Update README.md
Browse files
README.md
CHANGED
|
@@ -31,7 +31,7 @@ A functional “blank mind” before reasoning specialization
|
|
| 31 |
|
| 32 |
What this model is not
|
| 33 |
|
| 34 |
-
❌ Not a chatbot
|
| 35 |
|
| 36 |
❌ Not instruction-tuned
|
| 37 |
|
|
@@ -58,7 +58,7 @@ A clear separation between language competence and reasoning behavior
|
|
| 58 |
Many projects hide their base models.
|
| 59 |
Axion does the opposite.
|
| 60 |
|
| 61 |
-
Intended use
|
| 62 |
|
| 63 |
Research and experimentation
|
| 64 |
|
|
@@ -68,7 +68,7 @@ Studying the effects of reasoning-oriented datasets
|
|
| 68 |
|
| 69 |
Serving as a backbone for Axion1.5-Reasoning variants
|
| 70 |
|
| 71 |
-
Limitations
|
| 72 |
|
| 73 |
Because this model is trained only for next-token prediction:
|
| 74 |
|
|
@@ -82,7 +82,7 @@ It may hallucinate or contradict itself
|
|
| 82 |
|
| 83 |
These limitations are expected and acknowledged.
|
| 84 |
|
| 85 |
-
Future work
|
| 86 |
|
| 87 |
This release is part of a broader project:
|
| 88 |
|
|
@@ -94,7 +94,7 @@ Experiments with short, verifiable reasoning traces
|
|
| 94 |
|
| 95 |
The base model will remain unchanged to preserve its value as a reference.
|
| 96 |
|
| 97 |
-
Philosophy
|
| 98 |
|
| 99 |
Scale is not intelligence.
|
| 100 |
Structure matters.
|
|
@@ -104,6 +104,6 @@ Axion explores whether smaller models, trained with the right constraints, can d
|
|
| 104 |
This is an experiment.
|
| 105 |
And experiments are allowed to fail.
|
| 106 |
|
| 107 |
-
Acknowledgements
|
| 108 |
|
| 109 |
Created as an independent research project focused on understanding how reasoning emerges in language models.
|
|
|
|
| 31 |
|
| 32 |
What this model is not
|
| 33 |
|
| 34 |
+
❌ **Not a chatbot**
|
| 35 |
|
| 36 |
❌ Not instruction-tuned
|
| 37 |
|
|
|
|
| 58 |
Many projects hide their base models.
|
| 59 |
Axion does the opposite.
|
| 60 |
|
| 61 |
+
**Intended use**
|
| 62 |
|
| 63 |
Research and experimentation
|
| 64 |
|
|
|
|
| 68 |
|
| 69 |
Serving as a backbone for Axion1.5-Reasoning variants
|
| 70 |
|
| 71 |
+
**Limitations**
|
| 72 |
|
| 73 |
Because this model is trained only for next-token prediction:
|
| 74 |
|
|
|
|
| 82 |
|
| 83 |
These limitations are expected and acknowledged.
|
| 84 |
|
| 85 |
+
****Future work**
|
| 86 |
|
| 87 |
This release is part of a broader project:
|
| 88 |
|
|
|
|
| 94 |
|
| 95 |
The base model will remain unchanged to preserve its value as a reference.
|
| 96 |
|
| 97 |
+
**Philosophy**
|
| 98 |
|
| 99 |
Scale is not intelligence.
|
| 100 |
Structure matters.
|
|
|
|
| 104 |
This is an experiment.
|
| 105 |
And experiments are allowed to fail.
|
| 106 |
|
| 107 |
+
**Acknowledgements**
|
| 108 |
|
| 109 |
Created as an independent research project focused on understanding how reasoning emerges in language models.
|