neuralmagic/speculators

AI Summary: This task requires extending the EagleSpeculator in the Speculators library to support multiple transformer layer architectures beyond LlamaDecoder. This involves updating the configuration to allow selection of different architectures (from a list of the top 5-10 most popular ones like Mistral, Qwen, etc.), modifying the Speculator to instantiate the correct layer type based on the configuration, and adding or updating relevant unit tests to ensure functionality and compatibility with existing features.

AI Summary: Expand the Speculators library's Eagle Speculator to support Hugging Face Transformers models beyond Llama 3, including Llama 4, Mistral, Qwen, and DeepSeek. This involves updating the EagleSpeculatorConfig and EagleSpeculator to handle these new verifier model types while retaining the Llama decoder layer for the speculator. Thorough testing is required to ensure compatibility with existing library functions.