Draft: Integrate TQ2_0 into vulkan#22

Closed

makaveli10 wants to merge 23 commits intotetherto:temp-latestfrom

makaveli10:vineet/vulkan_tq2_0

Commits on Jul 30, 2025

Add lora finetuning from adapter
makaveli10
committed

Commits on Aug 1, 2025

Add: create new lora adapter for target modules to finetune if no lora is provided
makaveli10
committed

Commits on Aug 5, 2025

Fix identical loss over epochs; fix garbage lora initization
makaveli10
committed

Commits on Aug 6, 2025

Commits on Aug 7, 2025

Add dequantization to out_prod cuda kernel
makaveli10
committed

Commits on Aug 8, 2025

Update README with finetune-lora
makaveli10
committed

Commits on Aug 13, 2025

Vulkan: add support for fp32 OUT_PROD op
Italo Nicola
committed
CPU: add support for fp16_fp32 OUT_PROD op
Italo Nicola
committed
Vulkan: add support for f16_f32 OUT_PROD op
Italo Nicola
committed
Vulkan: Add Q4_0/Q8_0 OUT_PROD Vulkan support

zoq
authored and
Italo Nicola
committed

Commits on Aug 14, 2025

Commits on Aug 28, 2025

Commits on Sep 4, 2025

Fix out_prod; vulkan ci issues
makaveli10
committed
Add GEGLU backward (Vulkan) to enable Gemma training.

zoq
authored and
makaveli10
committed

Commits on Sep 5, 2025

ggml-vulkan: Add TQ2_0 dequantize and mul_mat vec
makaveli10
committed

Commits on Sep 9, 2025

ggml-vulkan: Enable coopmat support for Android
makaveli10
committed

Commits on Sep 11, 2025

ggml-vulkan: Add mul_mm path for TQ2_0
makaveli10
committed