Skip to content
This repository was archived by the owner on May 21, 2025. It is now read-only.

Enable ipex xpu op #2

Closed
wants to merge 22 commits into from
Closed

Enable ipex xpu op #2

wants to merge 22 commits into from

Conversation

jiqing-feng
Copy link

@jiqing-feng jiqing-feng commented Apr 24, 2025 •
edited
Loading

  1. update dockerfile: I installed the current repo in the dockerfile. This docker file will not introduce any cuda libs! I am not in favor of using docker-compose as the absolute path may be changed in the future. We'd better have a dockerfile that includes all our needed libs, so we can just build and run this dockerfile.
    BTW, I didn't see any bnb_intel path which is in the entrypoint and docker compose yaml file.
  2. Enable ipex op for cpu and xpu.

It could pass all transformers tests

@matthewdouglas @Titus-von-Koeller

@jiqing-feng
install the current repo in dockerfile
4d88df2 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng jiqing-feng marked this pull request as draft April 29, 2025 08:24
@jiqing-feng jiqing-feng changed the title Install the current repo in dockerfile Enable ipex op Apr 29, 2025
@jiqing-feng
register cpu/xpu ops
59f8f97 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
add de/quantize blockwise op
ae76703 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
fix dequantize blockwise register
803b429 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
fix quantize 4bit
0bc2a21 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
fix quantize 4bit out
bb22037 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng jiqing-feng changed the title Enable ipex op Enable ipex xpu op May 8, 2025
@jiqing-feng jiqing-feng marked this pull request as ready for review May 8, 2025 07:26
jiqing-feng added 16 commits May 8, 2025 12:39
@jiqing-feng
fix xpu ops and precommit
8b02dcf 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
fix op name
51f260f 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
add xpu ops
dfa8235 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
fix tuple
ae25c78 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
fix gemv 4bit
84bace7 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
fix quant table device
e3fd20f 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
fix dequantize blockwise
a151b86 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
fix dequantize 4bit
7a2175f 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
register ipex op
41359ea 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
fix dequantize_nf4_ipex
20d4bdf 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
simplify gemv
30fa5c0 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
check xpu
9141674 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
fix quantize blockwise output shape
7eb98c4 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
fix quant_storage bf16
a7d314e 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
fix xpu ops
6b31eb3 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng
fix xpu ops
c23930c 
Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
@jiqing-feng jiqing-feng mentioned this pull request May 15, 2025
Open
@jiqing-feng
Copy link
Author

jiqing-feng commented May 21, 2025 •
edited
Loading

Closed, xpu backend will be implemented in bitsandbytes main repo.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
1 participant
@jiqing-feng