Update model card: remove specific-LLM references, clarify variant purpose
Browse files
README.md
CHANGED
|
@@ -18,23 +18,6 @@ tags:
|
|
| 18 |
- reasoning-budget
|
| 19 |
base_model: Qwen/Qwen3.5-0.8B
|
| 20 |
pipeline_tag: text-classification
|
| 21 |
-
model-index:
|
| 22 |
-
- name: brick-complexity-2-max
|
| 23 |
-
results:
|
| 24 |
-
- task:
|
| 25 |
-
type: text-classification
|
| 26 |
-
name: Query Complexity Classification
|
| 27 |
-
dataset:
|
| 28 |
-
name: MMLU-Pro labeled 2K benchmark
|
| 29 |
-
type: regolo/brick-mmlu-pro-2k
|
| 30 |
-
split: test
|
| 31 |
-
metrics:
|
| 32 |
-
- type: accuracy
|
| 33 |
-
value: 0.7716
|
| 34 |
-
name: Accuracy (3-class)
|
| 35 |
-
- type: f1
|
| 36 |
-
value: 0.7707
|
| 37 |
-
name: Macro F1
|
| 38 |
---
|
| 39 |
|
| 40 |
<div align="center">
|
|
@@ -52,7 +35,7 @@ Classifier v2 is a family of small adapters that score each incoming prompt as *
|
|
| 52 |
|
| 53 |
<div align="center">
|
| 54 |
|
| 55 |
-
Maximum-accuracy variant tuned to classify query complexity as precisely as possible. Prioritizes routing quality over cost
|
| 56 |
|
| 57 |
**[Regolo.ai](https://regolo.ai) | [Brick SR1 on GitHub](https://github.com/regolo-ai/brick-SR1)**
|
| 58 |
|
|
@@ -74,22 +57,6 @@ Maximum-accuracy variant tuned to classify query complexity as precisely as poss
|
|
| 74 |
| **Output classes** | 3 (`easy`, `medium`, `hard`) |
|
| 75 |
| **License** | CC BY-NC 4.0 |
|
| 76 |
|
| 77 |
-
## Benchmark (MMLU-Pro labeled 2K)
|
| 78 |
-
|
| 79 |
-
| Metric | Value |
|
| 80 |
-
|---|---:|
|
| 81 |
-
| Accuracy (3-class) | 77.16% |
|
| 82 |
-
| Macro F1 | 0.7707 |
|
| 83 |
-
| Overestimate rate | 6.23% |
|
| 84 |
-
| Underestimate rate | 16.60% |
|
| 85 |
-
|
| 86 |
-
## Family Members
|
| 87 |
-
|
| 88 |
-
| Variant | Target | Accuracy | Macro F1 |
|
| 89 |
-
|---|---|---:|---:|
|
| 90 |
-
| [brick-complexity-2-eco](https://huggingface.co/regolo/brick-complexity-2-eco) | Cost savings (route toward easy tier) | 72.77% | 0.4246 |
|
| 91 |
-
| [brick-complexity-2-max](https://huggingface.co/regolo/brick-complexity-2-max) | Max routing accuracy | 77.16% | 0.7707 |
|
| 92 |
-
|
| 93 |
## Available Formats
|
| 94 |
|
| 95 |
| Format | Link |
|
|
|
|
| 18 |
- reasoning-budget
|
| 19 |
base_model: Qwen/Qwen3.5-0.8B
|
| 20 |
pipeline_tag: text-classification
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
---
|
| 22 |
|
| 23 |
<div align="center">
|
|
|
|
| 35 |
|
| 36 |
<div align="center">
|
| 37 |
|
| 38 |
+
Maximum-accuracy variant tuned to classify query complexity as precisely as possible. Prioritizes routing quality over cost.
|
| 39 |
|
| 40 |
**[Regolo.ai](https://regolo.ai) | [Brick SR1 on GitHub](https://github.com/regolo-ai/brick-SR1)**
|
| 41 |
|
|
|
|
| 57 |
| **Output classes** | 3 (`easy`, `medium`, `hard`) |
|
| 58 |
| **License** | CC BY-NC 4.0 |
|
| 59 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
## Available Formats
|
| 61 |
|
| 62 |
| Format | Link |
|