rebuild llama-ts#13039

Closed

mryvae wants to merge 33 commits intoggml-org:masterfrom

lanhin:dev_pim_rebuild

Commits on Nov 25, 2024

Add parallelised gemv op
Yinan
committed

Commits on Nov 26, 2024

Commits on Nov 27, 2024

Fix build error
Yinan
committed

Commits on Dec 4, 2024

Utility for tensor export/import
Yinan
committed
Fix and add llama-ts for tensor import and dump test
Yinan
committed
Add mul_add for 1st element of two quantified tensors
Yinan
committed

Commits on Dec 18, 2024

Build goes well for host app as PIM_KERNEL enabled.
Yinan
committed

Commits on Dec 19, 2024

Add dpu_main.c for gemv computing on DPU.
Yinan
committed

Commits on Jan 11, 2025

UPMEM gemv op runs for a single call in each generate iteartion. But not works correctly with 4\32\64 DPUs.
Yinan
committed

Commits on Jan 13, 2025

More friendly log print; Tensor differ function; Bug fix for weight transfer to DPUs; Transfer API replace by dpu_broadcast_to.
Yinan
committed

Commits on Jan 15, 2025

Bug fix for DPU kernel weight offset and layer id for multiple layers.
Yinan
committed

Commits on Jan 23, 2025

Update PIM_README.md and some formats polished.
Yinan
committed

Commits on Jan 26, 2025

Update PIM_README.md
Yinan
committed

Commits on Feb 8, 2025

Refact ts.cpp as a scaffold for dpu kernel
Yinan
committed

Commits on Feb 20, 2025

完成dpu的多线程并行 & 增加tensor的误差对比函数
 nonestack
committed

Commits on Mar 5, 2025

添加了部分注释
 nonestack
committed

Commits on Mar 31, 2025

Merge pull request #1 from lanhin/dev_pim_multithread
lanhin
authored

Commits on Apr 10, 2025

添加测试cpu的执行时间
 nonestack
committed

Commits on Apr 14, 2025

Merge pull request #2 from lanhin/dev_pim_multithread
lanhin
authored
Turn off tensor_export by a temp macro TENSOR_EXPORT
lanhin
committed
Fix compile when undefine PIM_KERNEL
lanhin
committed
add PIM_DEBUG_PERF_PRINT

mryvae
authored and
lanhin
committed
Multi-threads for PIM calculation.
lanhin
committed
Fix ts compile when PIM_KERNEL is off
lanhin
committed
Open offset print for 1st time when call PIM functions.
lanhin
committed
use row_segment for supporting 16 tasklets

mryvae
authored and
lanhin
committed
use mul_table_int4_int8

mryvae
authored and
lanhin
committed

Commits on Apr 16, 2025

Merge pull request #3 from lanhin/tensor_export_turnoff
lanhin
authored

Commits on Apr 17, 2025

add pim-llm-framework
mryvae
committed

Commits on Apr 18, 2025

Commits on Apr 21, 2025

add ts-multi-thread
mryvae
committed