PR CI Security Gate

Workflow runs for PR CI Security Gate

• Exporters: load optimum-intel, optimum-onnx and optimum-executorch exports

  skipped:
  PR CI Security Gate#16203:Pull request#50269synchronizebyIlyasMoutawwakil
  Nov 15, 20261s
  More actions for workflow run 16203

• Exporters: run generation exports through generate (ExportedModelForCausalLM, ExportedMultimodalModelForCausalLM, ExportedModelForSeq2SeqLM)

  skipped:
  PR CI Security Gate#16202:Pull request#50254synchronizebyIlyasMoutawwakil
  Nov 15, 20262s
  More actions for workflow run 16202

• Add generic heterogeneous model support (per-layer config and modeling)

  completed successfully:
  PR CI Security Gate#16201:Pull request#45332synchronizebyeladsegal
  Nov 14, 202639s
  More actions for workflow run 16201

• Fix Fp8Dequantize mis-dequantizing FP8 weights with partial blocks

  completed successfully:
  PR CI Security Gate#16200:Pull request#50767openedby1OF1Research
  Nov 14, 202633s
  More actions for workflow run 16200

• POC: Experts offload

  skipped:
  PR CI Security Gate#16199:Pull request#50717synchronizebytarekziade
  Nov 14, 20261s
  More actions for workflow run 16199

• Fix OpenVINO generate export tests for xLSTM and Bark

  skipped:
  PR CI Security Gate#16198:Pull request#50765synchronizebyydshieh
  Nov 14, 20261s
  More actions for workflow run 16198

• Warn when resuming from an older checkpoint

  completed successfully:
  PR CI Security Gate#16197:Pull request#50766openedbybezabyte
  Nov 14, 202640s
  More actions for workflow run 16197

• Exporters: load optimum-intel, optimum-onnx and optimum-executorch exports

  skipped:
  PR CI Security Gate#16196:Pull request#50269synchronizebyIlyasMoutawwakil
  Nov 14, 20261s
  More actions for workflow run 16196

• Exporters: run generation exports through generate (ExportedModelForCausalLM, ExportedMultimodalModelForCausalLM, ExportedModelForSeq2SeqLM)

  skipped:
  PR CI Security Gate#16195:Pull request#50254synchronizebyIlyasMoutawwakil
  Nov 14, 20261s
  More actions for workflow run 16195

• Exporters: run generation exports through generate (ExportedModelForCausalLM, ExportedMultimodalModelForCausalLM, ExportedModelForSeq2SeqLM)

  skipped:
  PR CI Security Gate#16194:Pull request#50254synchronizebyIlyasMoutawwakil
  Nov 14, 20268s
  More actions for workflow run 16194

• 🚨 Exporters: decompose generative models into one decode graph

  skipped:
  PR CI Security Gate#16193:Pull request#50252synchronizebyIlyasMoutawwakil
  Nov 14, 20269s
  More actions for workflow run 16193

• 🚨 Exporters: decompose generative models into one decode graph

  skipped:
  PR CI Security Gate#16192:Pull request#50252synchronizebyIlyasMoutawwakil
  Nov 14, 20267s
  More actions for workflow run 16192

• Exporters: run multimodal exports (ExportedMultimodalModel)

  skipped:
  PR CI Security Gate#16191:Pull request#50251synchronizebyIlyasMoutawwakil
  Nov 14, 20262s
  More actions for workflow run 16191

• 🚨 Exporters: export multimodal models per component (export_multimodal)

  skipped:
  PR CI Security Gate#16190:Pull request#50249synchronizebyIlyasMoutawwakil
  Nov 14, 202610s
  More actions for workflow run 16190

• Add Ovis2.5 model support

  completed successfully:
  PR CI Security Gate#16189:Pull request#47659synchronizebysushmanthreddy
  Nov 14, 202641s
  More actions for workflow run 16189

• Fix OpenVINO generate export tests for xLSTM and Bark

  skipped:
  PR CI Security Gate#16188:Pull request#50765openedbyydshieh
  Nov 14, 20262s
  More actions for workflow run 16188

• Reshard FSDP2 modules before creating the optimizer

  skipped:
  PR CI Security Gate#16187:Pull request#50535synchronizebyqgallouedec
  Nov 14, 20261s
  More actions for workflow run 16187

• [CB] Scale the per-token memory estimate by the TP size

  skipped:
  PR CI Security Gate#16186:Pull request#48566synchronizebyqgallouedec
  Nov 14, 20262s
  More actions for workflow run 16186

• Continuous batching for linear attention layers

  skipped:
  PR CI Security Gate#16185:Pull request#50350synchronizebyqgallouedec
  Nov 14, 20261s
  More actions for workflow run 16185

• Decode multi-head latent attention from the cache pages with FlashMLA

  skipped:
  PR CI Security Gate#16184:Pull request#50369synchronizebyqgallouedec
  Nov 14, 20261s
  More actions for workflow run 16184

• Use 32-row flex attention blocks for head_dim > 256

  skipped:
  PR CI Security Gate#16182:Pull request#49444synchronizebyqgallouedec
  Nov 14, 20262s
  More actions for workflow run 16182

• Size the paged cache from the rows and states the layers declare

  skipped:
  PR CI Security Gate#16183:Pull request#50356synchronizebyqgallouedec
  Nov 14, 20262s
  More actions for workflow run 16183

• Fuse the MoE top-k router into one kernel, served from the Hub

  skipped:
  PR CI Security Gate#16181:Pull request#48751synchronizebyqgallouedec
  Nov 14, 20262s
  More actions for workflow run 16181

• Cast the unquantized weights of a dequantized FP8 checkpoint to the model's dtype

  skipped:
  PR CI Security Gate#16180:Pull request#50595synchronizebyqgallouedec
  Nov 14, 20262s
  More actions for workflow run 16180

• Quantize on load from the checkpoint tensor, not its cast to the quantized dtype

  skipped:
  PR CI Security Gate#16179:Pull request#50301synchronizebyqgallouedec
  Nov 14, 20262s
  More actions for workflow run 16179