Spaces:

---

huggingface-projects

/

gemma-3-12b-it

like 131

Running on Zero

Q: Describe the image

Here's a description of the image you sent:

Overall Description:

The image is a diagram illustrating the evolution of Large Language Model (LLM) usage from 2023 to January 2025. It depicts a progression from basic inference to more complex reasoning using agents. The diagram is divided into three columns, each representing a year and showcasing a different approach to LLM utilization.

Column Breakdown (Left to Right):

2023: Basic LLM Inference
    Process: The diagram shows a simple flow: "Query" goes into an LLM, which has undergone "PreTrain," "Supervised Finetuning," and "Direct Preference Optimization." The LLM then produces an "Answer."
    Time: The inference is done in "milliseconds."
    Label: "Basic LLM Inference"
2024: RAG based LLM Inference
    Process: Similar to 2023, but with the addition of "Retrieval Augmented Generation." This involves retrieving "Additional data" from a database (represented by a cylinder icon) and using it in conjunction with the LLM to generate an "Answer."
    Time: The inference takes "10s of seconds."
    Label: "RAG based LLM Inference"
Jan 2025: Advanced Reasoning with Agents
    Process: This is the most complex flow. The "Query" is processed through the LLM (with the same PreTrain, Finetuning, and Optimization steps). However, instead of a direct answer, the LLM engages in "Reasoning" and then interacts with "Search Agent" and "Code Agent" through further "Reasoning" steps. Finally, an "Answer" is produced.
    Time: The inference takes "Minutes."
    Label: "Advanced Reasoning with Agents"

Common Elements Across Columns:

LLM: Each column features a box labeled "LLM" representing the Large Language Model.
PreTrain, Supervised Finetuning, Direct Preference Optimization: These are listed as steps that the LLM undergoes.
Components: Each column has a "Components" section with a vertical list of letters