[FEATURE] Truncation support for non vllm_server providers
- 主要語言
- Python
- 星號
- 70
- 分支
- 21
- 平均合併
- 2 天 11 小時
- 30 天內合併 PR
- 41
描述
## Feature Description
Add in support for the sampling_params variables truncation_side and truncate_prompt_tokens to the vllm, huggingface, litellm, and olmo_core providers to match the vllm_server implementation (https://github.com/allenai/olmo-eval/pull/208).
## Use Case
Currently, any setting of truncation_side or truncate_prompt_tokens for a non-vllm_server provider generates a warning. This limits providers that can be used for the safety judge, and any other length-limited model that is run
貢獻指南
這個儲存庫沒有索引到貢獻指南
研究方向
Compare the vllm_server implementation referenced in PR #208 with the vllm, huggingface, litellm, and olmo_core provider paths. Start by locating how sampling_params are handled and where warnings are generated for truncation_side and truncate_prompt_tokens. Done means these settings work for all four providers without the current warning.
由索引模型根據 Issue 內容生成。
評估
- 技術堆疊
- huggingface, python
- 領域
- machine-learning
- Issue 類型
- 功能
- 難度
- 3/5
- 預估耗時
- 1-2 天
- 活躍度
- 冷清
- 描述清晰度
- 基本清楚
- 新手友好度
- 55/100