Rohan Varma

icon indicating an email [email protected]

icon company Nvidia icon location Santa Clara, CA, USA icon location AI Developer Technology Engineer

Results 9 comments of Rohan Varma

[TRTLLM-5273]feat/Use full attention mask if Llama3 is used as encoder and fix EarlyStopDecoder unsqueeze bug

/bot run

[TRTLLM-5273]feat/Use full attention mask if Llama3 is used as encoder and fix EarlyStopDecoder unsqueeze bug

/bot run

[TRTLLM-5273]feat/Use full attention mask if Llama3 is used as encoder and fix EarlyStopDecoder unsqueeze bug

/bot run --disable-fail-fast

[TRTLLM-5273]feat/Use full attention mask if Llama3 is used as encoder and fix EarlyStopDecoder unsqueeze bug

/bot run --disable-fail-fast

[TRTLLM-5273]feat/Use full attention mask if Llama3 is used as encoder and fix EarlyStopDecoder unsqueeze bug

/bot run --disable-fail-fast

[TRTLLM-5273]feat/Use full attention mask if Llama3 is used as encoder and fix EarlyStopDecoder unsqueeze bug

/bot run --disable-fail-fast

[TRTLLM-5273]feat/Use full attention mask if Llama3 is used as encoder and fix EarlyStopDecoder unsqueeze bug

/bot run --disable-fail-fast

Workload API gaps for disaggregated/multi-component inference workloads

/sig scheduling

Workload API gaps for disaggregated/multi-component inference workloads

@ahg-g I'm a bit confused, I believe > the intent was for it to represent the minimum number of pods within an LWS "replica". is exactly what I am saying...