Update README.md
Browse files
README.md
CHANGED
|
@@ -19,8 +19,6 @@ library_name: transformers
|
|
| 19 |
|
| 20 |
This is a **base scientific language model** (not instruction-tuned).
|
| 21 |
|
| 22 |
-
---
|
| 23 |
-
|
| 24 |
## Overview
|
| 25 |
|
| 26 |
KiteFish-A1-1.5B explores what it takes to train a domain-specialized scientific language model directly from structured LaTeX archives.
|
|
@@ -35,8 +33,6 @@ KiteFish-A1-1.5B explores what it takes to train a domain-specialized scientific
|
|
| 35 |
|
| 36 |
The focus of this project is *scientific language modeling robustness*, not benchmark optimization.
|
| 37 |
|
| 38 |
-
---
|
| 39 |
-
|
| 40 |
## Model Architecture
|
| 41 |
|
| 42 |
- 24 Transformer layers
|
|
@@ -57,8 +53,6 @@ The focus of this project is *scientific language modeling robustness*, not benc
|
|
| 57 |
|
| 58 |
**Validation Perplexity:** ~4.2 (held-out scientific corpus)
|
| 59 |
|
| 60 |
-
---
|
| 61 |
-
|
| 62 |
## Intended Use
|
| 63 |
|
| 64 |
KiteFish-A1-1.5B is suitable for:
|
|
@@ -76,8 +70,6 @@ It is **not optimized for:**
|
|
| 76 |
- General conversational AI
|
| 77 |
- Benchmark leaderboard performance
|
| 78 |
|
| 79 |
-
---
|
| 80 |
-
|
| 81 |
## Performance Notes
|
| 82 |
|
| 83 |
This model was trained under moderate compute constraints and without instruction tuning or alignment stages.
|
|
@@ -92,8 +84,6 @@ Observed characteristics:
|
|
| 92 |
|
| 93 |
Performance improves significantly with supervised fine-tuning (SFT), LoRA adaptation, or domain-specific instruction tuning.
|
| 94 |
|
| 95 |
-
---
|
| 96 |
-
|
| 97 |
## Limitations
|
| 98 |
|
| 99 |
- Not instruction-tuned
|
|
@@ -105,8 +95,6 @@ Performance improves significantly with supervised fine-tuning (SFT), LoRA adapt
|
|
| 105 |
|
| 106 |
This release is intended primarily for research and experimentation.
|
| 107 |
|
| 108 |
-
---
|
| 109 |
-
|
| 110 |
## Example Usage
|
| 111 |
|
| 112 |
```python
|
|
|
|
| 19 |
|
| 20 |
This is a **base scientific language model** (not instruction-tuned).
|
| 21 |
|
|
|
|
|
|
|
| 22 |
## Overview
|
| 23 |
|
| 24 |
KiteFish-A1-1.5B explores what it takes to train a domain-specialized scientific language model directly from structured LaTeX archives.
|
|
|
|
| 33 |
|
| 34 |
The focus of this project is *scientific language modeling robustness*, not benchmark optimization.
|
| 35 |
|
|
|
|
|
|
|
| 36 |
## Model Architecture
|
| 37 |
|
| 38 |
- 24 Transformer layers
|
|
|
|
| 53 |
|
| 54 |
**Validation Perplexity:** ~4.2 (held-out scientific corpus)
|
| 55 |
|
|
|
|
|
|
|
| 56 |
## Intended Use
|
| 57 |
|
| 58 |
KiteFish-A1-1.5B is suitable for:
|
|
|
|
| 70 |
- General conversational AI
|
| 71 |
- Benchmark leaderboard performance
|
| 72 |
|
|
|
|
|
|
|
| 73 |
## Performance Notes
|
| 74 |
|
| 75 |
This model was trained under moderate compute constraints and without instruction tuning or alignment stages.
|
|
|
|
| 84 |
|
| 85 |
Performance improves significantly with supervised fine-tuning (SFT), LoRA adaptation, or domain-specific instruction tuning.
|
| 86 |
|
|
|
|
|
|
|
| 87 |
## Limitations
|
| 88 |
|
| 89 |
- Not instruction-tuned
|
|
|
|
| 95 |
|
| 96 |
This release is intended primarily for research and experimentation.
|
| 97 |
|
|
|
|
|
|
|
| 98 |
## Example Usage
|
| 99 |
|
| 100 |
```python
|