Commit Graph

7 Commits

Author SHA1 Message Date
Michael Yang
7460259eb3 feat: qwen3 embed (#12301)
* cleanup

* use pooling.TypeNone

* pooling test

* qwen3 embed
2025-09-18 15:50:32 -07:00
Michael Yang
564b558c92 fix(llama): other llama flavours (#12308)
* fix(llama): rope scale

* spm llama

* skip moe models

* cleanup
2025-09-17 12:12:21 -07:00
Michael Yang
ad95d5b30b use split activations when possible (#12293)
* use ggml_*_split activations when possible

* forward qkv
2025-09-16 09:51:19 -07:00
Michael Yang
6f7117145f batch: use tensors for outputs (#12185)
this cleans up the model interface slightly without too much impact in
other areas
2025-09-15 14:33:06 -07:00
Michael Yang
2e77aa1ae7 use nn.Linear in place of ml.Tensor (#11049)
while nn.Linear.Forward isn't applicable for sparse MLP, it's still
a nice container for the tensors
2025-06-11 12:10:15 -07:00
Jesse Gross
1f371ea92f ml: Panic rather than return error on tensor allocation failure
FromFloatSlice and FromIntSlice return an error if the shape doesn't
match the passed data or if memory can't be allocated. Since these
are inputs, the memory being allocated is system memory rather than VRAM.

In many cases, the caller can't really handle the error and panics.

Empty and Zeros directly panic if they can't allocate memory.

This makes things consistent by panicing for the first two cases,
removing a fair amount of error handling code. This is also consistent
with how Go typically handles these situations.
2025-05-22 14:38:09 -07:00
Michael Yang
e0ed984cde feat: qwen3 dense and sparse models (#10708)
* feat: qwen3 dense
* feat: qwen3moe
* fix llama4 moe
2025-05-21 10:21:07 -07:00