Text Classification
Transformers
Safetensors
English
qwen3
text-generation
decision-model
structured-decisions
calibration
probability-estimation
text-embeddings-inference
Instructions to use TriusAI/kapteeni-v1-meticulous with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use TriusAI/kapteeni-v1-meticulous with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="TriusAI/kapteeni-v1-meticulous")# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("TriusAI/kapteeni-v1-meticulous") model = AutoModelForCausalLM.from_pretrained("TriusAI/kapteeni-v1-meticulous", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Card: GitHub repo + variant cross-links
Browse files- README.md +5 -3
- kapteeni/pack.py +13 -7
README.md
CHANGED
|
@@ -21,7 +21,7 @@ interface (the wire format of Jev, `POST /v1/systemone`).
|
|
| 21 |
|
| 22 |
**This variant: kapteeni-v1-meticulous** — conservative confidence. The safe default for unknown or messy traffic.
|
| 23 |
|
| 24 |
-
Kapteeni v1 ships as two variants of the same architecture. The other, kapteeni-v1-intuit, decides more accurately on well-formed numeric, temporal, and multi-step policy traffic but is less careful about its confidence there.
|
| 25 |
|
| 26 |
## Quickstart
|
| 27 |
|
|
@@ -50,8 +50,10 @@ m = SystemOneModel.from_dist("<snapshot dir>")
|
|
| 50 |
answers, usage = m.evaluate(state, questions)
|
| 51 |
```
|
| 52 |
|
| 53 |
-
The `kapteeni/` Python package is included in this snapshot; the full
|
| 54 |
-
(data pipeline, training, tests
|
|
|
|
|
|
|
| 55 |
|
| 56 |
## Question types (the Jev contract)
|
| 57 |
|
|
|
|
| 21 |
|
| 22 |
**This variant: kapteeni-v1-meticulous** — conservative confidence. The safe default for unknown or messy traffic.
|
| 23 |
|
| 24 |
+
Kapteeni v1 ships as two variants of the same architecture. The other, [kapteeni-v1-intuit](https://huggingface.co/TriusAI/kapteeni-v1-intuit), decides more accurately on well-formed numeric, temporal, and multi-step policy traffic but is less careful about its confidence there.
|
| 25 |
|
| 26 |
## Quickstart
|
| 27 |
|
|
|
|
| 50 |
answers, usage = m.evaluate(state, questions)
|
| 51 |
```
|
| 52 |
|
| 53 |
+
The `kapteeni/` Python package is included in this snapshot; the full
|
| 54 |
+
repository (data pipeline, training, tests, benchmark history and
|
| 55 |
+
pre-registrations) lives at https://github.com/TriusAI/Kapteeni
|
| 56 |
+
(Apache-2.0).
|
| 57 |
|
| 58 |
## Question types (the Jev contract)
|
| 59 |
|
kapteeni/pack.py
CHANGED
|
@@ -84,8 +84,10 @@ m = SystemOneModel.from_dist("<snapshot dir>")
|
|
| 84 |
answers, usage = m.evaluate(state, questions)
|
| 85 |
```
|
| 86 |
|
| 87 |
-
The `kapteeni/` Python package is included in this snapshot; the full
|
| 88 |
-
(data pipeline, training, tests
|
|
|
|
|
|
|
| 89 |
|
| 90 |
## Question types (the Jev contract)
|
| 91 |
|
|
@@ -162,9 +164,11 @@ _VARIANT_TEXT = {
|
|
| 162 |
"**This variant: kapteeni-v1-meticulous** — conservative "
|
| 163 |
"confidence. The safe default for unknown or messy traffic.\n\n"
|
| 164 |
"Kapteeni v1 ships as two variants of the same architecture. "
|
| 165 |
-
"The other,
|
| 166 |
-
"
|
| 167 |
-
"
|
|
|
|
|
|
|
| 168 |
),
|
| 169 |
"bench": (
|
| 170 |
"| JevBench-style score | Intelligence | top-label ECE | public accuracy |\n"
|
|
@@ -197,8 +201,10 @@ _VARIANT_TEXT = {
|
|
| 197 |
"policy traffic (measured skills-slice accuracy 0.81 vs 0.66 "
|
| 198 |
"for -meticulous, with better calibration on that slice).\n\n"
|
| 199 |
"Kapteeni v1 ships as two variants of the same architecture. "
|
| 200 |
-
"The other,
|
| 201 |
-
"
|
|
|
|
|
|
|
| 202 |
),
|
| 203 |
"bench": (
|
| 204 |
"| JevBench-style score | Intelligence | top-label ECE | public accuracy |\n"
|
|
|
|
| 84 |
answers, usage = m.evaluate(state, questions)
|
| 85 |
```
|
| 86 |
|
| 87 |
+
The `kapteeni/` Python package is included in this snapshot; the full
|
| 88 |
+
repository (data pipeline, training, tests, benchmark history and
|
| 89 |
+
pre-registrations) lives at https://github.com/TriusAI/Kapteeni
|
| 90 |
+
(Apache-2.0).
|
| 91 |
|
| 92 |
## Question types (the Jev contract)
|
| 93 |
|
|
|
|
| 164 |
"**This variant: kapteeni-v1-meticulous** — conservative "
|
| 165 |
"confidence. The safe default for unknown or messy traffic.\n\n"
|
| 166 |
"Kapteeni v1 ships as two variants of the same architecture. "
|
| 167 |
+
"The other, "
|
| 168 |
+
"[kapteeni-v1-intuit](https://huggingface.co/TriusAI/kapteeni-v1-intuit), "
|
| 169 |
+
"decides more accurately on well-formed numeric, temporal, "
|
| 170 |
+
"and multi-step policy traffic but is less careful about its "
|
| 171 |
+
"confidence there."
|
| 172 |
),
|
| 173 |
"bench": (
|
| 174 |
"| JevBench-style score | Intelligence | top-label ECE | public accuracy |\n"
|
|
|
|
| 201 |
"policy traffic (measured skills-slice accuracy 0.81 vs 0.66 "
|
| 202 |
"for -meticulous, with better calibration on that slice).\n\n"
|
| 203 |
"Kapteeni v1 ships as two variants of the same architecture. "
|
| 204 |
+
"The other, "
|
| 205 |
+
"[kapteeni-v1-meticulous](https://huggingface.co/TriusAI/kapteeni-v1-meticulous), "
|
| 206 |
+
"is the conservative-confidence default for unknown or messy "
|
| 207 |
+
"traffic."
|
| 208 |
),
|
| 209 |
"bench": (
|
| 210 |
"| JevBench-style score | Intelligence | top-label ECE | public accuracy |\n"
|