69 lines
4.9 KiB
Markdown
69 lines
4.9 KiB
Markdown
|
---
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
license: apache-2.0
|
||
|
|
tags:
|
||
|
|
- text-generation-inference
|
||
|
|
- transformers
|
||
|
|
- unsloth
|
||
|
|
- mistral
|
||
|
|
- trl
|
||
|
|
base_model: unsloth/mistral-7b-bnb-4bit
|
||
|
|
datasets:
|
||
|
|
- gate369/Alpaca-Star
|
||
|
|
---
|
||
|
|
- QSTAR
|
||
|
|
- I think this model is proof of my theory that you dont need a special architecture to train a llm to reason. The techniques i imployed to make this could
|
||
|
|
be greatly expanded on and coupled with a agent system to achieve a functioning low cost agi system.
|
||
|
|
|
||
|
|
- Steps: -Loss:
|
||
|
|
- 1 - 1.373000
|
||
|
|
- 2 - 1.551400
|
||
|
|
- 3 - 1.083100
|
||
|
|
- 4 - 1.164900
|
||
|
|
- 5 - 1.196500
|
||
|
|
- 6 - 1.015400
|
||
|
|
- ....
|
||
|
|
- 363 - 0.019000
|
||
|
|
- 364 - 0.017600
|
||
|
|
- 365 - 0.015600
|
||
|
|
- 366 - 0.016700
|
||
|
|
- 367 - 0.016400
|
||
|
|
- 368 - 0.016600
|
||
|
|
- 369 - 0.017600
|
||
|
|
- heres a conversation i had with the model:
|
||
|
|
- user: "hi there. id like to build a time machine. how might i go about achieving this?"
|
||
|
|
- assistant: "|startthought| To build a time machine, I need to figure out how to move through time just like we move through space. That means finding a way to change my
|
||
|
|
frame of reference from the stationary Earth to one that's moving forward in time. |endthought| One approach is to use Einstein's theory of relativity and massive
|
||
|
|
objects with strong gravitational fields, like black holes or neutron stars. By warping spacetime, these objects can slow down time for an observer passing by.
|
||
|
|
Another idea is to explore quantum mechanics and the notion of entanglement between particles. If I could create a pair of entangled particles and keep one, the
|
||
|
|
other might appear elsewhere in space-time, effectively teleporting it through time. Yet another approach is to look at cosmic rays and high-energy particle
|
||
|
|
collisions, which can create short-lived particles that seem to violate the forward-in-time direction of time. Understanding these phenomena might lead to a
|
||
|
|
way to control time flow. Ultimately, building a functional time machine will require advancing our understanding of these fundamental physical principles and
|
||
|
|
finding a way to apply them technologically. It's a grand challenge that has captivated scientists and science fiction writers alike for centuries."
|
||
|
|
- user: "how might we make use of this phenomena to create a functioning time machine?"
|
||
|
|
- assistant: "|startthought| To make use of these phenomenas to create a funcitoning time machine, we need to find a way to control and harness them technologically.
|
||
|
|
That could mean developing advanced gravitational lensing techniques to manipulate space-time around massive objects, exploring quantum entanglement
|
||
|
|
in more complex systems, or finding a way to generate the high-energy particle collision conditions needed to create tachyons. It's likely that we'll need to
|
||
|
|
understand and combine multiple approaches to make a functional time machine a reality. |endthought| One approach might be to focus on developing advanced
|
||
|
|
gravitational lensing techniques. This could allow us to control the warping of space-time around massive objects like black holes or neutrino stars, creating
|
||
|
|
stable time vortexes that we can pass through. Another idea is to explore quantum entanglement in more complex systems. If we could create and maintain a
|
||
|
|
large-scale entangled state between two particles, we might be able to use one particle as a clock while the other appears elsewhere in space-time, effectively
|
||
|
|
teleporting it through time. Yet another approach is to look at high-energy particle collision phenomenas like those created in large hadron colliders or
|
||
|
|
supernova explosions. These events can generate short-lived particles called tachyons, which seem to move faster than the speed of light and violate the
|
||
|
|
forward-in-time direction of time. If we could understand and control this phenomena, we might be able to create a stable source of tachyons that could be
|
||
|
|
used to power a time machine. Ultimately, making a functional time machine will require advancing our understanding of these fundamental physical principles
|
||
|
|
and finding a way to apply them technologically. It's a grand challenge that has captivaed scientists and science fiction writers alike for centuries."
|
||
|
|
|
||
|
|
|
||
|
|
|
||
|
|
- the data set will need more work but this shows insane potential
|
||
|
|
# Uploaded model
|
||
|
|
|
||
|
|
- **Developed by:** liminerity
|
||
|
|
- **License:** apache-2.0
|
||
|
|
- **Finetuned from model :** unsloth/mistral-7b-bnb-4bit
|
||
|
|
This model was trained on a alpaca based dataset i made using perplexity and claude 3. The dataset is meant to encourage the llm the think before speaking
|
||
|
|
This mistral model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
|
||
|
|
|
||
|
|
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|