torshie

Add Android TV support on the server side

10

With Android TV support on the server side, we can enjoy all the goodies of NymphCast without the burden (cable hazard, space, money ) of an extra device.

Export the network architecture and re-train with a different dataset.

1

I'm testing AutoKeras on a recommender system. Usually recommenders are developed with a fixed training set. When a good result is found, the architecture is finalized. When deployed the model...

[BUG] FP32 gradient accumulation result in crash.

3

**Describe the bug** * Enable BF16 training * Set gradient accumulation types to FP32 * Enable ZeRO-1, and CPU offload * Enable overlap_comm * Tune train batch size so that...

Identify code snippet in "input" fields

1

I want to translate the training data into another language with Google translate, but code snippets should not be translated, so I have to replace code snippets with placeholders before...

Thresholds for all quality signals

2

After all quality signals are generated, what are the thresholds used to classify a document as good/bad for each quality signal ?

Does mimalloc support shared memory blocks ? If not, what kind of effort is required to make it do ?

Test case `test_hstu_attention_tma.HSTUAttentionTmaTest.test_attn_triton_tma` seems always fail on my environment: fbgemm_gpu 1.3.0+cu128 torch 2.8.0+cu128 torchrec 1.3.0+cu128 triton 3.4.0 Python 3.12.3 OS: Ubuntu 24.04 ``` python3: /root/.triton/llvm/llvm-8957e64a-almalinux-x64/include/llvm/ADT/SmallVector.h:296: const_reference llvm::SmallVectorTemplateCommon::operator[](size_type) const [T =...