Readon

> Why is `pastValid()` implemented with `initial()` instead of `init()` ? > > See [`formal/package.scala#L20`](https://github.com/SpinalHDL/SpinalHDL/blob/42b22b3eea7db0d9570ba3d17c2828673c47dd3a/core/src/main/scala/spinal/core/formal/package.scala#L20) > > ```scala > def pastValid() : Bool = signalCache(ClockDomain.current -> "formal.pastValid")(RegNext(True) initial(False) setWeakName("formal_with_past")) >...

> About CI.. I don't think your pr #1455 did what you think it did? It is now using the `latest` tag that was last pushed/released to one year ago....

you can use the master version to test first.

> Ah, so I have to change the image tag that is used in the action yaml? Can do... that commit shall never enter dev though. I can also test...

Sorry for the late reply. I just want to release the docker image first then merge this before. However, it seems that ghdl would not release frequently. So we should...

Now all environment has been prepared, and I tried it on my own machine. One problem we need to concern about is that all ghdl supported formal verification runs in...

> > Is it possible to split the ghdl related test case into standalone CI jobs? > > Currently there is no difference made between default symbiyosys or ghdl formal...

Does llama.cpp support to offload experts to GPU standalone? If so we could offload all 8 shared experts to GPU as possible and leave the dynamically selected expert to CPU...

> looking at ktransformers, it seems like they have figured out which layers to load to GPU for improved performance: > > https://github.com/kvcache-ai/ktransformers/blob/main/ktransformers/optimize/optimize_rules/DeepSeek-V3-Chat.yaml > > Looks like this PR will...

I think this function might lead to misunderstanding, user might need to trace which signals are unassigned manually, which is hard. Give out info to explain which signals connected to...