Repository navigation
Actions: google/tunix
Actions
Showing runs from all workflows
2,500+ workflow runs
2,500+ workflow runs
StandardRLProgram step shows up only as perf/global_step_time. Telling whether rollouts, training, or weight sync were slow, and which rollouts held up the batch, meant scanning trajectory logs. Log that directly:
GitHub Actions Scan
#4188:
Pull request #2826
opened
by
copybara-service
Bot
StandardRLProgram step shows up only as perf/global_step_time. Telling whether rollouts, training, or weight sync were slow, and which rollouts held up the batch, meant scanning trajectory logs. Log that directly:
Tunix Package Tests
#11280:
Pull request #2826
opened
by
copybara-service
Bot
StandardRLProgram step shows up only as perf/global_step_time. Telling whether rollouts, training, or weight sync were slow, and which rollouts held up the batch, meant scanning trajectory logs. Log that directly:
Check the Documentation Build
#7722:
Pull request #2826
opened
by
copybara-service
Bot
StandardRLProgram step shows up only as perf/global_step_time. Telling whether rollouts, training, or weight sync were slow, and which rollouts held up the batch, meant scanning trajectory logs. Log that directly:
Auto-assign issues and PRs
#1727:
Pull request #2826
opened
by
copybara-service
Bot
TrajectoryCollectEngine enforced its timeout budget only on env.step. A slow model call, e.g. a sampler request parked across a long weight sync, held the episode and its rollout slot past the budget (indefinitely if it never returned), and a call that returned after the budget ran out made the next env.step start with a negative deadline: the episode was mislabeled ENV_TIMEOUT, and the abandoned env.step could keep running while the final reward was computed.
Tunix Package Tests
#11279:
Pull request #2821
opened
by
copybara-service
Bot
TrajectoryCollectEngine enforced its timeout budget only on env.step. A slow model call, e.g. a sampler request parked across a long weight sync, held the episode and its rollout slot past the budget (indefinitely if it never returned), and a call that returned after the budget ran out made the next env.step start with a negative deadline: the episode was mislabeled ENV_TIMEOUT, and the abandoned env.step could keep running while the final reward was computed.
Check the Documentation Build
#7716:
Pull request #2821
opened
by
copybara-service
Bot
TrajectoryCollectEngine enforced its timeout budget only on env.step. A slow model call, e.g. a sampler request parked across a long weight sync, held the episode and its rollout slot past the budget (indefinitely if it never returned), and a call that returned after the budget ran out made the next env.step start with a negative deadline: the episode was mislabeled ENV_TIMEOUT, and the abandoned env.step could keep running while the final reward was computed.
Auto-assign issues and PRs
#1722:
Pull request #2821
opened
by
copybara-service
Bot
TrajectoryCollectEngine enforced its timeout budget only on env.step. A slow model call, e.g. a sampler request parked across a long weight sync, held the episode and its rollout slot past the budget (indefinitely if it never returned), and a call that returned after the budget ran out made the next env.step start with a negative deadline: the episode was mislabeled ENV_TIMEOUT, and the abandoned env.step could keep running while the final reward was computed.
GitHub Actions Scan
#4187:
Pull request #2821
opened
by
copybara-service
Bot
BaseChatTemplateParser strips every TokenConfig string from message content to block control-token injection. For Qwen that set included tool_start_token ("\n<tool_call>\n") and tool_end_token ("\n</tool_call>"), which are ordinary text in tool-use system prompts and in assistant tool-call turns. Stripping them removed the <tool_call> wrapper from the function-calling format example that the Qwen3.5 chat template writes into the system prompt, so the model emi...
Tunix Package Tests
#11277:
Pull request #2820
synchronize
by
copybara-service
Bot
BaseChatTemplateParser strips every TokenConfig string from message content to block control-token injection. For Qwen that set included tool_start_token ("\n<tool_call>\n") and tool_end_token ("\n</tool_call>"), which are ordinary text in tool-use system prompts and in assistant tool-call turns. Stripping them removed the <tool_call> wrapper from the function-calling format example that the Qwen3.5 chat template writes into the system prompt, so the model emitted bare <function=...> blocks and often failed to end its turn. In agentic RL on DeepSWE with Qwen3.5-35B-A3B, about a fifth of rollouts degenerated into repeated </function> until they hit the response-length cap, which more than doubled step time.
Check the Documentation Build
#7713:
Pull request #2820
synchronize
by
copybara-service
Bot
BaseChatTemplateParser strips every TokenConfig string from message content to block control-token injection. For Qwen that set included tool_start_token ("\n<tool_call>\n") and tool_end_token ("\n</tool_call>"), which are ordinary text in tool-use system prompts and in assistant tool-call turns. Stripping them removed the <tool_call> wrapper from the function-calling format example that the Qwen3.5 chat template writes into the system prompt, so the model emitted bare <function=...> blocks and often failed to end its turn. In agentic RL on DeepSWE with Qwen3.5-35B-A3B, about a fifth of rollouts degenerated into repeated </function> until they hit the response-length cap, which more than doubled step time.
GitHub Actions Scan
#4186:
Pull request #2820
synchronize
by
copybara-service
Bot
BaseChatTemplateParser strips every TokenConfig string from message content to block control-token injection. For Qwen that set included tool_start_token ("\n<tool_call>\n") and tool_end_token ("\n</tool_call>"), which are ordinary text in tool-use system prompts and in assistant tool-call turns. Stripping them removed the <tool_call> wrapper from the function-calling format example that the Qwen3.5 chat template writes into the system prompt, so the model emi...
Tunix Package Tests
#11276:
Pull request #2820
synchronize
by
copybara-service
Bot
BaseChatTemplateParser strips every TokenConfig string from message content to block control-token injection. For Qwen that set included tool_start_token ("\n<tool_call>\n") and tool_end_token ("\n</tool_call>"), which are ordinary text in tool-use system prompts and in assistant tool-call turns. Stripping them removed the <tool_call> wrapper from the function-calling format example that the Qwen3.5 chat template writes into the system prompt, so the model emitted bare <function=...> blocks and often failed to end its turn. In agentic RL on DeepSWE with Qwen3.5-35B-A3B, about a fifth of rollouts degenerated into repeated </function> until they hit the response-length cap, which more than doubled step time.
Check the Documentation Build
#7712:
Pull request #2820
synchronize
by
copybara-service
Bot