Chen Qian
Chen Qian
It's very hard to 100% reproduce a distributed training workloads with the infra we have access to, but we can try spinning up multiple processes (macbook pro should be sufficient)...
Unfortunately it's annoyingly complex, I don't think we can get this infra set up within 1-2 weeks. IMO it will have the components below: - Testing trigger: we can use...
this is interesting, will be a bit weird for batched inputs, but I can see it's useful for single string input.
An easy way to debug is turning on MLflow tracing: https://dspy.ai/tutorials/observability/. > The error I'm getting from dspy shows that dspy is constructing content as just a string. In our...
@lambda-science Thanks for the detailed issue report! Looks like Bedrock can have None as the value, while most other providers sets a default value.
@Haoping-Xiao Thanks for reporting the issue! Although we don't capture the function name, we do provide a hook for all the methods we want to trace, e.g., `on_adapter_format_start`: https://github.com/stanfordnlp/dspy/blob/9b258e390307d95e83d029be35d57d7092b2aa85/dspy/utils/callback.py#L360. Are...
Our current trace happens at the method level, and we use decorator `@with_callbacks` for methods, which already captures the method name. How do you implement your tracing? If you write...
I actually don't think this is blocking/unblocking what you are trying to do. From your code, you are using the fact that `_get_on_start_handler` runs `on_tool_start`/`on_tool_end` hooks for any methods inside...
@fchollet Yea, this should be a simple fix. We can probably add a util `maybe_disable_xla()`, and further disabling situation could be covered by the util. I will benchmark RNN with...
@ivanmilevtues Thanks for working on this! I like diagrams or knowledge graph, but I don't think the rendered diagram is correct though. Since DSPy is using MIT license, and you...