Search results

[TPU] FP8 blockwise quantization: block_size vs MXU utilization tradeoff — how to reconcile in Pallas kernels?

Status: Open.

#839 In openxla/tokamax;

· pengchengneo opened on Apr 8, 2026

Persist autotuning cache

Status: Open.

#792 In openxla/tokamax;

· emergenz opened on Mar 24, 2026

Mosaic GPU implementation of mamba2 ssd layer

Status: Open.

#750 In openxla/tokamax;

· AlexG1105 opened on Mar 5, 2026

B200 Mosaic ragged_dot VJP unsupported

Status: Open.

#628 In openxla/tokamax;

· changlan opened on Feb 11, 2026

result mismatch for mosaic gpu attention on sm100

Status: Open.

#574 In openxla/tokamax;

· qelk123 opened on Jan 31, 2026

[Feature Request] Support for Custom DeepSeek-V3 Model Sizes (30B & 155B) in Tokamax GMM

Status: Open.

#385 In openxla/tokamax;

· bzantium opened on Nov 24, 2025

splash attention not compatible with shard_map(check_vma=True)

Status: Open.

#382 In openxla/tokamax;

· hb-scaler opened on Nov 21, 2025

Error tuning ragged_dot backward

Status: Open.

#375 In openxla/tokamax;

· hb-scaler opened on Nov 20, 2025

Trying to access flag --tokamax_xxx before flags were parsed

Status: Open.

#374 In openxla/tokamax;

· hb-scaler opened on Nov 20, 2025

When would the performance benchmark is available?

Status: Open.

#298 In openxla/tokamax;

· MoFHeka opened on Nov 1, 2025

Fixed typeguard version prevents installation

Status: Open.

#240 In openxla/tokamax;

· adonath opened on Oct 22, 2025