Skip to content

Pull towards Junji accross the Flags model - #1626

Open
sheikhsajid69 wants to merge 912 commits into
FlagOpen:JUNJIE99-patch-1from
medusaa-oss:master
Open

sheikhsajid69 wants to merge 912 commits into
FlagOpen:JUNJIE99-patch-1from
medusaa-oss:master

Conversation

@sheikhsajid69

Copy link
Copy Markdown

speed goes 4x but utilization comes more; output increases by 20%;

545999961 and others added 30 commits November 21, 2024 11:18
Fix air-bench eval bugs: AIRBenchEvalArgs
Fix air-bench eval bugs: AIRBenchEvalArgs
update code and README for scripts
hanhainebula and others added 30 commits March 26, 2026 20:15
fix(m3): make bf16 inference/train loading safe and unify numpy conversion
fix(eval): pass use_bf16 through AbsEvalRunner.get_model and COIR evaluator model initialization
 feat: add use_mrl interface of finetune for embedder and truncate_dim for inference of embedder(MRL)
…ting support

  - pass domain_for_pseudo_moe through evaluation runner to embedder initialization
  - add new decoder-only embedder implementation for pseudo MoE models
feat(embedder): Add decoder-only pseudo_moe inference with domain routing support
chore: update version to 1.4.0 in setup.py
…ma-import

Fix Gemma reranker imports with Transformers v5
Fix Transformers 5 Trainer compatibility
fix(reranker): support transformers v5 tokenizer preparation
fix(reranker): normalize model mapping names
Fix Transformers 5 tokenizer compatibility
feat: Add hard negative mining script

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.