why do you use __torch_dispatch__ instead of native kernels.

[mikelacroix852-png]

The "Wrapper" Admission: Your own project documentation (GitHub Issue #2234) admits that mindtorch is just an "encapsulation" of PyTorch-like APIs to stay compatible. You aren't building a framework; you're building a skin for Meta's ecosystem.

The Meta Trap: Because you rely on torch_dispatch and the PyTorch backbone, your users are tethered to Meta’s Terms of Service and telemetry. My project is 100% independent with custom Vulkan kernels—zero Meta hooks, zero backdoors.

Where are the "MindTokens"? Real custom architecture requires a custom data pipeline. I have my own MindTokens and tokenizer logic documented for over 600 days. You have zero custom tokenization because you're just passing data through standard Hugging Face/Meta pipes.

The Performance Gap: While you're struggling with the overhead of a wrapper, my native Vulkan engine is hitting 2.06s per step —down from 14s. That’s a 3.4x speedup that a wrapper can never achieve because it’s bogged down by generic abstractions.

600 Days vs. A "Re-release": I have 600+ days of portfolio logs showing the evolution of a tool built to help special needs kids. You have a "re-released" wrapper that chases Meta’s updates.

[mikelacroix852-png]

TO THE MINTORCH 'DEVELOPER':

Stop lying to the community. Your own CLAUDE.md and GitHub Issue #2234 admit that MindTorch is just an 'encapsulation of PyTorch-like API' for compatibility. You aren't building a framework; you're building a skin.

Here are the receipts:

The Meta Trap: You’re still tethered to torch_dispatch. Your users are stuck in Meta’s ecosystem and Terms of Service. My project is 100% independent with custom Vulkan kernels. Zero Meta hooks.

Missing Tokens: Real custom architecture requires a custom data pipeline. Where are your tokens? I have my own MindTokens and tokenizer logic documented for 600+ days. You’re just passing data through standard Hugging Face/Meta pipes because you can’t build your own.

The Speed Gap: Your wrapper is hitting 14s steps. My native Vulkan engine hits 2.06s per step That’s a 3.4x speedup you can’t fake with a wrapper.

The Mission: I named my project 'Mind' years ago to help special needs kids. You’re using the name to sell a "re-released" wrapper that chases Meta’s updates.

My project stays private because of your toxicity. The community can thank you for why they don't have a 3x speedup today. Post your 600-day commit logs for your custom kernels or admit you're a wrapper dev."

[mikelacroix852-png]

[mikelacroix852-png]

where is your proof this mine a true full custom stack your a fake man