-
Notifications
You must be signed in to change notification settings - Fork 1.3k
Pull requests: FlashML-org/FreeToken
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
fix(swa): keep the prompt head's windowed KV so fan-out can reuse a shared prefix
#488
opened Sep 15, 2026 by
huhoo
Loading…
fix(server): hoist system messages to front for templates requiring system-first
#487
opened Sep 15, 2026 by
rafaeldrincon
Loading…
feat(scheduler): interleave decode steps into long prefill runs
#484
opened Sep 15, 2026 by
huhoo
Loading…
feat(server): expose POST /v1/tokenize endpoint for client-side token counting
feature
New feature or request
#477
opened Sep 14, 2026 by
rafaeldrincon
Loading…
feat(server): add --default-thinking-mode flag for server-wide thinking default
#476
opened Sep 14, 2026 by
rafaeldrincon
Loading…
fix(tokenizer): respect explicit enable_thinking=false even when tools present
#475
opened Sep 14, 2026 by
rafaeldrincon
Loading…
fix(engine): preserve greedy sampling in mixed batches
#471
opened Sep 14, 2026 by
taking-lying-flat
Contributor
Loading…
fix(models): preserve glm4 router correction bias in fp32
#469
opened Sep 14, 2026 by
taking-lying-flat
Contributor
Loading…
fix(engine): reject fa attention backend below sm_90 at config time
#468
opened Sep 14, 2026 by
SanjayS45
Loading…
fix(utils): tolerate coalesced msgpack frames in zmq pull queues
#466
opened Sep 14, 2026 by
Cyber-Marty
Loading…
fix(scheduler): match stop strings with incremental decoding
#464
opened Sep 14, 2026 by
taking-lying-flat
Contributor
Loading…
fix(server): loopback-bind the TP rendezvous store, add --dist-port
#461
opened Sep 13, 2026 by
shikhakath
Loading…
feat(models): support deepseek v4.1 with native fp8-fp4 kv storage
#460
opened Sep 13, 2026 by
ArqAlice
Loading…
docs(moe): report shared expert banks across replicas
#451
opened Sep 12, 2026 by
kuangrenaigc-stack
Loading…
feat(moe): owner-local expert parallelism and tensor parallelism
#447
opened Sep 11, 2026 by
leiyu1980
Loading…
fix(models): build o_proj row-parallel in the remaining column-parallel families
#440
opened Sep 11, 2026 by
gberasmus87
Loading…
fix(scheduler): read the prefill chunk cap from the pool, not a snapshot
#439
opened Sep 10, 2026 by
gberasmus87
Loading…
fix(server): resolve JSON Schema refs and union types in tool arguments
#435
opened Sep 10, 2026 by
Retloldin
Loading…
fix(glm4_moe): load compressed-tensors NVFP4 expert checkpoints
#430
opened Sep 10, 2026 by
gberasmus87
Loading…
qwen4_exp: build o_proj row-parallel, as every other family does
#429
opened Sep 10, 2026 by
gberasmus87
Loading…
Previous Next
ProTip!
Adding no:label will show everything without a label.