Posted here first: https://www.reddit.com/r/Oobabooga/comments/192qb2c/mermaidmistral_a_work_in_progress_model_for_flow/
It's kinda barely under 16GB in full precision, you can give it a try using I threw this together but be warned it will OOM on long context due to Free Tier being limited to 15GB VRAM.
Introduction:
Introducing MermaidMistral, a powerful yet compact 7-billion-parameter language model adept at Python code understanding and crafting engaging story flow maps. Trained on a meticulously hand curated dataset of 478 diverse Python examples and hand crafted mermaid flow maps utilizing https://mermaid.live, this model goes beyond its size to deliver exceptional performance in code understanding and story visualization.
Key Features:
MermaidMistral is not a "Chatty Kathy" and should only respond with a mermaid code block with a flow diagram in mermaid js syntax and nothing more.
1. Code Understanding:
- Grasps Python intricacies with finesse.
- Generates clear and accurate Mermaid Diagram Flow Charts.
- Ideal for developers seeking visual representations of their code's logic.
2. Storytelling Capabilities:
- Converts narrative inputs into captivating Mermaid Diagrams.
- Maps character interactions, plot developments, and narrative arcs effortlessly.
3. Unmatched Performance:
- Surpasses larger models, like GPT-4, in generating well-organized and detailed Mermaid Diagrams for story flows.
4. Training Insights:
- Trained on a 478 Python examples for just under three epochs on a single RTX 3090 following batch size equal to 1, known as stochastic gradient descent.
- Exhibited emergent properties in story-to-flow map translations.
- Adaptable and efficient in resource utilization
- Due to hardware constraints this fine tune has a token limit of 2048.
Collaboration:
MermaidMistral is open to collaboration to further strengthen its capabilities. The dataset, formatted in Alpaca, provides a unique foundation for understanding Python intricacies. If you're interested in contributing or collaborating to enhance the model's performance, feel free to reach out to [email protected]. Your expertise could play a pivotal role in refining MermaidMistral.
Example Use Cases:
1. Code Documentation:
- Developers can use MermaidMistral to automatically generate visual flow charts from their Python code, aiding in documentation and code understanding.
2. Storyboarding:
- Storytellers and writers can input their narrative and receive visually appealing Mermaid Diagrams, offering a structured overview of character interactions and plot progression.
3. Project Planning:
- Project managers can leverage MermaidMistral to create visual project flow maps, facilitating effective communication and planning among team members.
4. Learning Python:
- Students and beginners can use MermaidMistral to visually understand Python code structures, enhancing their learning experience.
5. Game Design:
- Game developers can utilize MermaidMistral for visualizing game storylines, ensuring a coherent narrative structure and character development.
Proof of Concept:
MermaidMistral proves that innovation thrives in compact packages, delivering exceptional performance across diverse applications. Its adaptability and efficiency showcase the potential for groundbreaking results even in resource-constrained environments.
These mermaid codeblocks can be converted directly into images using mermaid cli tool found here: https://github.com/mermaid-js/mermaid-cli
I plan to release my working proof of concept VSCode Extension that currently displays the Live Flow Map
every time a user stops typing for more than 10 seconds.
Stay tuned.
Example Story -> Flow
https://chat.openai.com/share/e3163857-981b-4968-b2db-98ad869c9259
Insights on how to get best results
For best results use full precision using one of the 3 different instruction types:
- "instruction": "Create the mermaid diagram for the following code:",
- "instruction": "Create the mermaid diagram for the following story:",
- "instruction": "Create the mermaid diagram for the following:",
- Downloads last month
- 233