Llm¶

LLM¶

CrossAttentionGPT2¶

        classDiagram
    class CrossAttentionGPT2 {
        +embed_tokens() Tensor
        +forward() Tensor
    }
    class Module
    <<abstract>> Module
    Module <|-- CrossAttentionGPT2 : herits

CrossAttentionGPT2

A GPT2 with cross attention to allow vision/weather data injection as key/values into some of the transformer block.

GPT2¶

        classDiagram
    class GPT2 {
        +download_weights_from_tf_ckpt() None
        +embed_tokens() Tensor
        +forward() Tensor
        +forward_vectors() Tensor
        +load_weights_from_dict()
        +reset_kv_cache() None
    }
    class Module
    <<abstract>> Module
    Module <|-- GPT2 : herits

GPT2

GPT implementation - Based on Sebastian Raschka's book and github repo : https://github.com/rasbt/LLMs-from-scratch/.

Llama2¶

        classDiagram
    class Llama2 {
        +embed_tokens() Tensor
        +forward() Tensor
        +forward_vectors() Tensor
    }
    class Module
    <<abstract>> Module
    Module <|-- Llama2 : herits

Llama2

Llama2 implementation - Based on Sebastian Raschka's book and github repo : https://github.com/rasbt/LLMs-from-scratch/.

Llama3¶

        classDiagram
    class Llama3 {
        +embed_tokens() Tensor
        +forward() Tensor
        +forward_vectors() Tensor
        +reset_kv_cache() None
    }
    class Module
    <<abstract>> Module
    Module <|-- Llama3 : herits

Llama3

Qwen3_5¶

        classDiagram
    class Qwen3_5 {
        +compute_memory_size() float
        +create_mask() Tensor
        +download_weights_from_hf() None
        +forward() Tensor
        +generate_output_stream() Iterator
        +generate_text() str
        +load_weights_from_dict() None
        +reset_kv_cache() None
    }
    class Module
    <<abstract>> Module
    Module <|-- Qwen3_5 : herits

Qwen3_5