Kuiz berkaitan panduan · Sederhana Tahap
Kuiz LLMOps lwn MLOps
Bandingkan kebimbangan operasi khusus LLM dengan amalan MLOps yang dikongsi merentasi gesaan, perolehan semula, penilaian, kos dan pemantauan.
soalan 1 daripada 8
Perubahan apa yang boleh mengubah tingkah laku aplikasi LLM tanpa menanaman semula berat model asas?
Pratonton setiap soalan8 soalan
Jawapan dan penjelasan muncul semasa anda mengambil kuiz.
- Perubahan apa yang boleh mengubah tingkah laku aplikasi LLM tanpa menanaman semula berat model asas?
- A prompt template or retrieval index update
- A dashboard color change only
- A cloud region name in documentation
- A model card font size
- Mengapa versi korpus atau indeks pengekstrakan secara berasingan daripada penjana?
- It makes the index part of model weights
- It lets teams isolate whether changed evidence retrieval caused an output regression
- It eliminates document access controls
- It guarantees factual answers
- Metric apa yang khusus untuk penjejakan kos API LLM biasa?
- GPU occupancy only
- Gini impurity
- Input and output token usage
- Number of class labels
- Mengapa menjalankan kes penilaian selepas menukar prompt?
- Evaluation is only for GPU kernels
- Prompts cannot affect model responses
- A prompt change retrains the model automatically
- Prompt wording can alter outputs, tool use and refusal behavior
- Apa yang boleh terlepas daripada penghakiman berasaskan model?
- Prompt version changes
- Factual errors or evaluator bias shared with the candidate
- Token counts
- Retrieval index updates
- Which shared MLOps practice remains useful for LLM applications?
- Using only training accuracy
- Ignoring data governance
- Staged release, monitoring and rollback
- Always retraining full weights per request
- Why can a hosted model alias limit reproducibility?
- Hosted models never return metadata
- Aliases contain the full model weights
- The prompt cannot be stored
- Provider behavior may change behind an identifier
- Which security issue can arise from retrieved external documents?
- Untrusted content can include prompt-injection instructions
- Documents automatically retrain the model
- Retrieval guarantees valid access permissions
- Embeddings remove sensitive data