-
Notifications
You must be signed in to change notification settings - Fork 0
All issues
Issue creation is restricted in this repository
Issues
is:issue state:open
is:issue state:open
Search results
llm-inference/gemma: three tests use JUnit 4 Assume under the Jupiter engine — aborts record as failures
bugSomething isn't workingSomething isn't workingStatus: Open.#261 In SKaiNET-developers/SKaiNET-transformers;FunctionGemma export materializes the tied embedding twice (77% of the weight archive)
bugSomething isn't workingSomething isn't workingenhancementNew feature or requestNew feature or requestStatus: Open.#260 In SKaiNET-developers/SKaiNET-transformers;- Status: Open.#249 In SKaiNET-developers/SKaiNET-transformers;
- Status: Open.#248 In SKaiNET-developers/SKaiNET-transformers;
- Status: Open.#235 In SKaiNET-developers/SKaiNET-transformers;
- Status: Open.#234 In SKaiNET-developers/SKaiNET-transformers;
- Status: Open.#194 In SKaiNET-developers/SKaiNET-transformers;
- Status: Open.#184 In SKaiNET-developers/SKaiNET-transformers;
Eager NATIVE_OPTIMIZED: keep Q8_0 matmul weights packed (pre-transpose marker) so gemma fits + runs fast on the SL2610
enhancementNew feature or requestNew feature or requestStatus: Open.#178 In SKaiNET-developers/SKaiNET-transformers;Packed SIMD matmul kernel for Q5_1 / Q5_0 (keep NATIVE_OPTIMIZED fast, not dequant-to-FP32)
enhancementNew feature or requestNew feature or requestStatus: Open.#170 In SKaiNET-developers/SKaiNET-transformers;- Status: Open.#118 In SKaiNET-developers/SKaiNET-transformers;
- Status: Open.#104 In SKaiNET-developers/SKaiNET-transformers;