[WIP] feat: support bfclv4#178
Open
123liuziming wants to merge 3 commits intomainfrom
Open
Conversation
…ard v4 Introduce loongsuite-instrumentation-bfclv4 covering BFCL v4 (bfcl_eval) per the design in llm-dev/bfclv4/execute.md: * ENTRY span around bfcl_eval._llm_response_generation.generate_results, with a narrow swap of that module's ThreadPoolExecutor name to a contextvars-propagating subclass so worker threads inherit the ENTRY trace context. * AGENT span around BaseHandler.inference (kind=AGENT, op=invoke_agent), picking up token usage from the metadata BFCL writes back. * STEP spans created reflectively for every concrete handler discovered via bfcl_eval.constants.model_config.MODEL_CONFIG_MAPPING; each STEP re-invokes the handler's _parse_query_response_* to harvest token counts and latency. * Per-call TOOL spans emitted from bfcl_eval.eval_checker.multi_turn_eval.multi_turn_utils.execute_multi_turn_func_call (one span per func_call entry in the batch). * Provider override mapping that routes OSSMODEL handlers to vllm/sglang based on args.backend, plus contextvars-based bfcl.turn_idx / gen_ai.react.round tracking. LLM spans are intentionally not created by this plugin; they continue to be produced by the downstream vendor SDK probes (OpenAI / Anthropic / DashScope / etc.).
|
musi seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account. You have signed the CLA already but the status is still pending? Let us recheck it. |
… responses When ``_query_FC`` / ``_query_prompting`` returns a streaming wrapper (e.g. ``openai-v2`` ``ChatStreamWrapper``), the LLM span and its OTel context attach are kept alive until the stream is consumed by BFCL's ``_parse_query_response_*`` after the STEP context manager has already exited. Non-LIFO context detach then leaves the prior LLM span as the "current" span, which causes subsequent STEP and TOOL spans to be parented under the previous STEP rather than under AGENT. Force-consume the streaming response inside the STEP context and replace it with a plain iterator over the cached chunks so that ``stop_llm`` (which detaches LLM context) runs in LIFO order before STEP detaches.
17 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
Please include a summary of the change and which issue is fixed. Please also include relevant motivation and context. List any dependencies that are required for this change.
Fixes # (issue)
Type of change
Please delete options that are not relevant.
How Has This Been Tested?
Please describe the tests that you ran to verify your changes. Provide instructions so we can reproduce. Please also list any relevant details for your test configuration
Does This PR Require a Core Repo Change?
Checklist:
See contributing.md for styleguide, changelog guidelines, and more.