You can deploy this model and start earning money today!

mombip/
Llama-3.1-8B-q4_k_m-manim

Name: Llama-3.1-8B-q4_k_m-manim
Availability: InStock

transformers

gguf

llama

text-generation-inference

Views

No views yet

Uploaded model

Developed by: mombip
License: apache-2.0
Finetuned from model : unsloth/meta-llama-3.1-8b-instruct-bnb-4bit

Description:

I'm now testing if it works so treat it as preview!!! This is a fine-tuned version of the LLaMA 3.1-8B model, specialized in the original 3b1b - Manim and ManimCE libraries.
The goal of this fine-tuning is to create a universal model that is both partially instructive and partially chat-like, allowing it to provide helpful and informative responses related to the Manim library.

Training Data:

Merged dataset from four sources:

Merge methodology:

From each above certain columns were mapped as Instruction (User) or Output (Assistant). Then all even elements of merged dtaset were formated as simplified Instructional input ("Instruction" and "Output"). Odd elements were fomratted as chat-like input (with User and Assistant flag)

Dataset size:

train split: 3,865
eval split: 430

Training method:

Fine-tuning of the unsloth/meta-llama-3.1-8b-instruct-bnb-4bit model using SFTTrainer

Metrics

eval_loss: 0.6

This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.

mombip/Llama-3.1-8B-q4_k_m-manim