AlwaysFHao

如题，HttpRequest中的getOkHttpClient方法中设置的TimeOut未与HttpRequest.TIME_OUT_DURATION字段同步，同时MultiFileRequest中的OkHttpClient对象是通过new OkHttpClient()实例化的，如需同步也应该使用build构造

发起多次请求后有概率抛出D/TrafficStats: tagSocket(89) with statsTag=0xffffffff, statsUid=-1 D/OkHttp: Callback failure for call to 我的请求url D/OkHttp: java.net.ProtocolException: unexpected end of stream D/OkHttp: at baseokhttp3.internal.http1.Http1Codec$FixedLengthSource.read(Http1Codec.java:409) D/OkHttp: at baseokio.Buffer.writeAll(Buffer.java:1016) D/OkHttp: at baseokio.RealBufferedSource.readString(RealBufferedSource.java:223) D/OkHttp: at baseokhttp3.ResponseBody.string(ResponseBody.java:174)...

On the small model, the actual GPU memory usage of Mamba2 is much higher than that of Mamba1.

17

The parameters of the Mamba2 model are d_state=32, d_conv=4, expand=2, and head_dim=32 (using "nn. Conv1d" with padding method, without the constraint of d_model/head_dim%8==0). Mamba1 maintains the same parameters except for...

Questions about Chunk_size using Triton optimization in SSD kernel

2

Thank you for your great work，I found that in Mamba2, chunk_size defaults to 256, while my sequence length is only 200 and still runs normally. In issue #439 , you...