Upload README.md with huggingface_hub

README.md

@@ -94,6 +94,22 @@ pipe({
 
 Optional `criteria` overrides the yes/no wording: `{"true": "...", "false": "..."}`.
 
+Another text, different questions: tone and content checks.
+
+```python
+pipe({
+    "state": "Thanks so much for the quick help, the new dashboard looks great!",
+    "questions": {
+        "polite": {"type": "noul", "instructions": "Is the text polite?"},
+        "complaint": {"type": "noul", "instructions": "Does the text contain a complaint?"},
+        "question": {"type": "noul", "instructions": "Does the text ask a question?"},
+    },
+})["answers"]
+# {"polite":    {"type": "noul", "noul": 0.9991},
+#  "complaint": {"type": "noul", "noul": 0.0009},
+#  "question":  {"type": "noul", "noul": 0.0080}}
+```
+
 ### `choice`: pick one option
 
 `criteria` maps each option label to a description. You get the top label,
@@ -120,6 +136,30 @@ pipe({
 #           "confidence": 0.984}}
 ```
 
+Intent classification with four options:
+
+```python
+pipe({
+    "state": "Can you add a dark mode to the mobile app? My eyes hurt at night.",
+    "questions": {
+        "intent": {
+            "type": "choice",
+            "instructions": "What is the intent of the message?",
+            "criteria": {
+                "bug_report": "Something is broken or not working as expected",
+                "feature_request": "Asks for new functionality or an improvement",
+                "praise": "Compliments or thanks",
+                "question": "Asks how to do something",
+            },
+        },
+    },
+})["answers"]
+# {"intent": {"type": "choice", "choice": "feature_request",
+#             "probabilities": {"feature_request": 0.9781, "bug_report": 0.0094,
+#                               "question": 0.0079, "praise": 0.0046},
+#             "confidence": 0.907}}
+```
+
 ### `score`: ordered scale
 
 `criteria` is an ordered list of levels (lowest first, 2 to 26 levels). `score`
@@ -146,6 +186,27 @@ pipe({
 #  "probabilities": {"0": 0.0018, "1": 0.0542, "2": 0.9440}, "confidence": 0.796}
 ```
 
+A five-point scale on mixed feedback. The distribution is spread out, so
+`score` lands between levels and `confidence` is low:
+
+```python
+pipe({
+    "state": "The delivery was two days late, the box was dented, but the product itself works fine.",
+    "questions": {
+        "satisfaction": {
+            "type": "score",
+            "instructions": "How satisfied is the customer overall?",
+            "criteria": ["Very dissatisfied", "Dissatisfied", "Neutral", "Satisfied", "Very satisfied"],
+        },
+    },
+})["answers"]["satisfaction"]
+# {"type": "score", "score": 1.408,
+#  "legend": {"0": "Very dissatisfied", "1": "Dissatisfied", "2": "Neutral",
+#             "3": "Satisfied", "4": "Very satisfied"},
+#  "probabilities": {"0": 0.1144, "1": 0.5705, "2": 0.1635, "3": 0.0957, "4": 0.0560},
+#  "confidence": 0.223}
+```
+
 ### Mixing types in one call
 
 All questions in a call see the same `state` and run in one batched forward