Not a task — informational heads-up for the maintainers, nothing to implement. Close it freely.
Hi — I curate StackMap, a hand-curated knowledge graph of open-source AI/agent tools. Every entry is human-reviewed: a summary, an opinionated note on when to use it (and when not), and typed edges to what it pairs with or competes against — with the why written down.
Model-Optimizer earned a place on the map:
How we mapped it:
- pairs well with
vllm — ModelOpt is the compress half and vLLM the serve half of the same pipeline: quantized checkpoints export directly into vLLM, which is where the FP8/NVFP4 throughput win actually shows up.
- alternative to
airllm — Both make an oversized model fit on hardware it shouldn't. AirLLM streams layers and explicitly refuses to quantize, preserving quality at the cost of latency; ModelOpt compresses the weights and keeps throughput.
If you want it visible in your README, every mapped repo has a badge:
[](https://stackmap.shipwithai.xyz/repos/NVIDIA/model-optimizer?utm_source=badge)
It renders as StackMap | ↔ 7 typed links and links to the page above. Static JSON — the count updates itself as the map changes, nothing for you to maintain.
If any of this misrepresents the project — an edge you'd dispute, a sharper "when NOT to use" — reply here and a human fixes it. And feel free to just close this issue; it's only a heads-up.
Hi — I curate StackMap, a hand-curated knowledge graph of open-source AI/agent tools. Every entry is human-reviewed: a summary, an opinionated note on when to use it (and when not), and typed edges to what it pairs with or competes against — with the why written down.
Model-Optimizer earned a place on the map:
How we mapped it:
vllm— ModelOpt is the compress half and vLLM the serve half of the same pipeline: quantized checkpoints export directly into vLLM, which is where the FP8/NVFP4 throughput win actually shows up.airllm— Both make an oversized model fit on hardware it shouldn't. AirLLM streams layers and explicitly refuses to quantize, preserving quality at the cost of latency; ModelOpt compresses the weights and keeps throughput.If you want it visible in your README, every mapped repo has a badge:
It renders as
StackMap | ↔ 7 typed linksand links to the page above. Static JSON — the count updates itself as the map changes, nothing for you to maintain.If any of this misrepresents the project — an edge you'd dispute, a sharper "when NOT to use" — reply here and a human fixes it. And feel free to just close this issue; it's only a heads-up.