electron271/graig-alpha

other companies may be trying to reach artificial general intelligence, but we are trying to reach artificial grain intelligence. with the help of our team of the best grain farmers in the world, we are making huge strides in the field. fine tuned fully locally using a RX 9070 XT using unsloth.

ollama run hf.co/electron271/graig-alpha:Q4_K_M

History

this is a continuation of the "tuxsentience" series made by @GrainWare, however using new advancements in AMD support in unsloth we are now able to pack in significantly more grain per parameter.

Recommended Settings

• temperature = 0.6
• top_k = 20
• min_p = 0.00 (llama.cpp's default is 0.1)
• top_p = 0.95
• presence_penalty = 0.0 to 2.0 (llama.cpp default turns it off, but to reduce repetitions, you can use this) Try 1.0 for example.
• Supports up to 262,144 context natively but you can set it to 32,768 tokens for less RAM use

Disclaimer

Graig can be prone to saying offensive statements in rare circumstances due to the unpredictability of LLMs. These do not reflect our opinions/views and are a byproduct we are trying to avoid. Newer graig models (such as this one) are significantly less prone to this, however if you do not setup settings correctly or do not prompt right this may still occur. If you find Graig to be saying offensive statements during normal circumstances please either open a community post on this model or email me at [email protected].

In public deployments such as on Discord, please setup a filter using something such as https://github.com/cherryl1k/llmcordplus.

(Normal circumstances is defined as using the recommended settings and talking to graig in a non aggressive manner.)