LFM-2.5-1.2b-Instruct-roleplay-tuned-v2

INTRODUCTION

This model is further improvement over the success of my previous LFM2.5 1.2b finetune, it was trained on a dataset thats over 3 times bigger than the one I used for v1(5m vs 17m tokens), and it serves to improve creative writing further.

I will have to admit that my gemma 3 4b it v2 finetune performed better, I'd say even surprisingly so, if you experience any issues with this model, consider getting my bigger tunes.

How good is it?

Advantages:

• Excellent writing
• Improved emotional understanding
• Size remains the same regardless, meaning anyone can run this model anywhere!

Disadvantages:

• Architecture remains the same, its still a 1.2b model. Though, I did do my best!

Quants(Speaking from personal experience with this specific model):

• BF16- Recommended, highest quality, least logical mistakes.
• Q8_0- Recommended, high quality, makes slightly more mistakes but nonetheless near lossless.
• Q6_K- Recommended if Q8_0 is too much, degradation begins, not exactly notable here, but you will notice minor detail loss.(When Mradermacher quantizes this model, I recommend getting his i1 Q6_K quant instead of the one I got in my repo, but in any case I still recommend Q8_0 or BF16)
• Q5_K_M- Recommended if hardware is really, REALLY bad, degradation becomes noticeable.
• Q4_K_M- Not recommended for most use cases, degradation is clearly noticeable.

Quants can be found in the repository, along with safetensors.

Thank you!

I appreciate that people actually found my initially released model interesting enough to download it, I didnt even think that it would attract any attention. But somehow, it did, I wanted anyone to be able to experience local roleplay, without the need for expensive hardware, without the need to pay for API keys, without the need to purchase subscriptions, and without the need to risk their data, and I believe small models like this one are very valuable for this task, because I know perfectly well that not everybody can run a 8b model.

v3:

v3 is ready, and it will be called Super-Slop-Machina-Roleplay-1.2b, I took a little different approach to it, the release is soon, wait until my horrible wifi finally loads it, but right now I can say enhancing it worked.