Load json config#27226

Closed

yugong333 wants to merge 114 commits intovllm-project:mainfrom

dcmaddix:load_json_config

Commits on Jul 23, 2025

Ignore yapf config

wuchen
authored and
wcwuwc
committed
to support fusedmoe with lora

wuchen
authored and
wcwuwc
committed
added test_moe_lora_align_sum.py

wuchen
authored and
wcwuwc
committed
fixed typing import error

wuchen
authored and
wcwuwc
committed

Commits on Jul 24, 2025

Commits on Jul 25, 2025

Commits on Jul 29, 2025

Commits on Aug 6, 2025

Commits on Aug 11, 2025

Commits on Aug 12, 2025

Commits on Aug 14, 2025

moe fixes for deepseek r1
vangheem
committed

Commits on Aug 22, 2025

added add_lora_fused_moe method for punica_gpu

banjuede
authored and
wcwuwc
committed

Commits on Aug 29, 2025

merge branch 'main' into myfusedmoe

wcwuwc
authored and
banjuede
committed
merge branch with added MergedReplicatedLinearWithLoRA

wcwuwc
authored and
banjuede
committed

Commits on Sep 19, 2025

Fix error with building vllm from source based on https://github.com/vllm-project/vllm/pull/23929
Danielle Robinson
committed
Add function and SupportLoRA to enable multi-lora for gpt-oss
Danielle Robinson
committed
Modify packed_modules_mapping to extract q,k,v projs and fuse with lora projection
Danielle Robinson
committed
Fix bug that accessed field of quant_config even if it was None
Danielle Robinson
committed

Commits on Sep 22, 2025

Clean up file
Danielle Robinson
committed
Clean up file
Danielle Robinson
committed

Commits on Sep 23, 2025

Debug moe layer lora support for gpt-oss
Danielle Robinson
committed
specific to match weights config
Danielle Robinson
committed

Commits on Sep 24, 2025

unmerge gate_proj_up for lora support
Danielle Robinson
committed

Commits on Sep 25, 2025

Add mask to avoid out of bounds memory access for expert_ids and sorted_token_ids in fused_moe_lora triton kernel
Danielle Robinson
committed
Clean up debugging printing
Danielle Robinson
committed
drop base layer terms from lora adapter
Danielle Robinson
committed
update file name to experts to match adapter
Danielle Robinson
committed
update expert naming format for lora module
Danielle Robinson
committed

Commits on Sep 26, 2025

include experts/base_layer
Danielle Robinson
committed
Add experts.base_layer and experts for gate_up and down proj
Danielle Robinson
committed
Adding in gnovack's splitting moe weights into each per expert
Danielle Robinson
committed

Commits on Sep 27, 2025

Add MoE bias terms back will now run inference with valid output:
Danielle Robinson
committed
clean up comments
Danielle Robinson
committed
clean up comments
Danielle Robinson
committed
Adding gnovack's changes to support FSDP file format where experts.base_layer corresponds to the gate_up_proj and experts corresponds to the down_proj. Tested and works with various prompts for TP=…
Danielle Robinson
committed
Clean up comments
Danielle Robinson
committed
Revert back expert_params to take 3 proj weights
Danielle Robinson
committed
Revert get_expert_mapping function to take gate_proj, down_proj and up_proj to be compatible with deepseek PR
Danielle Robinson
committed
clean up comments
Danielle Robinson
committed
Remove fused_qkv_a_proj since we only need the qkv projs
Danielle Robinson
committed

Commits on Sep 30, 2025

fix index in expert ids
Danielle Robinson
committed

Commits on Oct 1, 2025

remove printing log
Danielle Robinson
committed
Merge pull request #2 from dcmaddix/gpt_oss_multi_lora
wcwuwc
authored

Commits on Oct 3, 2025

fixing conflict
wcwuwc
committed

Commits on Oct 4, 2025

Commits on Oct 6, 2025

Commits on Oct 7, 2025

Adding support for lora activation adapter
Danielle Robinson
committed
add support for moe_sum through fused_marlin_moe
Danielle Robinson
committed
Merge pull request #6 from dcmaddix/marlin_experts_mxfp4
wcwuwc
authored

Commits on Oct 9, 2025

Commits on Oct 10, 2025

cleanup and update activation_func
Danielle Robinson
committed
Fix conflict
jeejeelee
committed
cleanup
Danielle Robinson
committed
Add TODO comment on refactor moe_lora_align_sum_kernels with moe_align_sum_kernels
Danielle Robinson
committed

Commits on Oct 11, 2025

Commits on Oct 12, 2025

Commits on Oct 13, 2025

Commits on Oct 14, 2025

Merge branch 'main' into fused_moe_lora
jeejeelee
authored
add unit testing
linitra24
committed
Add test
jeejeelee
committed
Adding test for gptoss
Danielle Robinson
committed
add test
linitra24
committed
Update to default_act_function and pass as callable
Danielle Robinson
committed
remove None type
Danielle Robinson
committed
Update default_act function signature
Danielle Robinson
committed
Merge pull request #16 from dcmaddix/remove_torch_ops
wcwuwc
authored
clean up the code
wcwuwc
committed
represent invalid expert ids with -1
wcwuwc
committed

Commits on Oct 15, 2025

Merge pull request #15 from dcmaddix/test_pr_final
wcwuwc
authored

Commits on Oct 16, 2025

Add test
jeejeelee
committed

Commits on Oct 17, 2025

Adding json config loading for fused_moe_lora kernel
Ubuntu
committed