Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,85 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: apache-2.0
|
3 |
+
language:
|
4 |
+
- en
|
5 |
+
tags:
|
6 |
+
- creative
|
7 |
+
- story
|
8 |
+
- writing
|
9 |
+
- fiction
|
10 |
+
- roleplaying
|
11 |
+
- rp
|
12 |
+
- horror
|
13 |
+
- science fiction
|
14 |
+
- fiction writing
|
15 |
+
- scene generation
|
16 |
+
- scene continue
|
17 |
+
- brainstorm 5x
|
18 |
+
- enhanced
|
19 |
+
- llama3
|
20 |
+
pipeline_tag: text-generation
|
21 |
+
---
|
22 |
+
|
23 |
+
<H3>BRAINSTORM 5x: Meta-Llama-3-8B Instruct, Formula #7 </H3>
|
24 |
+
|
25 |
+
This repo contains quants 5x of Meta-Llama-3-8B Instruct using the "Brainstorm" method of augmenting reasoning in a LLM
|
26 |
+
to increase it's performance at the core level for ANY creative use case(s).
|
27 |
+
|
28 |
+
This specific version has calibrations that allow it to exceed the logic solving abilities of the original model.
|
29 |
+
|
30 |
+
The BRAINSTORM process was developed by David_AU.
|
31 |
+
|
32 |
+
Some of the core principals behind this process are discussed in this <a href="https://arxiv.org/pdf/2401.02415">
|
33 |
+
scientific paper : Progressive LLaMA with Block Expansion </a>.
|
34 |
+
However I went in a completely different direction from what was outlined in this paper.
|
35 |
+
|
36 |
+
<B>What is "Brainstorm" ?</b>
|
37 |
+
|
38 |
+
The reasoning center of an LLM is taken apart, reassembled, and expanded by 5x.
|
39 |
+
Then these centers are individually calibrated. These "centers" also interact with each other. This introduces
|
40 |
+
subtle changes into the reasoning process. The calibrations further adjust - dial up or down - these "changes" further. The
|
41 |
+
number of centers (5x,10x etc) allow more "tuning points" to further customize how the model reasons so to speak.
|
42 |
+
|
43 |
+
The core aim of this process is to increase the model's detail, concept and connection to the "world", general concept connections, prose quality and prose length without affecting
|
44 |
+
instruction following. This will also enhance any creative use case(s) of any kind, including "brainstorming", creative art form(s) and like case uses.
|
45 |
+
|
46 |
+
Here are some of the enhancements this process brings to the model's performance:
|
47 |
+
|
48 |
+
- Prose generation seems more focused on the moment to moment.
|
49 |
+
- Sometimes there will be "preamble" and/or foreshadowing present.
|
50 |
+
- Fewer or no "cliches"
|
51 |
+
- Better overall prose and/or more complex / nuanced prose.
|
52 |
+
- A greater sense of nuance on all levels.
|
53 |
+
- Coherence is stronger.
|
54 |
+
- Description is more detailed, and connected closer to the content.
|
55 |
+
- Simile and Metaphors are stronger and better connected to the prose, story, and character.
|
56 |
+
- Sense of "there" / in the moment is enhanced.
|
57 |
+
- Details are more vivid, and there are more of them.
|
58 |
+
- Prose generation length can be long to extreme.
|
59 |
+
- Emotional engagement is stronger.
|
60 |
+
- The model will take FEWER liberties vs a normal model: It will follow directives more closely but will "guess" less.
|
61 |
+
- The MORE instructions and/or details you provide the more strongly the model will respond.
|
62 |
+
- Depending on the model "voice" may be more "human" vs original model's "voice".
|
63 |
+
|
64 |
+
Other "lab" observations:
|
65 |
+
|
66 |
+
- This process does not, in my opinion, make the model 5x or 10x "smarter" - if only that was true!
|
67 |
+
- However, a change in "IQ" was not an issue / a priority, and was not tested or calibrated for so to speak.
|
68 |
+
- From lab testing it seems to ponder, and consider more carefully roughly speaking.
|
69 |
+
- You could say this process sharpens the model's focus on it's task(s) at a deeper level.
|
70 |
+
|
71 |
+
The process to modify the model occurs at the root level - source files level. The model can quanted as a GGUF, EXL2, AWQ etc etc.
|
72 |
+
|
73 |
+
Other technologies developed by David_AU like "Ultra" (precision), "Neo Imatrix" (custom imatrix datasets), and "X-quants" (custom application of the imatrix process)
|
74 |
+
can further enhance the performance of the model along with the "Brainstorm" process.
|
75 |
+
|
76 |
+
The "Brainstorm" process has been tested on multiple LLama2, Llama3, and Mistral models of various parameter sizes, as well as on
|
77 |
+
"root" models like "Llama3 Instruct", "Mistral Instruct", and "merged" / "fine tuned" models too.
|
78 |
+
|
79 |
+
For original model specifications, usage information and other important details:
|
80 |
+
|
81 |
+
Special thanks to the model creators at META for making such a fantastic model:
|
82 |
+
|
83 |
+
[ https://huggingface.co/meta-llama/Meta-Llama-3-8B ]
|
84 |
+
|
85 |
+
More to follow...
|