DreamGenX

Link: https://www.youtube.com/watch?v=EMOFRDOMIiU

Really solid interview, made me feel hopeful about the future of open source models at Mistral. He also elaborates how they think about model safety. In summary: base model should not have training targeted at safety as it degrades performance, but tools will be provided to make safety filtering on top easier, so users can opt into it. He also hinted at some sort of platform they are building (to actually make money, I guess).

I hope they can keep their current approach. It might be hard as they are VC funded, and as the company grows.

TL;DR:

• Uncensored, Llama 2 70B based model for collaborative/steerable story writing and chat / role-play.
• Hugging Face link:
  • dreamgen/opus-v0-70b
  • dreamgen/opus-v0-70b-awq
  • dreamgen/opus-v0-70b-gguf
• Try for free on dreamgen.com. Don't forget to select it in the editor, otherwise you get the 7B.

Hey everyone, I am following up on my post from last week, and releasing a 70B version of DreamGen Opus V0.

What is collaborative / steerable story writing?

You can find more details about prompting the model in the official prompting guide. Make sure to checkout the examples, like “You can stop time. What will you do?”, “Natasha Romanoff and me…” or “The mischievous Coven”.

In summary: It allows you to describe a story, generate / write a part of it, and then further steer the story with inline instructions, like:

(Initial description of the story)
(Can include character description)


(Previously generated or written part of the story)


(Inline instructions describing how the story should develop next)
(The model will steer it in the direction within the next few sentences / paragraphs)

How to do chat / role-play?

Very similar to regular story writing, but:

• In the story description, mention that it’s written in the form of a dialog or in first person (depending on preferences).
• Preferably include character definitions.
• After ``, kick off the chat / role-play with 2-3 example dialog turns, using the format you want, e.g.:

See the examples tagged as "chat" from the official guide linked above.

How to try it out

The model should work great with any tool that supports the Llama 70B base model.

The model is large, therefore not everyone might be able to run it locally. To that end, you can try it for free on dreamgen.com (at least while I can support it :)).

What’s next

Last time I asked the community what I should work on next (you can still vote here). The most requested thing was better support for creating & sharing characters, both on the tooling side & model side (there’s tons of headroom when it comes to data, etc.) — stay tuned for updates in the coming weeks.

Want to join the fun and influence the direction of DreamGen? Join our community server.

Hello!

By popular demand I am planning a fine-tune of https://huggingface.co/dreamgen/opus-v0-7b on top of Yi-34B and wonder whether to use the 200K as the base.

The regular Yi-34B seems slightly better than Yi-34B-200K on standard benchmarks, but I wonder how it "feels" and whether the loss of performance on short context is worth it, given that the regular version can be used up to 32K tokens.

(Yi-34B vs Yi-34B-200K)

Did anyone try an analysis of these 2 models on various sequence lengths (<4K, <8K, <16K, etc.)?

​

TL;DR:

• Uncensored, Mistral 7B based model that lets you write stories in collaborative fashion, but also works nicely for chat / (E)RP
• Hugging Face link: https://huggingface.co/dreamgen/opus-v0-7b

Hey everyone, I am excited to share with you the first release of “DreamGen Opus”, an uncensored model that lets you write stories in collaborative fashion, but also works nicely for chat / (E)RP.

Specifically, it understands the following prompt syntax (yes, another one — please don’t hate :D):

(Description of the story, can also optionally include information about characters) 

...


(Instructions as you write the story, to guide the next few sentences / paragraphs)

You can find more details about prompting the model in the official prompting guide, including a few examples (like for chat / ERP).

The initial model is based on Mistral 7B, but Llama 2 70B version is in the works and if things go well, should be out within 2 weeks (training is quite slow :)).

The model is based on a custom dataset that has >1M tokens of instructed examples like the above, and order of magnitude more examples that are a bit less instructed.

How to try it out

The model should work great with any tool that supports the Mistral 7B base model. It will work well with oobabooga/text-generation-webui and many other tools. I like vLLM.

Using vLLM

• Install vLLM following the instructions in the repo
• Run python -u -m vllm.entrypoints.openai.api_server --host 0.0.0.0 --model dreamgen/opus-v0-7b

Using DreamGen.com website (free)

You can also try the model on dreamgen.com for free (but it requires a registration with email).

What’s next

I believe that for story telling & character creation it’s especially important to have access to the model weights, otherwise you run the risk of losing your plot or virtual companion (as already happened a few times before on various closed platforms that suddenly changed their rules or got shut down by their API provider). Hence DreamGen.

Here’s a high level overview of what I would like to do next under the DreamGen umbrella:

On the model side:

• (Soon) Larger story models
• Fine tune the model for even better character chat & roleplay
• Longer context windows, at least for smaller models (8-16K depending on how experiments go)

On the application side, I am thinking about these features:

• Character editor, chat & roleplay
• Ability to share your stories privately & publicly (not sure about this one, to be honest :))
• Image generation to go alongside with story generation & chat
• API so that you can use the model more easily if you don’t have a GPU

For all of these, I would love your input! You can vote on the roadmap here.

For more updates, join the community server or follow updates on Twitter.