Load and Tune an Open-Source Model with Python | Dataford Interview Questions

Write a Python script to load an open-source model and use it for text generation while tweaking hyperparameters.

Asked in the Technical Round 1 stage. For deterministic grading, represent the loaded model as a JSON checkpoint whose transitions map each token to candidate next-token logits. Return only generated text, excluding the prompt.

Function Contract

Implement generate_text(model_json, prompt, max_new_tokens, temperature, top_p). The JSON string contains a transitions object. Select tokens using temperature-scaled logits and a top_p nucleus, choosing the highest-probability remaining token at each step. Use <BOS> for an empty prompt and stop when no transition exists.

Constraints

model_json is a valid JSON string containing a transitions object
Each transition maps token strings to numeric logits
0 <= max_new_tokens <= 100
temperature > 0
0 < top_p <= 1
Tokens and prompts are separated by single spaces

Function Signature

def generate_text(model_json, prompt, max_new_tokens, temperature, top_p):

Your question is Load and Tune an Open-Source Model with Python. Start with the requirements in the Question tab.

Run and submit as often as you like. When you're ready, talk me through your approach or go straight to the code.

Problem