-
Notifications
You must be signed in to change notification settings - Fork 195
All issues
Issue creation is restricted in this repository
Issues
is:issue state:open
is:issue state:open
Search results
[PJRT] Neuron 2.31 increases inter-executable latency from ~10-13 us to 170–190 us
bugSomething isn't workingSomething isn't workingStatus: Open.#1370 In aws-neuron/aws-neuron-sdk;[vllm-neuron 0.21.0.1.0.0] neuronx-cc compilation failed with 70 at tp_barrier — intermittent cold-compile crash at TP=64 (trn2.48xlarge)
bugSomething isn't workingSomething isn't workingStatus: Open.#1369 In aws-neuron/aws-neuron-sdk;[vllm-neuron 0.21.0.1.0.0] Compiled NEFF cache not reused across pods — VLLM_NEURON_REMOTE_CACHE has no effect (full recompile every cold start)
bugSomething isn't workingSomething isn't workingStatus: Open.#1368 In aws-neuron/aws-neuron-sdk;MXFP4 blockwise MoE kernel (bwmm_shard_on_block_mx) fails to compile for hidden_size>=6144: O(hidden_size) stack buffer exceeds capacity
bugSomething isn't workingSomething isn't workingStatus: Open.#1366 In aws-neuron/aws-neuron-sdk;neuronx-cc [F134] crash: getLatency(InstRangeSelect&) not implemented in Hwm.cpp:325 — inf2.24xlarge, Qwen2.5-32B, SDK 2.31.0, --enable-ccop-compute-overlap
bugSomething isn't workingSomething isn't workingStatus: Open.#1365 In aws-neuron/aws-neuron-sdk;neuronx-cc INTERNAL_ERROR NCC_IBIR297 (bir::samePartitionsAll in ModuleForkPass) compiling an MLA + 256-expert MoE decode graph on trn2
bugSomething isn't workingSomething isn't workingStatus: Open.#1362 In aws-neuron/aws-neuron-sdk;One-time NeuronCore runtime init (~6.7s in nrt_init/tdrv_init) dominates process startup — can it be shared/reduced?
bugSomething isn't workingSomething isn't workingStatus: Open.#1361 In aws-neuron/aws-neuron-sdk;[Feature Request] NxD Inference support for MiniMax-Text-01 (hybrid lightning/linear attention)
bugSomething isn't workingSomething isn't workingStatus: Open.#1355 In aws-neuron/aws-neuron-sdk;Using dynamic_range loops with tensor_copy can cause errors with no information
bugSomething isn't workingSomething isn't workingStatus: Open.#1354 In aws-neuron/aws-neuron-sdk;- Status: Open.#1346 In aws-neuron/aws-neuron-sdk;
Support StableHLO/PJRT graph-level memory placement and data movement betwen memories
documentationImprovements or additions to documentationImprovements or additions to documentationStatus: Open.#1340 In aws-neuron/aws-neuron-sdk;Neuron PJRT StableHLO compile rewrites sort+slice sampling path to AwsNeuronTopK with full vocab size as k
bugSomething isn't workingSomething isn't workingStatus: Open.#1339 In aws-neuron/aws-neuron-sdk;