Dolphin 2.9.4 Gemma2 2b 🐬

Curated and trained by Eric Hartford and Cognitive Computations.

This one is special because I used GrokAdamW and Liger Kernel

GrokAdamW is intended to enable fast Grokking, to increase generalization. (I am not certain this occurred because this checkpoint is 4 epochs, and it probabaly take more epochs to achieve grok.)

Discord: https://discord.gg/h3K4XGj2RH

Our appreciation for the sponsors of Dolphin 2.9.4:

Crusoe Cloud - provided excellent on-demand 8xL40S node

This model is based on Google Gemma2 2b, and is governed by the Gemma license.

The base model has 128K context, and our finetuning used 8192 sequence length.

ollama run CognitiveComputations/dolphin-gemma2:2b

https://ollama.com/CognitiveComputations/dolphin-gemma2

Dolphin 2.9.4 uses ChatML prompt template format.

example:

<|im_start|>system
You are Dolphin, a helpful AI assistant.<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

Dolphin-2.9.4 has a variety of instruction following, conversational, and coding skills. It also has agentic abilities and supports function calling. It is especially trained to obey the system prompt, and follow instructions in many languages.

Dolphin is uncensored. We have filtered the dataset to remove alignment and bias. This makes the model more compliant. You are advised to implement your own alignment layer before exposing the model as a service. It will be highly compliant with any requests, even unethical ones. Please read my blog post about uncensored models. https://erichartford.com/uncensored-models You are responsible for any content you create using this model. Enjoy responsibly.