Re Releasing Our Blackwell
The author ships a verifier for LLM-generated PTX kernels targeting NVIDIA Blackwell — a serious attempt to fix the weakest link in the 'AI as a Compiler' idea: trusting machine-written assembly. The timing is right, since labs are already pushing models to write kernels directly, and bypassing compiler abstractions promises real gains on scarce GPU capacity.… more
In our view the hard part is exactly what the verifier claims to check — symbolic correctness and race conditions at the PTX level are notoriously difficult to prove at scale, so independent stress tests should precede any talk of replacing compiler stacks. Still, if the guarantees hold, verified hardware-level code generation becomes a genuinely practical path to more efficient models.
We’re releasing our Blackwell kernel verifier, building on our recently released work on "AI as a Compiler." In this approach, an LLM generates optimized PTX for a given workload, and our verifier checks it for correctness. By skipping intermediate compiler layers, we enable finer-grained control over the hardware at the PTX level, paving the way f…
Source:
X — Azalia Mirhoseini
Discussion (0)