning β 4 Benchmarks, All Domains Retained
We tested sequential fine-tuning on Mistral-7B across 4 independent benchmarks (5, 4, 5, and 8 domains).
Standard LoRA forgets 38β49% of prior knowledge per domain. Our continual learning adapter: -0.17% drift.
The Salesforce 5-domain test showed positive backward transfer β the model got better at old domains as it learned new ones (retention BERTScore: 0.889 β 0.907).
No replay buffers. No EWC. No knowledge distillation. Spectral norm locked at 1.0. Naive LoRA crashed at gradient norm 263. Ours: under 6.
Interactive benchmark dashboard: Fourwheels2512/zero-forgetting-benchmarks
Live product (free tier): https://mhc-finetune-saas-zrtokzlkbnue9zsk7jfgad.streamlit.app
Patent pending. 7 technical reports. 196 automated tests. Solo founder, 6 months of R&D.
We tested sequential fine-tuning on Mistral-7B across 4 independent benchmarks (5, 4, 5, and 8 domains).
Standard LoRA forgets 38β49% of prior knowledge per domain. Our continual learning adapter: -0.17% drift.
The Salesforce 5-domain test showed positive backward transfer β the model got better at old domains as it learned new ones (retention BERTScore: 0.889 β 0.907).
No replay buffers. No EWC. No knowledge distillation. Spectral norm locked at 1.0. Naive LoRA crashed at gradient norm 263. Ours: under 6.
Interactive benchmark dashboard: Fourwheels2512/zero-forgetting-benchmarks
Live product (free tier): https://mhc-finetune-saas-zrtokzlkbnue9zsk7jfgad.streamlit.app
Patent pending. 7 technical reports. 196 automated tests. Solo founder, 6 months of R&D.
updated a
Space 24 days ago
published a
Space 24 days ago
CRMA: Stable Fine-Tuning + Continual Learning for Small LLMs
#2 opened about 1 month ago
by Fourwheels2512
CRMA Fine-Tuner: QLoRA + gradient stability adapter (CRMA + ZClip + PiSSA) for TinyLlama, Gemma, Mistral β ablation results
#1153 opened about 1 month ago
by Fourwheels2512
Built a stability adapter on top of TRL's SFTTrainer (CRMA + ZClip) β sharing ablation results
#1 opened about 1 month ago
by Fourwheels2512
Welcome to CRMA Fine-Tuner β what it does and how to use it
#1 opened about 1 month ago
by Fourwheels2512
updated a
Space about 1 month ago
published a
Space about 1 month ago