Repository navigation
LiteLLM silently drops GenerateContentConfig.seed #6513
Description
🔴 Required Information
Describe the Bug:
GenerateContentConfig(seed=42) is not forwarded by LiteLlm. The seed is still present in the request config, but missing from parameters passed to LiteLLM.
Steps to Reproduce:
- install
google-adk==2.5.0 - run code:
import asyncio
from google.adk.models.lite_llm import _get_completion_inputs
from google.adk.models.llm_request import LlmRequest
from google.genai import types
async def main():
req = LlmRequest(
model="openai/gpt-4o-mini",
config=types.GenerateContentConfig(temperature=0.0, seed=42),
)
result = await _get_completion_inputs(req, model=req.model)
print(result[3])
asyncio.run(main())Expected Behaviour:
{"temperature": 0.0, "seed": 42}Observed Behaviour:
{"temperature": 0.0}There is no error, seed is just silently dropped.
Environment Details:
- ADK: 2.5.0
- OS: macOS
- Python: 3.13
Model Information:
- LiteLLM: Yes
- Model:
openai/gpt-4o-mini, but repro does not call the model
🟡 Optional Information
Additional Context:
Looks like seed is missing from explicit parameter mapping in lite_llm.py. I also checked current main and it seems to have the same problem. I can submit a small fix and test if the approach is ok.
How often has this issue occurred?:
- Always (100%)
Activity
[-]LiteLlm silently drops GenerateContentConfig.seed[/-][+]LiteLLM silently drops GenerateContentConfig.seed[/+]fix(litellm): forward GenerateContentConfig.seed to LiteLLM
fix(litellm): forward GenerateContentConfig.seed to LiteLLM
surajksharma07 commentedon Jul 30, 2026
@svfat seed just isn't in the param allowlist inside _get_completion_inputs, so it gets silently dropped before reaching LiteLLM.
There's already a fix up in #6516 (adds seed to that list) that we've verified locally resolves it — could you pull that branch and confirm it works for your gpt-4o-mini setup as well?
Once you've tested and confirmed it on your side, we can move forward with getting it merged!
fix(litellm): forward GenerateContentConfig.seed to LiteLLM
fix(litellm): forward GenerateContentConfig.seed to LiteLLM
svfat commentedon Jul 30, 2026
@surajksharma07 Confirmed. I tested PR #6516 with openai/gpt-4o-mini, and seed=42 is now forwarded correctly in the LiteLLM generation parameters. The original reproduction now returns {"temperature": 0.0, "seed": 42}
surajksharma07 commentedon Aug 4, 2026
Appreciate you testing that on your actual gpt-4o-mini setup and reporting back — that's exactly the confirmation we needed!
#6516 is clean on CI, lint, and CLA, so it's just waiting on a review now.
I'll nudge GWeale to take a look and get it merged.
Clearing the clarification flag since you've confirmed it!
Team is looking more into it.
Thanks!
fix: forward GenerateContentConfig.seed to LiteLLM