| .. |
|
nki
|
Neuron MoE: skip unrouted experts inside the fp4 kernels at decode sizes (DEEPINFER_NEURON_GATED_EXPERTS, default on)
|
2026-10-26 06:36:48 +08:00 |
|
vendor
|
neuron: Engram gather via numpy (torch index_select waited 12-26 ms on its contended intra-op pool); DEEPINFER_NEURON_TORCH_THREADS; CPU bind mode 'all'
|
2026-10-28 02:10:43 +08:00 |
|
__init__.py
|
Add AWS Trainium (Neuron) platform, basic mode — part 1
|
2026-10-20 18:34:51 +08:00 |
|
fp8_guard.py
|
Neuron: tiny model on a NeuronCore matches the reference (issues #1, #2, #4)
|
2026-10-21 13:22:59 +08:00 |
|
fp8.py
|
Neuron: full DeepSeek-V4.1-Flash load + decode region graphs at EP=64 (issues #21, #25)
|
2026-10-23 01:39:22 +08:00 |
|
neuron_compile.py
|
neuron: float-only fp8 codec in the compile backend (no GpSimd int ops / remainder)
|
2026-10-26 12:41:42 +08:00 |
|
NEURON_PORT.md
|
Neuron: tiny model on a NeuronCore matches the reference (issues #1, #2, #4)
|
2026-10-21 13:22:59 +08:00 |
|
nki_attention.py
|
neuron: dense prefill attention over small KV tables and prefill table bounds (both opt-in / dormant by default)
|
2026-10-27 14:11:06 +08:00 |
|
nki_gemm.py
|
neuron: norms feeding only NKI kernels skip their dead fp8 copy (TTFT 1387 -> 1229 ms, step 457 -> 450 ms, tokens identical); capacity MoE wired into the layer region graph with its overflow flag riding the MoE all-reduce (still off: per-layer host sync makes prefill slower, 1823 ms)
|
2026-10-28 05:29:40 +08:00 |
|
nki_hc.py
|
neuron: mHC collapse through torch by default (DEEPINFER_NEURON_HC_TORCH=collapse); fix Neuron bind_cpu core index
|
2026-10-27 09:18:56 +08:00 |
|
nki_norm.py
|
Neuron: full DeepSeek-V4.1-Flash load + decode region graphs at EP=64 (issues #21, #25)
|
2026-10-23 01:39:22 +08:00 |
|
nki_topk.py
|
Neuron: nkilib rotational_topk for the sampler; single-host shared engram tables (#27)
|
2026-10-23 07:23:32 +08:00 |
|
op_census.py
|
neuron: sampling flags per request batch, NUMA-local CPU binding, sync/op-site tracing
|
2026-10-26 19:04:53 +08:00 |
|
regions.py
|
neuron: shared expert sliced over ranks, summed in the MoE all-reduce (DEEPINFER_NEURON_SHARED_SLICE=1)
|
2026-10-27 01:31:41 +08:00 |
|
sampling_mirror.py
|
neuron: compiled decode verification behind DEEPINFER_NEURON_COMPILE_VERIFY (default off)
|
2026-10-26 21:33:34 +08:00 |
|
static_moe.py
|
neuron: norms feeding only NKI kernels skip their dead fp8 copy (TTFT 1387 -> 1229 ms, step 457 -> 450 ms, tokens identical); capacity MoE wired into the layer region graph with its overflow flag riding the MoE all-reduce (still off: per-layer host sync makes prefill slower, 1823 ms)
|
2026-10-28 05:29:40 +08:00 |
|
transfer_to_neuron.py
|
Neuron: static decode graphs, static-shape MoE, trn2 e4m3 fp8 (issues #9, #12, #24)
|
2026-10-22 11:59:37 +08:00 |
