GPT-2 Medium
Open-weight text completion model for generation and fine-tuning
OpenAI's GPT-2 Medium open-weight model for text generation and completion. Can be used with the base pretrained weights or with a fine-tuned checkpoint from a training run.
When to use:
- Text continuation and autocompletion
- Running locally or on-premise without API calls
- Using a fine-tuned checkpoint for a specific domain
Input: Text prompt + optional fine-tuned checkpoint Output: Generated text continuation and generation metadata
Inference Settings
Temperature (default: 0.7) Sampling temperature for generation.
- 0.0: Greedy decoding — most likely token at every step
- 0.7: Creative, varied output
- 1.0+: High diversity, may be incoherent
Max Length (default: 100) Maximum number of tokens to generate (prompt tokens + generated tokens).
- Set higher for longer completions
- Keep low for speed and cost
Top P (default: 0.9, range: 0.0–1.0) Nucleus sampling threshold. Only tokens in the top P probability mass are considered.
- 0.9: Good default — avoids very unlikely tokens
- 1.0: No filtering, full distribution