Barrett Kepler

Similar here but with example-00-helloworld.ninja:48 and linux release64 ``` ninja --version 1.10.1 ```

Sounds awesome! But as for someone who just want to try it out I'm not able to figure it out fast enough how to use offloading Could you please add...

> Could this even get the full one working on a much smaller GPU VRAM too ? the full mega checkpoint instead of just the fp? I guess so: >...

Got something that look like the same. Here's my record: ```lua global record Array add: function(self: Array, val: T) get: function(self: Array, index: number): T metamethod __call: function(self: Array, type:...

@hishamhm Thank you for your reply! Sorry for posting it here - in the wrong issue. Still, it's not an issue as it turns out. This actually works. Would be...

> Use go install, as the help text suggests. > […](#) > On Mon, 15 Aug 2022, 17:00 Barrett Kepler, ***@***.***> wrote: > [stage-0 12/34] RUN go get -u github.com/valyala/quicktemplate...

It's been a long time since initial discussion, but the problem seems to persist In my case I'm trying to mirror https://nvidia.github.io/nvidia-docker/ubuntu18.04/amd64 My command is ```bash aptly -config=/app/aptly.json -keyring=/app/trustedkeys.gpg -architectures="amd64"...

> ..and GPU support! But I suppose the big problem is that the profiles might have to be per-runner as they all might have different hardware setups? So I guess...

Same Using [2025.10.9-pt2.8.0-cu12.8-update-entrypoint](https://hub.docker.com/layers/unsloth/unsloth/2025.10.9-pt2.8.0-cu12.8-update-entrypoint/images/sha256-998709111ce28908cf23bad558a1bc5748b9d35ff3b3cfa63ac01c773f9ab889) docker image If I try to load in torch.float16 the warnings are gone, but trainer fails with `AssertionError: wrong number of dimensions2 for op: torch.ops.bitsandbytes.int8_mixed_scaled_mm.default` Tried Llama-3.2-1B-Instruct,...