Elinas

Machine Learning/MLOps Engineer | LLM focus | |

[Elinas @officialelinas]

Chronos 70B v2 release! Thanks to Pygmalion for generously providing the compute and @TheBlokeAI for quantizing the model. As usual, the model optimized for chat, roleplay, storywriting, and now includes vastly improved reasoning skills.
https://t.co/P5NLl9fMSB

[Elinas @officialelinas]

@jobmontefalco @Teknium codex is for engineering, not for your fiction

[Elinas @officialelinas]

@Teknium @ar0cket1 Can I get access? I promise I won't do harm, just pentesting my lab. ;)

[Elinas @officialelinas]

@clashreport Here we go again.

[Elinas @officialelinas]

@Teknium Dirty deletes, nice.

[Elinas @officialelinas]

@hkeydesign @Teknium Try opencode `opencode agent create` - i use obsidian and it has my stand up talking points ready daily.

[Elinas @officialelinas]

@Teknium Opencode has changed me as an engineer. It's the engineer/PO/PM now.

[Elinas @officialelinas]

@Teknium Why does it being a reasoning model or not matter? Model performance on the leaderboards is what every model is chasing now ;)

[Elinas @officialelinas]

@InglesonSi44993 @RealAlexJones @caseyplummer4 It's iPhones with Apple intelligence and iPad pros. Even if you don't have Apple intelligence enabled, it still does it. Turning off WiFi makes it so it doesn't happen.

[Elinas @officialelinas]

@valhalla_dev @yacineMTB So is X, just a different format.

[Elinas @officialelinas]

@sainingxie Scores highest in mathematics, though I don't believe transformers can actually do it. Don't know what the new OAI models bench at but likely eval fodder.

[Elinas @officialelinas]

@sainingxie Everyone sleeping on Gemini...

[Elinas @officialelinas]

Chronos has returned, but new model series, using Mistral Nemo 12B as a base. I think most people who enjoy RP and storywriting will be happy with this model. More details on the card as well as quants. Thanks and enjoy! https://t.co/GXQbC31AGa
https://t.co/mfyGnVEiyX

[Elinas @officialelinas]

@kaiokendev1 @kalomaze I'm not being antagonistic, if anything you are. All I stated is that the LLM can output many interesting tokens. What is verifiable you haven't produced yet.

[Elinas @officialelinas]

@kaiokendev1 @kalomaze Realistically, this means nothing. Any closed LLM can produced whatever. Do you work at Anthropic?

[Elinas @officialelinas]

@kaiokendev1 @kalomaze You're using 2 different APIs. The second is chat completions. This makes no sense.

[Elinas @officialelinas]

Llama-3-13B-Instruct & finetune

The goal was to create a "mid" size model as Meta has released before. See the model cards for more detail, particularly on the finetuning of the base model and preliminary results.

Base: https://t.co/Ahx4djSlOD
Finetune: https://t.co/3ktdhSYiVW

[Elinas @officialelinas]

Releasing Chronos-Mistral-7B after holding onto it for weeks. Better performance compared to Chronos 13B-v2. Check out the model card. Using the original dataset, but very cleaned up. This is an experimental model and more will come in the future.
https://t.co/MHtzPS3P4L

[Elinas @officialelinas]

@teknium @teknium What about SOLAR?

[Elinas @officialelinas]

@teknium He said it's just a csv, and got a takedown for that? Open source is slowly getting dismantled by corporations with their own interests in mind.

[Elinas @officialelinas]

@DanielCardena @TheBlokeAI Primarily because it's 70B and I've tweaked the dataset to be cleaner. You can also try a merge I did to attempt to get some more logical variation in the model which can be found here: https://t.co/QUpU1f9Rzu

[Elinas @officialelinas]

Chronos 70B v2 release! Thanks to Pygmalion for generously providing the compute and @TheBlokeAI for quantizing the model. As usual, the model optimized for chat, roleplay, storywriting, and now includes vastly improved reasoning skills.
https://t.co/P5NLl9fMSB