🐛 Bug Report: LiteLLM instrumentation does not record seed, stop, frequency_penalty, presence_penalty

未关闭 适合新手
#4,598 2 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

@shantam77 已经在做这个了。

开始于 2026年10月11日。

  • #4607 来自 @shantam77 —— 未关闭

评估

难度
2/5
预计耗时
1-3 小时
新手友好度
85/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
活跃
技术栈
python
领域
ai, observability

调研方向

从 opentelemetry/instrumentation/litellm/__init__.py 中的 _set_request_attributes 开始,将其请求属性映射与 LiteLLM callback 进行比较。使用提供的复现来确认发出了哪些 GenAI 属性,然后在保留现有属性的同时,为 seed、stop、frequency_penalty 和 presence_penalty 添加测试覆盖。完成的标准是,当提供这四个参数时,它们都使用指定的 gen_ai.request.* 名称出现。

由索引模型根据 Issue 内容生成。

描述

Which component is this bug for?

AlephAlpha Instrumentation

📜 Description

Package: opentelemetry-instrumentation-litellm 0.62.4 (not in the component list, so I picked the closest option).

_set_request_attributes in the LiteLLM instrumentation (opentelemetry/instrumentation/litellm/__init__.py, lines 166-178 at 0.62.4) records only gen_ai.request.model, temperature, top_p and max_tokens. Request parameters that the GenAI semantic conventions define and that LiteLLM forwards to the provider are dropped: gen_ai.request.seed, gen_ai.request.stop_sequences, gen_ai.request.frequency_penalty, gen_ai.request.presence_penalty. A trace therefore cannot show why two otherwise identical calls behaved differently.

For comparison, LiteLLM's own OpenTelemetry callback records all of them for the same call. Related: #4538 covers seed / stop_sequences in the OpenAI package; this is the separate LiteLLM package, which also misses the penalties.

👟 Reproduction steps

pip install litellm opentelemetry-instrumentation-litellm opentelemetry-sdk httpx

import os
os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
import httpx, litellm, openai
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import SimpleSpanProcessor
from opentelemetry.sdk.trace.export.in_memory_span_exporter import InMemorySpanExporter
from opentelemetry.instrumentation.litellm import LiteLLMInstrumentor

exporter = InMemorySpanExporter()
provider = TracerProvider()
provider.add_span_processor(SimpleSpanProcessor(exporter))
LiteLLMInstrumentor().instrument(tracer_provider=provider)

client = openai.OpenAI(api_key="x", max_retries=0, http_client=httpx.Client(transport=httpx.MockTransport(
    lambda r: httpx.Response(200, json={"id": "c1", "object": "chat.completion", "created": 0, "model": "gpt-4o-mini",
        "choices": [{"index": 0, "message": {"role": "assistant", "content": "Hello"}, "finish_reason": "stop"}],
        "usage": {"prompt_tokens": 11, "completion_tokens": 5, "total_tokens": 16}}))))
litellm.completion(model="openai/gpt-4o-mini", messages=[{"role": "user", "content": "hi"}], client=client, api_key="x",
                   seed=7, stop=["END"], temperature=0.3, top_p=0.9, max_tokens=50,
                   frequency_penalty=0.1, presence_penalty=0.2)
print({k: v for k, v in exporter.get_finished_spans()[0].attributes.items() if k.startswith("gen_ai.request.")})
👍 Expected behavior

gen_ai.request.seed, gen_ai.request.stop_sequences, gen_ai.request.frequency_penalty and gen_ai.request.presence_penalty are set when the call passes them.

👎 Actual Behavior with Screenshots
gen_ai.request.model = openai/gpt-4o-mini
gen_ai.request.temperature = 0.3
gen_ai.request.top_p = 0.9
gen_ai.request.max_tokens = 50

(seed, stop, frequency_penalty and presence_penalty were sent but are not recorded)

🤖 Python Version

3.12

📃 Provide any additional context for the Bug.

opentelemetry-instrumentation-litellm 0.62.4 (the latest release), litellm 1.104.2.

👀 Have you spent some time to check if this bug has been raised before?
  • I checked and didn't find similar issue
Are you willing to submit PR?

None

主要语言
Python
星标
7.5k
派生
1.2k
平均合并
2 天 6 小时
30 天内合并 PR
2

环境准备

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

traceloop/openllmetry 的其他 Issue

查看 traceloop/openllmetry 的全部 Issue

相似的 Issue

更多 Python Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。