Repository Issue Activity (beta)

ztxz16/fastllm

Current issue state, recent activity, and per-issue timelines from the indexed issue data.

Open Issues
147
New in 7 Days
9
Closed in 7 Days
1
Average Open Age
336 days
Stale 30+ Days
134
Stale 90+ Days
119
Last 2 Weeks
DateOpenedClosedCommentsEventsOpen Backlog
2026-09-1500000
2026-09-1431133
2026-09-130002145
2026-09-1210000
2026-09-1100000
2026-09-1030000
2026-09-0920000
2026-09-0800000
2026-09-0700000
2026-09-0610000
2026-09-0500000
2026-09-0400000
2026-09-0320000
2026-09-0200000
This Week

Opened: 9

Closed: 1

Comments: 1

Events: 5

Top Labels

No label distribution is available yet.

Issue Explorer
IssueAuthorStateLabelsCommentsReactionsUpdated

#738 [讨论] 能否兼顾 MTP 无损投机解码与高接受率

Opened 19 hours ago
jklnvlink
open
No labels
0019 hours ago

#736 DFlash2 backbone TP 在全同构 GPU 上被 equal-ratio 守卫跳过(force 后 +8% decode,4×2080Ti)

Opened 20 hours ago
DarthCY-K
open
No labels
0020 hours ago

#735 4GPU 运行 Qwen3.8 27B FP8 ,第一个GPU的利用率持续攀高

Opened 22 hours ago
w62651515
open
No labels
0022 hours ago

#728 [Bug]--开启mtp会导致 (qwen3_5 / Qwen3.8-27B)长上下文下思考无限循环

Opened 3 days ago
jklnvlink
closed - completed
No labels
3323 hours ago

#636 不支持Qwen3-32B-GPTQ-Int4模型的部署

Opened 9 months ago
zhaojc001
open
No labels
203 days ago

#725 [custom all-reduce] arrival-order rendezvous can pair different tensors across ranks (silent wrong results / out-of-range read)

Opened 5 days ago
liaoyangr8-ai
open
No labels
103 days ago

#722 [Bug+修复方案] 双卡 TP + MTP 长上下文 prefill 永久卡死:偶数 TP 缺少 NCCL 主机提交会合 + 分页参数阻塞上传

Opened 6 days ago
spideytznn
open
No labels
103 days ago

#726 `cudaErrorIllegalAddress` / Xid 13 crash with DFlash2 speculative decoding caused by missing stream sync in `FastllmCudaDFlashAttention`

Opened 5 days ago
To0nyZ
open
No labels
005 days ago

#724 [custom all-reduce] missing pointer-alignment guard faults on the device (cudaErrorMisalignedAddress / Xid 13)

Opened 5 days ago
liaoyangr8-ai
open
No labels
005 days ago

#723 [Bug] --device cuda --moe_device cpu 在线性注意力 MoE 模型(qwen3_5)上崩溃

Opened 6 days ago
ruyinmicc
open
No labels
006 days ago

#719 项目主页的微信群二维码过期了

Opened 9 days ago
NumericalAnalysis
open
No labels
208 days ago

#718 Token Pay合作邀请

Opened 12 days ago
Musinibaba
open
No labels
0012 days ago

#717 [Bug] Qwen3.8-Flash-Next (qwen4_exp) --mtp 1 aborts during warmup: cudaErrorInvalidValue (cudaMemcpy2D) -> cublas error on RTX 3090 (SM86)

Opened 12 days ago
jyf2100
open
No labels
0012 days ago

#690 pip安装rocm版本只更新到0.1.5.1

Opened 4 months ago
bbslxj
open
No labels
2027 days ago

#666 【Bug】DeepSeek-V4-Flash-FP8-Q4 使用 multicuda:0,1 输出乱码,单卡正常

Opened 4 months ago
haixinliu7864-alt
open
No labels
3030 days ago

#714 [Bug] V100 (SM70) 加载 AWQ 模型:权重 Load 100% 后 GPU 静默挂死,服务永不启动

Opened 30 days ago
Tianshaji0
open
No labels
0030 days ago

#700 希望尽快发布 Windows 版 0.1.7.0 wheel,以支持 Qwen3.5/Qwen3.6 系列模型

Opened 2 months ago
yutyflynew
open
No labels
3030 days ago

#712 Heap buffer overflows loading a crafted .safetensors model

Opened 1 month ago
professor-moody
open
No labels
201 month ago

#711 deepseek v4工具調用期間卡死

Opened 1 month ago
win10ogod
open
No labels
001 month ago

#709 8259cl + 2080ti 混推 deepseek v4 flash 0731 moe dtype int4g 报错

Opened 1 month ago
yichengming
open
No labels
101 month ago

#670 ROCm编译问题

Opened 4 months ago
h4rm00n
open
No labels
301 month ago

#704 Deepseek V4 Flash 混合推理OOM

Opened 2 months ago
WangDianhui
open
No labels
102 months ago

#703 一个新的量化想法

Opened 2 months ago
a793181018
open
No labels
002 months ago

#701 Out-of-bounds read in ModelLoader::ReadString when loading a crafted .flm model

Opened 2 months ago
professor-moody
open
No labels
002 months ago

#652 Unsupported model type qwen3_5_moe

Opened 6 months ago
lovedheart
open
No labels
203 months ago

Rows per page:

1–25 of 241