Gemini Dispatch

Workflow runs for Gemini Dispatch

• fix(deps): skip nvidia-cuda-cccl on macOS (no wheels available)

  skipped:
  Gemini Dispatch#2139:Pull request#476submittedbyentrpn
  Nov 6, 202613s
  More actions for workflow run 2139

• feat(wan): token-sharded patch embedding (wan_patch_embed_mode="tokens")

  skipped:
  Gemini Dispatch#2138:Pull request#510submittedbyprishajain1
  Nov 6, 20262s
  More actions for workflow run 2138

• feat(wan): short-KV text cross-attention Pallas kernel (wan_cross_attn_kernel="pallas")

  skipped:
  Gemini Dispatch#2137:Pull request#509submittedbyprishajain1
  Nov 6, 20262s
  More actions for workflow run 2137

• perf(attention): shard-major Ulysses output all-to-all (wan_ulysses_out_a2a="chunked"/"shard_major")

  skipped:
  Gemini Dispatch#2136:Pull request#511submittedbyprishajain1
  Nov 5, 20262s
  More actions for workflow run 2136

• perf(wan): pad sequence to 128-lane multiple for self-attention (wan_seq_pad="lane")

  skipped:
  Gemini Dispatch#2135:Pull request#491submittedbyprishajain1
  Nov 5, 20262s
  More actions for workflow run 2135

• style: fix pyink formatting on main (2 files)

  skipped:
  Gemini Dispatch#2134:Pull request#503submittedbycsgoogle
  Oct 28, 20262s
  More actions for workflow run 2134

• chore: migrate gsutil usage to gcloud storage

  requires action with the application:
  Gemini Dispatch#2133:Pull request#506labeledbyentrpn
  gurusai-voleti:ai-gsutil-migration-9d3620735a024539bc218f65c07d86f1

  Oct 26, 2026Action required
  More actions for workflow run 2133

• fix(attention): dot-product fallback layout for 4-D inputs; tile_size_grid_search iters/bkv_compute guards

  skipped:
  Gemini Dispatch#2132:Pull request#507labeledbyPerseus14
  feat/attn-dot-fallback-fixes

  Oct 26, 20261s
  More actions for workflow run 2132

• chore: migrate gsutil usage to gcloud storage

  skipped:
  Gemini Dispatch#2131:Pull request#506submittedbyentrpn
  Oct 26, 20264s
  More actions for workflow run 2131

• feat(wan): fast serving with persistent AOT caching and tuned inference recipe

  skipped:
  Gemini Dispatch#2130:Pull request#479submittedbymbohlool
  Oct 26, 20261s
  More actions for workflow run 2130

• perf(wan): XLA-fused RMSNorm + RoPE producer for self-attention Q/K

  skipped:
  Gemini Dispatch#2129:Pull request#508submittedbymbohlool
  Oct 26, 202611s
  More actions for workflow run 2129

• perf(wan): pad sequence to 128-lane multiple for self-attention (wan_seq_pad="lane")

  skipped:
  Gemini Dispatch#2128:Pull request#491submittedbyprishajain1
  Oct 26, 20266s
  More actions for workflow run 2128

• perf(attention): shard-major Ulysses output all-to-all (wan_ulysses_out_a2a="chunked"/"shard_major")

  skipped:
  Gemini Dispatch#2127:Pull request#511submittedbyprishajain1
  Oct 26, 20261s
  More actions for workflow run 2127

• feat(wan): short-KV text cross-attention Pallas kernel (wan_cross_attn_kernel="pallas")

  skipped:
  Gemini Dispatch#2126:Pull request#509submittedbyprishajain1
  Oct 26, 20261s
  More actions for workflow run 2126

• feat(wan): fused RMSNorm+RoPE Pallas producer; Wan switches in config, not env

  skipped:
  Gemini Dispatch#2125:Pull request#488submittedbyprishajain1
  Oct 26, 20266s
  More actions for workflow run 2125

• feat(wan): fused RMSNorm+RoPE Pallas producer; Wan switches in config, not env

  skipped:
  Gemini Dispatch#2124:Pull request#488submittedbyprishajain1
  Oct 26, 20261s
  More actions for workflow run 2124

• fix(attention): dot-product fallback layout for 4-D inputs; tile_size_grid_search iters/bkv_compute guards

  skipped:
  Gemini Dispatch#2123:Pull request#507submittedbyprishajain1
  Oct 26, 20269s
  More actions for workflow run 2123

• feat(attention): fixed-m splash attention kernel with dynamic bounds and safety fallbacks

  skipped:
  Gemini Dispatch#2122:Pull request#477submittedbyprishajain1
  Oct 26, 20261s
  More actions for workflow run 2122

• perf(wan): batch-sharded cross-attn KV cache and batch-aware VAE mesh

  skipped:
  Gemini Dispatch#2121:Pull request#519createdbyPerseus14
  Oct 25, 20261s
  More actions for workflow run 2121

• perf(wan): batch-sharded cross-attn KV cache and batch-aware VAE mesh

  skipped:
  Gemini Dispatch#2120:Pull request#519submittedbyPerseus14
  Oct 25, 20268s
  More actions for workflow run 2120

• perf(wan): batch-sharded cross-attn KV cache and batch-aware VAE mesh

  skipped:
  Gemini Dispatch#2119:Pull request#519createdbygemini-code-assist[bot]Bot
  Oct 25, 20261s
  More actions for workflow run 2119

• perf(wan): batch-sharded cross-attn KV cache and batch-aware VAE mesh

  skipped:
  Gemini Dispatch#2118:Pull request#519submittedbygemini-code-assist[bot]Bot
  Oct 25, 20269s
  More actions for workflow run 2118

• Keep the full LoRA delta when merging into bfloat16 weights

  requires action with the application:
  Gemini Dispatch#2117:Pull request#513submittedbyboshjerns
  Oct 23, 2026Action required
  More actions for workflow run 2117

• Keep the full LoRA delta when merging into bfloat16 weights

  requires action with the application:
  Gemini Dispatch#2116:Pull request#513createdbyboshjerns
  Oct 23, 2026Action required
  More actions for workflow run 2116

• Keep the full LoRA delta when merging into bfloat16 weights

  requires action with the application:
  Gemini Dispatch#2115:Pull request#513submittedbyboshjerns
  Oct 23, 2026Action required
  More actions for workflow run 2115