-
Notifications
You must be signed in to change notification settings - Fork 37
Insights: nod-ai/shark-ai
Overview
14 Pull requests merged by 10 people
-
Add an e2e test for the shortfin flux pipeline
#998 merged
Feb 26, 2025 -
Add toy grok numerical tests
#999 merged
Feb 25, 2025 -
[sharktank] Coerce paged attention args' dtype to avoid mismatch
#994 merged
Feb 25, 2025 -
Add a toy llama test for numerics
#997 merged
Feb 24, 2025 -
[iree auto bump workflow] Put the bumped-to version in pr title
#996 merged
Feb 24, 2025 -
Bump the github-actions group with 3 updates
#993 merged
Feb 24, 2025 -
Bump IREE requirement pins to 3.3.0rc20250215
#973 merged
Feb 24, 2025 -
[sharktank] restore custom matmul kernel
#896 merged
Feb 21, 2025 -
Some dtype fixes for vae tests
#989 merged
Feb 21, 2025 -
Don't use GGUF but convert T5 directly from Hugging Face
#967 merged
Feb 20, 2025 -
[shortfin llm] Simplify interface between llm specific code and fastapi webapp
#985 merged
Feb 19, 2025 -
Explicitly check for None when using prefill
attn_mask
#983 merged
Feb 19, 2025 -
[shortfin] Implement async alloc/dealloc of buffers.
#507 merged
Feb 19, 2025
11 Pull requests opened by 8 people
-
Enable llama fp8 masked_flash_attention 8
#984 opened
Feb 19, 2025 -
Shortfin LLM Direct-to-batcher tests
#987 opened
Feb 20, 2025 -
Cache shortfin llm integration test model artifacts in `/shark-cache` if it exists
#988 opened
Feb 20, 2025 -
Moved sharktank runner to ossci cluster
#990 opened
Feb 21, 2025 -
Sharded integration tests
#995 opened
Feb 24, 2025 -
Fix llama fp8 benchmark test and remove decomposed benchmarking tests
#1000 opened
Feb 25, 2025 -
[sharktank] Enable f8 model perplexity tests
#1001 opened
Feb 25, 2025 -
[sharktank] Update perplexity README and enable torch attention-kernel
#1002 opened
Feb 25, 2025 -
Bump IREE requirement pins to 3.3.0rc20250225
#1003 opened
Feb 25, 2025 -
Initial pipeline parallelism support
#1008 opened
Feb 25, 2025 -
Refactor BatcherProcess and GenerateService between shortfin apps
#1009 opened
Feb 25, 2025
5 Issues closed by 4 people
-
SDXL model (shortfin_apps.sd.server) does not behave as its help says
#992 closed
Feb 25, 2025 -
[tracking] iree bump to 0225 failures
#1004 closed
Feb 25, 2025 -
[NOT-BUG] The purpose of this project
#755 closed
Feb 25, 2025 -
Refresh user guide for 3.1 release
#724 closed
Feb 24, 2025 -
Data-dependent VaeFluxDecoderTest fail on HEAD
#982 closed
Feb 21, 2025
3 Issues opened by 1 person
-
Sharktank Data-Dependent tests reporting workgroup distribution verification errors
#1007 opened
Feb 25, 2025 -
SDXL int8 export in Sharktank Model Integration Tests timing out
#1006 opened
Feb 25, 2025
15 Unresolved conversations
Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.
-
Make get_iree_devices read IREE_DEVICE env var if provided
#891 commented on
Feb 25, 2025 • 2 new comments -
[WIP] LLM Server Release v3.3.0
#921 commented on
Feb 19, 2025 • 0 new comments -
Job timeouts in "CI - sharktank / Data-dependent Tests" after updating IREE versions
#888 commented on
Feb 19, 2025 • 0 new comments -
Shortfin assumes that function results do not alias outside allocations
#980 commented on
Feb 19, 2025 • 0 new comments -
Streamline LLM import/compile/serve user experience
#691 commented on
Feb 21, 2025 • 0 new comments -
Test both pinned and unpinned versions of IREE dependencies
#760 commented on
Feb 24, 2025 • 0 new comments -
[Tuner] Automatic merging of tuner-generated td specs
#810 commented on
Feb 25, 2025 • 0 new comments -
[Tuner] Improving ease of use for the tuner
#814 commented on
Feb 25, 2025 • 0 new comments -
[tuner] Candidate specs should be annonated with default unit attr
#816 commented on
Feb 25, 2025 • 0 new comments -
Enable tokenizers in shortfin packages on Linux x86_64.
#688 commented on
Feb 19, 2025 • 0 new comments -
Add [pinned, unpinnned] matrix to ci_eval.yaml.
#767 commented on
Feb 24, 2025 • 0 new comments -
[DNM] cherry pick fp8 attn nonsense with hack cream
#907 commented on
Feb 25, 2025 • 0 new comments -
[CI][sharktank] Move Sharktank Data-Dependent Tests to OSSCI Cluster
#932 commented on
Feb 20, 2025 • 0 new comments -
[sharktank][ci] Re-enable t5 data-dependent tests
#969 commented on
Feb 24, 2025 • 0 new comments -
[shortfin-sd] Add exports and support for scheduled unet, batch sizes.
#972 commented on
Feb 26, 2025 • 0 new comments