AI Potluck
Model components / Fine-tuning code

Lamini

AMD

Enterprise fine-tuning stack acquired by AMD at Advancing AI 2025 (June 2025) and rolled into AMD's ROCm/AI software organization; the former CEO Sharon Zhou is now VP of AI at AMD. Picked when teams running on AMD Instinct MI300/MI325 accelerators want a first-party fine-tuning path optimized for ROCm; Lamini's Mixture of Memory Experts (MoME) technique trains thousands of LoRA-style memory experts per model to drive ~95% factual recall with ~10x fewer hallucinations than vanilla SFT. The discrete lamini.ai SaaS has been wound down; the technology now anchors AMD's enterprise fine-tuning story alongside ROCm 6.x and AMD's growing data-center AI revenue ramp.

Lamini enterprise LLM fine-tuning/inference platform; acquired by AMD June 2025 (founders incl. Sharon Zhou joined AMD). The fine-tuning/memory-tuning engine runs on Lamini's hosted servers; only the Python client SDK (github.com/lamini-ai/lamini, Apache-2.0) is open. Confirmed via repo + coverage June 2026.

Openness

1 high confidence
1.0
license
Proprietary
source
closed
engine
closed(hosted memory-tuning/fine-tuning service on Lamini/AMD servers)
client-sdk
open(Apache-2.0, lamini-ai/lamini)
data
closed
training-code
closed
access
API-key-gated

Proprietary managed fine-tuning service; the only open artifact is a thin API client SDK, so the product itself is closed.

Adoption

not assessed

No verified user/usage/download figure for the hosted platform; post-AMD-acquisition integration into AMD's AI stack obscures any standalone signal. Declining to assign a level rather than infer from press.

Capability

not assessed

Vendor markets memory-tuning (claimed ~95% factual accuracy, 32x compression) but these are unverified vendor claims with no independent benchmark; closed engine is not assessable on the recipe's feature matrix. Not scoring rather than fabricate.

Unchanged since 2026-07-30 (last edited, not re-checked)