Forskjeller

Her vises forskjeller mellom den valgte versjonen og den nåværende versjonen av dokumentet.

Lenk til denne sammenligningen

Begge sider forrige revisjonForrige revisjon
Neste revisjon		Forrige revisjon
start [2025/12/02 19:27] – admin99start [2025/12/02 22:53] (nåværende versjon) – admin99
Linje 1: Linje 1:
-====== **LET'S DEFINE THE PIECES** ======+====== LET'S DEFINE THE PIECES ======
  
 Everything you need to know falls into one of three categories: Everything you need to know falls into one of three categories:
  
 ---- ----
-**===== 1. MODELS (LLMs) ===== +===== 1. MODELS (LLMs) =====
-** +
 ---- ----
  
Linje 13: Linje 11:
  
  
-These are the brains. \\ +  * These are the brains. \\ 
-They are NOT software. \\ +  * They are NOT software. \\ 
-They are NOT programs. \\ +  * They are NOT programs. \\ 
-They are NOT plugins. \\ +  * They are NOT plugins. \\ 
-They are neural networks stored in a single file, usually: \\+  * They are neural networks stored in a single file, usually: \\
 **model.gguf \\ **model.gguf \\
 model.safetensors \\ model.safetensors \\
Linje 43: Linje 41:
  \\  \\
 **A model does NOT:** \\ **A model does NOT:** \\
-Have a UI \\ +  * Have a UI \\ 
-Open PDFs \\ +  * Open PDFs \\ 
-Connect to NASA \\ +  * Connect to NASA \\ 
-Run indexing \\ +  * Run indexing \\ 
-Provide a chat window \\+  * Provide a chat window \\
  \\  \\
 It only takes text in → text out. \\ It only takes text in → text out. \\
Linje 56: Linje 54:
  
 ---- ----
-**===== 2. MODEL RUNTIMES (ENGINES) =====** \\+===== 2. MODEL RUNTIMES (ENGINES) ===== 
 +----
 These are the programs that LOAD and RUN the model's brain. \\ These are the programs that LOAD and RUN the model's brain. \\
 Think of a “runtime” as the machine that runs a model file. \\ Think of a “runtime” as the machine that runs a model file. \\
  \\  \\
-**Runtimes include:** \\ +**===== Runtimes include ===== 
-Ollama+:** \\ 
 +**✔ Ollama** 
 +  * Terminal-based 
 +  * Local API 
 +  * Can fine-tune 
 +  * Good for automation 
 +  * Good for pipelines 
 +  * Very flexible 
 +  * Acts like a backend server
  
-Terminal-based 
  
-Local API 
  
-Can fine-tune+**✔ LM Studio** 
 +  * GUI desktop app 
 +  * Easy model downloading 
 +  * Drag-and-drop PDFs 
 +  * File chat 
 +  * Rudimentary RAG 
 +  * Easy to test many models 
 +  * Great for tinkering
  
-Good for automation+**✔ GPT4All** 
 +  * GUI 
 +  * Also a runtime 
 +  * Similar to LM Studio 
 +  * Not as modern
  
-Good for pipelines+**✔ koboldcpp** 
 +  * Runtime specialized for story-writing/roleplay 
 +  * GUI 
 +  * Some fine-tuning tools
  
-Very flexible+✔ Faraday / Unsloth / Axolotl (training tools, not chatbots)
  
-Acts like a backend server+These are **training engines**, not chat apps.
  
 +You don’t talk to them;
 +you use them to train a model.
 +
 +==== Summary ====
 +
 +**Models** = The brains \\
 +**Runtimes** = The engines that run the brains \\
 +**Training tools** = The machines that modify the brains \\
 ---- ----
 +===== 3. TRAINING TOOLS (NEURAL LEARNING, “NEW SYNAPSES”) =====
 +----
 +This is where real learning happens. Not prompts. Not memory. \\
 +Not RAG. (library model), Actual neural adjustment. \\
 +  \code\
  
 +Tools include:
 +**✔ Axolotl**
 +
 +  * THE standard for fine-tuning LLaMA, Mistral, etc.
 +  * Used by researchers, labs, and giants.
 +
 +**✔ Unsloth**
 +
 +  * A faster, more GPU-efficient training toolkit.
 +  * Great for consumer GPUs like yours.
 +
 +**✔ Faraday**
 +  * A more experimental training tool, but advanced.
 + \\
 +**  * These tools:
 +**  * Read your dataset
 +  * Modify the model’s weights
 +  * Produce a new personalized model
  
 +This is real learning.