mambiux/lumina-lexiR1-8B
Text Generation
•
Updated
•
32
mbx
mambiux
AI & ML interests
Synthetic Consciousness , AGI, Johnny-5, Mega Man, Tron, Wavelength-based computing, Analog computing, DIY Cyberpunk Home-Lab, you get the idea.
Recent Activity
updated
a model
27 days ago
mambiux/lumina-lexiR1-8B
published
a model
about 1 month ago
mambiux/lumina-lexiR1-8B
Organizations
None yet
mambiux's activity
updated
a
model
27 days ago
Check out this model merge, really good for Agentic Traits and Self Awareness
2
#657 opened 28 days ago
by
mambiux
published
a
model
about 1 month ago
reacted to
dylanebert's
post with 🔥
6 months ago
Post
2557
Here's a 1-minute video tutorial on how to fine-tune
unsloth/llama-3-8b-bnb-4bit with unsloth
Using Roller Coaster Tycoon peep thoughts as an example
Using Roller Coaster Tycoon peep thoughts as an example
reacted to
Undi95's
post with ❤️
8 months ago
Post
15417
Hello there,
New model released, my goal was to try finetune on the last Llama-3.1-8B-Instruct but not a small train, I wanted to do something useful.
One of the rare model that I didn't made for RP, or in the goal to uncensor it (but I did anyway kek).
The model was trained on 9M Claude conversations ONLY, giving him another writting style.
Undi95/Meta-Llama-3.1-8B-Claude > OG release fp32, it's the epoch 2
Undi95/Meta-Llama-3.1-8B-Claude-bf16 > Base model resharded in bf16 waiting for available quant without issues
Since it's frustrating to be censored using a local model, orthogonal activation steering was used, trying to force the model to never refuse a prompt.
Undi95/Meta-Llama-3.1-8B-Claude-68fail-3000total > Uncensored model, refuse 68 times on 3000 toxic prompt
Undi95/Meta-Llama-3.1-8B-Claude-39fail-3000total > Uncensored model, refuse 39 times on 3000 toxic prompt
It still refuse some prompt but the majority of them is uncensored. OAS can make a model more dumb or make the base perplexity go higher, so I didn't snipe for 0 refusal.
I don't do non-RP model a lot so any feedback is welcome, I would like to re-use this base for some others future project if needed.
New model released, my goal was to try finetune on the last Llama-3.1-8B-Instruct but not a small train, I wanted to do something useful.
One of the rare model that I didn't made for RP, or in the goal to uncensor it (but I did anyway kek).
The model was trained on 9M Claude conversations ONLY, giving him another writting style.
Undi95/Meta-Llama-3.1-8B-Claude > OG release fp32, it's the epoch 2
Undi95/Meta-Llama-3.1-8B-Claude-bf16 > Base model resharded in bf16 waiting for available quant without issues
Since it's frustrating to be censored using a local model, orthogonal activation steering was used, trying to force the model to never refuse a prompt.
Undi95/Meta-Llama-3.1-8B-Claude-68fail-3000total > Uncensored model, refuse 68 times on 3000 toxic prompt
Undi95/Meta-Llama-3.1-8B-Claude-39fail-3000total > Uncensored model, refuse 39 times on 3000 toxic prompt
It still refuse some prompt but the majority of them is uncensored. OAS can make a model more dumb or make the base perplexity go higher, so I didn't snipe for 0 refusal.
I don't do non-RP model a lot so any feedback is welcome, I would like to re-use this base for some others future project if needed.
- 4 replies
liked
a
Space
11 months ago
updated
a
model
12 months ago
Whats up with TheBloke ?
30
#2 opened about 1 year ago
by
Languido
Broken model!
1
#1 opened over 1 year ago
by
wolfram
Awesome work
1
#1 opened over 1 year ago
by
mambiux
airoboros
1
#1 opened over 1 year ago
by
mambiux
Holy stuff this is huge!!!! Cant wait for 70b GGML model!!!!
11
#1 opened over 1 year ago
by
rombodawg
Holy stuff this is huge!!!! Cant wait for 70b GGML model!!!!
11
#1 opened over 1 year ago
by
rombodawg
Holy stuff this is huge!!!! Cant wait for 70b GGML model!!!!
11
#1 opened over 1 year ago
by
rombodawg