-
03c28d2b15
fix(kernel): make Triton top-k/top-p sampling exact (#329)
main
cherry77-cloud
2026-09-03 10:01:57 +08:00
-
6eca2d7d2b
fix(models): detect nvfp4 experts behind a mixed-precision compressed-tensors format (#343)
Xiaoze Fan
2026-09-02 00:11:14 -07:00
-
a80b4d308a
fix(hf): download the shards the safetensors index names (#336)
Xiaoze Fan
2026-09-01 19:43:19 -07:00
-
a2538a428b
feat(glm5_next): support GLM-5.3-Flash (#332)
Xiaoze Fan
2026-09-01 15:18:15 -07:00
-
4c0bad3f56
feat(qwen4_exp): stream the PLE n-gram table from disk (#311)
Xiaoze Fan
2026-09-01 13:35:14 -07:00
-
e05cff83a0
perf(moe): route fused_topk through the in-repo triton router (#319)
Xiaoze Fan
2026-08-31 20:21:11 -07:00
-
3a20a79038
fix(kernel): unbreak the nightly kernel-cache wheel build (#310)
Xiaoze Fan
2026-08-30 22:43:50 -07:00
-
4b94bdc38a
docs: add SECURITY.md
Xiaoze Fan
2026-08-30 01:00:07 -07:00
-
58f4b9ec0e
fix(kernel): avoid row-wise _scaled_mm stall on sm_89 with torch<2.12 (#243)
Berni McCoy
2026-08-29 00:20:16 -04:00
-
a05c26543f
docs(models): add Qwen3.8-Flash-Next & Qwen3.8 27b
Xiaoze Fan
2026-08-28 15:49:00 -07:00
-
bd8f3d519a
feat(qwen4_exp): support Qwen3.8-Flash-Next (#257)
Xiaoze Fan
2026-08-28 15:33:15 -07:00
-
9ef3651309
chore(assets): update wechat group QR code
Xiaoze Fan
2026-08-26 00:41:24 -07:00
-
f7c31e92db
fix(fla): stop l2norm recompiling per token count
Xiaoze Fan
2026-08-25 05:36:19 +00:00
-
2757bb5f91
feat(server)!: add --gpu to choose the GPU on multi-GPU machines (#117)
Xiaoze Fan
2026-08-24 16:44:01 -07:00
-
bd372b630a
fix(checkpoint): FTW truncated shard raises OSError instead of silent garbage load (#25)
Lancer
2026-08-24 08:53:27 +08:00
-
e0a3bbc046
Merge pull request #112 from FlashML-org/feat/split-residency
Shuo Yang
2026-08-23 17:39:31 -07:00
-
-
831d38a66b
fix(moe): report the residency banks actually settle at
Xiaoze Fan
2026-08-23 23:06:56 +00:00
-
eebb3f53f3
feat(moe): estimate expert-bank bytes from the model config
Xiaoze Fan
2026-08-23 23:06:56 +00:00
-
c41833b717
feat(moe): per-layer host-bank residency -- split lock-CPU/pin-GPU layers, auto selection under capped pin quota
Xiaoze Fan
2026-08-23 08:15:17 +00:00
-
184a4f114d
build(kernel-cache): add sm_80 (A100/A800) to default arches (#75)
jasonfan
2026-08-22 23:38:42 -07:00
-
-
f0abe587a1
docs: add CONTRIBUTING.md
jasonfan
2026-08-22 22:22:15 -07:00
-
0ab982f109
chore(README): Add Community WeChat link to README
Shuo Yang
2026-08-20 11:43:47 -07:00
-
bc6b269cbc
chore(assets): Add FreeToken wechat group invite
Shuo Yang
2026-08-20 11:41:16 -07:00
-
9db1a39455
chore(release): 0.1.2
v0.1.2
Xiaoze Fan
2026-08-19 05:57:18 +00:00
-
3f02e3c5a8
docs(README): use absolute URLs so PyPI renders assets and links
jason-fxz
2026-08-18 20:49:53 -07:00
-
c0a4a60b2c
Merge branch 'dev/refine-readme': README overhaul + quick start guide
jason-fxz
2026-08-18 20:38:49 -07:00
-
-
9ecbc9f33f
docs(README): add a Desktop screenshot to Getting Started
jason-fxz
2026-08-18 20:36:08 -07:00
-
12ed695386
feat(stats): record per-request TTFT and report its mean
jason-fxz
2026-08-18 19:08:52 -07:00
-
586a6c16ce
perf(fp8): run per-tensor fp8 as W8A8 via scaled_mm (#7)
Xiaoze Fan
2026-08-18 18:04:39 -07:00
-
f4373f8a45
docs(README): add install commands to Getting Started
jason-fxz
2026-08-18 16:09:32 -07:00
-
5528af06a4
docs(README): add Citation
jason-fxz
2026-08-18 15:50:32 -07:00
-
fca27580a1
docs: refine README.md, add quick start guide
jason-fxz
2026-08-18 15:24:47 -07:00
-
-
5812fbd084
chore(README.md): add links.
Xiaoze Fan
2026-08-18 13:25:02 -07:00
-
6223cda1b0
Support muse glimmer (#4)
Shuo Yang
2026-08-18 12:26:05 -07:00
-
b86a904f86
build(release): pypi-ready wheels, metadata, and publish workflow (#6)
Xiaoze Fan
2026-08-17 23:03:44 -07:00
-
-
76edc16282
build(release): pypi-ready wheels, metadata, and publish workflow
Xiaoze Fan
2026-08-14 07:25:47 +00:00
-
-
711325d17e
fix(server): stop API server when backend dies
jason-fxz
2026-08-16 14:34:17 -07:00
-
2635246ba2
ci: move nightly wheels to 11:00 UTC
jason-fxz
2026-08-16 11:56:22 -07:00
-
d7cd6ed88b
[Feature] unify reasoning effort on one shared scale (#5)
Xiaoze Fan
2026-08-15 23:40:52 -07:00
-
a2572083e9
feat(launch): support launch deepseek harness (#3)
Xiaoze Fan
2026-08-14 23:50:44 -07:00
-
4f510f1ba6
feat(server): enable reasoning_effort on /v1/chat/completions (#2)
Xiaoze Fan
2026-08-14 00:02:32 -07:00
-
c28e93a8c0
chore(assets): update icons & logos
jasonfan
2026-08-13 18:09:01 -07:00
-
fdfcacd491
chore: Enhance README with logo
Xiaoze Fan
2026-08-13 16:28:39 -07:00
-
9dc3ff4ca5
Support minimax m3 (#1)
Shuo Yang
2026-08-13 16:22:41 -07:00
-
17af154bb5
feat(assets): add dark/light logos
jasonfan
2026-08-13 16:10:28 -07:00
-
3e5214d4a3
build(deps): pin apache-tvm-ffi, flashlib, triton to exact versions
jasonfan
2026-08-13 14:51:26 -07:00
-
97eb4e25b5
Merge branch 'main' of github.com:FlashML-org/FreeToken
jasonfan
2026-08-13 13:27:51 -07:00
-
-
a8669e4950
chore(assets): add freetoken icon & logo
jasonfan
2026-08-13 13:26:53 -07:00
-
ea5348b4e3
feat(kvcache): checkpoint decode state at the tool-call opener token
Xiaoze Fan
2026-08-13 18:32:02 +00:00
-
ee54f638fd
fix(gemma4): read head_dim / kv heads per attention type from per_layer_config
Xiaoze Fan
2026-08-13 03:55:59 +00:00
-
-
3af9d90ee5
feat: initial open-source release
Xiaoze Fan
2026-08-11 22:43:57 +00:00