Gemini Dispatch

Workflow runs for Gemini Dispatch

• Fix the Wan VAE attention block, which never attends over spatial positions

  skipped:
  Gemini Dispatch#1969:Pull request#514submittedbygemini-code-assist[bot]Bot
  Oct 18, 20264s
  More actions for workflow run 1969

• Keep the full LoRA delta when merging into bfloat16 weights

  requires action with the application:
  Gemini Dispatch#1968:Pull request#513createdbygemini-code-assist[bot]Bot
  Oct 18, 2026Action required
  More actions for workflow run 1968

• Keep the full LoRA delta when merging into bfloat16 weights

  requires action with the application:
  Gemini Dispatch#1967:Pull request#513createdbygemini-code-assist[bot]Bot
  Oct 18, 2026Action required
  More actions for workflow run 1967

• Keep the full LoRA delta when merging into bfloat16 weights

  requires action with the application:
  Gemini Dispatch#1966:Pull request#513submittedbygemini-code-assist[bot]Bot
  Oct 18, 2026Action required
  More actions for workflow run 1966

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

  skipped:
  Gemini Dispatch#1965:Pull request#511createdbyPerseus14
  Oct 15, 20263s
  More actions for workflow run 1965

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

  skipped:
  Gemini Dispatch#1964:Pull request#511submittedbyPerseus14
  Oct 15, 20261s
  More actions for workflow run 1964

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

  skipped:
  Gemini Dispatch#1963:Pull request#511submittedbyPerseus14
  Oct 15, 20262s
  More actions for workflow run 1963

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

  skipped:
  Gemini Dispatch#1962:Pull request#511createdbyPerseus14
  Oct 15, 20262s
  More actions for workflow run 1962

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

  skipped:
  Gemini Dispatch#1961:Pull request#511createdbyPerseus14
  Oct 15, 20261s
  More actions for workflow run 1961

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

  skipped:
  Gemini Dispatch#1960:Pull request#511submittedbyPerseus14
  Oct 15, 20261s
  More actions for workflow run 1960

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

  skipped:
  Gemini Dispatch#1959:Pull request#511createdbyPerseus14
  Oct 15, 20261s
  More actions for workflow run 1959

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

  skipped:
  Gemini Dispatch#1958:Pull request#511submittedbyPerseus14
  Oct 15, 20261s
  More actions for workflow run 1958

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

  skipped:
  Gemini Dispatch#1957:Pull request#509createdbyPerseus14
  Oct 15, 20261s
  More actions for workflow run 1957

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

  skipped:
  Gemini Dispatch#1956:Pull request#509submittedbyPerseus14
  Oct 15, 20261s
  More actions for workflow run 1956

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

  skipped:
  Gemini Dispatch#1955:Pull request#508submittedbyPerseus14
  Oct 15, 20262s
  More actions for workflow run 1955

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

  skipped:
  Gemini Dispatch#1954:Pull request#508createdbyPerseus14
  Oct 15, 20269s
  More actions for workflow run 1954

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

  skipped:
  Gemini Dispatch#1953:Pull request#508createdbyPerseus14
  Oct 15, 20261s
  More actions for workflow run 1953

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

  skipped:
  Gemini Dispatch#1952:Pull request#508submittedbyPerseus14
  Oct 15, 20261s
  More actions for workflow run 1952

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

  skipped:
  Gemini Dispatch#1951:Pull request#507createdbyPerseus14
  Oct 15, 20261s
  More actions for workflow run 1951

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

  skipped:
  Gemini Dispatch#1950:Pull request#507submittedbyPerseus14
  Oct 15, 20261s
  More actions for workflow run 1950

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

  skipped:
  Gemini Dispatch#1949:Pull request#507createdbyPerseus14
  Oct 15, 20262s
  More actions for workflow run 1949

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

  skipped:
  Gemini Dispatch#1948:Pull request#507submittedbyPerseus14
  Oct 15, 20262s
  More actions for workflow run 1948

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

  skipped:
  Gemini Dispatch#1947:Pull request#509createdbygemini-code-assist[bot]Bot
  Oct 15, 20262s
  More actions for workflow run 1947

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

  skipped:
  Gemini Dispatch#1946:Pull request#509submittedbygemini-code-assist[bot]Bot
  Oct 15, 20262s
  More actions for workflow run 1946

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

  skipped:
  Gemini Dispatch#1945:Pull request#511createdbygemini-code-assist[bot]Bot
  Oct 15, 20261s
  More actions for workflow run 1945