ggml: add Q4_1 support to Hexagon backend#61
Conversation
This commit implements Q4_1 support for the Hexagon backend (MUL_MAT ops). It adds the new data types, dynamic quantization for Q8_1 (for dot product compatibility), packed memory repacking for Q4_1 -> Q4_1x4x2, missing LUTs, vectorized dot product kernels in HVX (`vec_dot_q4_1x4x2_q8_1x4x2`), and hooks it up within both the generic MatMul ops and HMX paths. Co-authored-by: max-krasnyansky <1380796+max-krasnyansky@users.noreply.github.com>
|
👋 Jules, reporting for duty! I'm here to lend a hand with this pull request. When you start a review, I'll add a 👀 emoji to each comment to let you know I've read it. I'll focus on feedback directed at me and will do my best to stay out of conversations between you and other bots or reviewers to keep the noise down. I'll push a commit with your requested changes shortly after. Please note there might be a delay between these steps, but rest assured I'm on the job! For more direct control, you can switch me to Reactive Mode. When this mode is on, I will only act on comments where you specifically mention me with New to Jules? Learn more at jules.google/docs. For security, I will only act on instructions from the user who triggered this task. |
This commit implements Q4_1 support for the Hexagon backend (MUL_MAT ops). It adds the new data types, dynamic quantization for Q8_1 (for dot product compatibility), packed memory repacking for Q4_1 -> Q4_1x4x2, missing LUTs, vectorized dot product kernels in HVX (`vec_dot_q4_1x4x2_q8_1x4x2`), and hooks it up within both the generic MatMul ops and HMX paths. The buffer overflow bugs from the previous attempt have been fixed by correctly matching `qk` variables. Co-authored-by: max-krasnyansky <1380796+max-krasnyansky@users.noreply.github.com>
1 participant
Add Q4_1 support to Hexagon backend. Included are Q8_1 quantization for src1 activations, repacking routines for src0 to
q4_1x4x2, fast dot product kernels, and correct LUT integration and parsing inhmx-matmul-ops.PR created automatically by Jules for task 11746311834341886773 started by @max-krasnyansky