MCPcopy Create free account

hub / github.com/Tencent/hpc-ops / endpoints

Endpoints13 in github.com/Tencent/hpc-ops

Routefuse_allreduce_rmsnorm_high_throughput_fake
torch.library.register_fake("hpc::fuse_allreduce_rmsnorm_high_throughput")
hpc/allreduce.py:None
Routetest_act_mul_and_quant
pytest.mark.parametrize("use_output", [True, False])
tests/test_act.py:None
Routetest_attention_prefill_bf16
pytest.mark.parametrize("use_output", [True, False])
tests/test_attention_prefill_bf16.py:None
Routetest_attention_with_kvcache_prefill_bf16
pytest.mark.parametrize("use_output", [True])
tests/test_attention_with_kvcache_prefill_bf16.py:None
Routetest_attention_with_kvcache_prefill_fp8
pytest.mark.parametrize("use_output", [False])
tests/test_attention_with_kvcache_qpertoken_perhead_kvpertensor_prefill_fp8.py:None
Routetest_attention_with_kvcache_qkpertoken_perhead_vperhead_prefill_fp8
pytest.mark.parametrize("use_output", [False])
tests/test_attention_with_kvcache_qkpertoken_perhead_vperhead_prefill_fp8.py:None
Routetest_attn_fp8_sm90
pytest.mark.parametrize("use_output", [False])
tests/test_attention_decode_qpertoken_perhead_kvpertensor_fp8.py:None
Routetest_attn_fp8_sm90
pytest.mark.parametrize("use_output", [False])
tests/test_attention_decode_qkpertoken_perhead_vperhead_fp8.py:None
Routetest_fuse_moe_blockwise_fp8
pytest.mark.parametrize("has_shared_output", [False, True])
tests/test_fuse_moe_blockwise.py:None
Routetest_fuse_moe_cp_async
pytest.mark.parametrize("has_shared_output", [False, True])
tests/test_fuse_moe_cp_async.py:None
Routetest_fuse_moe_pertensor_fp8
pytest.mark.parametrize("has_shared_output", [False, True])
tests/test_fuse_moe_pertensor.py:None
Routetest_gemm_bf16xfp32_sm90
pytest.mark.skipif(torch.cuda.get_device_capability()[0] != 9, reason="skip on non sm90!")
tests/test_gemm_bf16xfp32.py:None
Routetest_rope_norm_store_kv_fp8
pytest.mark.skipif(bool(os.getenv("SANITIZER_CHECK")), reason="skip sanitizer")
tests/test_rope.py:None