A research firm wants a Foundry-hosted assistant to answer questions about individual 300-page technical reports. Each report is far larger than the deployed model's context window, and sending the full document in one prompt causes errors. What approach lets the assistant answer questions grounded in a specific report without exceeding the model's input limit?
Select an answer to reveal the explanation.
Short Explanation
A 300-page report will not fit into one prompt no matter how the request is phrased, so the fix is to stop trying to hand over the whole thing at once. Break the report into smaller chunks, find the handful of passages that actually answer the question being asked, and send only those along with the question. That keeps every request comfortably inside the model's input limit while still grounding the answer in the right document. Turning up randomness has nothing to do with how much text fits in a prompt. Baking every report into the model ahead of time through heavy retraining is slow, has to be repeated every time a new report arrives, and still does not guarantee it can recite exact passages on demand. And trimming how much the model is allowed to write back solves a totally different problem than an input that is already too large to send.
Full Explanation
The correct answer is A. Splitting each report into smaller sections and retrieving only the passages relevant to a given question keeps the amount of text sent to the model within its context window while still grounding the answer in the correct report, which is the retrieval-augmented pattern built for exactly this constraint. Option B is incorrect because temperature controls how varied or deterministic the wording of a response is; it has no effect on how much input text the model can accept. Option C is incorrect because fine-tuning every report into the model ahead of time is a heavy, slow process aimed at shaping general behavior or style, not a reliable way to guarantee the model recalls and cites the exact contents of a specific 300-page document on demand, and it would need to be redone whenever reports change. Option D is incorrect because reducing the requested output length addresses how much the model writes back, not the separate problem of the input document exceeding the model's context window in the first place.