MediaWiki - User contributions [en]

AI on demand: Mapika/decider-31b

2026-10-19T08:35:02Z

Sst-djo: /* Overview */

= Overview =
System One decision models such as [https://huggingface.co/Mapika/decider-31b Mapika/decider-31b] assess text and return structured answers and probabilities in a single forward pass.

They support three question types:
* '''Choice'''
** Selects the single best option (or returns a probability distribution) from a fixed set of 2 to 255 declared options.

* '''Noul (Yes/No)'''
** Returns a boolean probability distribution for a binary question.

* '''Score'''
** Assigns a numeric value along a defined ordered scale (typically 2 to 10 levels) to represent a grading rubric or rating.

These models suit classification, routing, and decision support. They do not generate conversational answers or explanations, and their predictions can be wrong. Reported probabilities require evaluation on representative data before using them for consequential automation.

Our [https://huggingface.co/Mapika/decider-31b Mapika/decider-31b] implementation has a multilingual checkpoint with a fixed 40960-token input limit.

= Usage =
== Usage - Choice ==
The following '''choice''' example routes a request to a department.

Example call:
<syntaxhighlight lang="bash">
# Set your personal key:
STONEY_KEY=sk-....

# Execute the actual call:
curl --fail-with-body -sS \
https://llm.stoney-cloud.com/v1/systemone \
-H "apikey: $STONEY_KEY" \
-H 'Content-Type: application/json' \
--data-binary '{
"model": "Mapika/decider-31b",
"state": {
"document": "Meine Rechnung wurde zweimal belastet. Bitte prüfen Sie die doppelte Zahlung."
},
"questions": {
"department": {
"type": "choice",
"instructions": "Which department should handle this request?",
"criteria": {
"technical": "Application errors and login problems",
"billing": "Invoices, payments and duplicate charges",
"support": "General customer support and newsletter preferences"
}
}
}
}' | jq .
</syntaxhighlight>

Example answer:
<syntaxhighlight lang="json">
{
"model": "Mapika/decider-31b",
"answers": {
"department": {
"type": "choice",
"choice": "billing",
"confidence": 0.9264,
"x_p_max": 0.9509,
"certainty": 0.795,
"probabilities": {
"technical": 0.0351,
"billing": 0.9509,
"support": 0.014
}
}
},
"usage": {
"input_tokens": 88,
"output_tokens": 0,
"prompt_tokens": 88,
"completion_tokens": 0
}
}
</syntaxhighlight>

== Usage - Noul ==
The following '''noul''' example detects if customer explicitly requests a refund.

Example call:
<syntaxhighlight lang="bash">
# Set your personal key:
STONEY_KEY=sk-....

# Execute the actual call:
{
"model": "Mapika/decider-31b",
"answers": {
"refund_requested": {
"type": "noul",
"noul": 0.9327
}
},
"usage": {
"input_tokens": 89,
"output_tokens": 0,
"prompt_tokens": 89,
"completion_tokens": 0
}
}
</syntaxhighlight>

Example answer:
<syntaxhighlight lang="json">
"model": "Mapika/decider-31b",
"answers": {
"refund_requested": {
"type": "noul",
"noul": 0.9851
}
},
"usage": {
"input_tokens": 60,
"output_tokens": 23,
"prompt_tokens": 60,
"completion_tokens": 23
},
"latency_ms": 86.9
}
</syntaxhighlight>

== Usage - Score ==
The following '''score''' example estimates the urgency of a support request. The criteria is arranged from lowest to highest.

Example call:
<syntaxhighlight lang="bash">
# Set your personal key:
STONEY_KEY=sk-....

# Execute the actual call:
curl --fail-with-body -sS \
https://llm.stoney-cloud.com/v1/systemone \
-H "apikey: $STONEY_KEY" \
-H 'Content-Type: application/json' \
--data-binary '{
"model": "Mapika/decider-31b",
"state": {
"document": "Unser gesamtes Produktionssystem ist ausgefallen. Alle Kunden sind betroffen und können keine Bestellungen aufgeben. Wir benötigen sofort Hilfe."
},
"questions": {
"urgency": {
"type": "score",
"instructions": "How urgent is this support request?",
"criteria": [
"Not urgent: informational request with no operational impact",
"Low urgency: minor inconvenience with a workaround",
"Moderate urgency: partial disruption affecting some users",
"High urgency: major disruption requiring prompt attention",
"Critical urgency: complete production outage affecting all customers"
]
}
}
}' | jq .
</syntaxhighlight>

Example answer:
<syntaxhighlight lang="json">
{
"model": "Mapika/decider-31b",
"answers": {
"urgency": {
"type": "score",
"score": 3.9,
"confidence": 0.915,
"x_p_max": 0.9466,
"certainty": 0.8337,
"legend": {
"0": "Not urgent: informational request with no operational impact",
"1": "Low urgency: minor inconvenience with a workaround",
"2": "Moderate urgency: partial disruption affecting some users",
"3": "High urgency: major disruption requiring prompt attention",
"4": "Critical urgency: complete production outage affecting all customers"
},
"probabilities": {
"0": 0.0076,
"1": 0.011,
"2": 0.0039,
"3": 0.031,
"4": 0.9466
}
}
},
"usage": {
"input_tokens": 140,
"output_tokens": 0,
"prompt_tokens": 140,
"completion_tokens": 0
}
}
</syntaxhighlight>

[[Category:AI on Demand]]
[[Category: System One decision models]]

AI on demand: convaiinnovations/laya

2026-10-09T12:37:12Z

Sst-djo:

= System One Model - Introduction =

System One models such as Laya assess text and return structured decisions in a single forward pass.
They support three question types:
* '''choice'''
**selects a category, such as the department responsible for a ticket

* '''noul'''
** returns the probability that a statement is true, such as whether a customer requests a refund.

* '''score'''
** assesses text against an ordered scale, such as urgency from low to critical.

These models suit classification, routing, and decision support. They do not generate conversational answers or explanations, and their predictions can be wrong. Reported probabilities require evaluation on representative data before using them for consequential automation.

Our service uses Laya’s multilingual checkpoint with a fixed 4,096-token input limit.

= System One Model - Usage =

To use a System One model set your stoney key

<syntaxhighlight lang="bash">

STONEY_KEY=sk-....

</syntaxhighlight>

== System One Model - Usage choice ==

Use this category for example to route a request to a department.

'''Input:'''

<syntaxhighlight lang="bash">
curl --fail-with-body -sS \
https://llm.stoney-cloud.com/v1/systemone \
-H "apikey: $STONEY_KEY" \
-H 'Content-Type: application/json' \
--data-binary '{
"model": "convaiinnovations/laya",
"state": {
"document": "Meine Rechnung wurde zweimal belastet. Bitte prüfen Sie die doppelte Zahlung."
},
"questions": {
"department": {
"type": "choice",
"instructions": "Which department should handle this request?",
"criteria": {
"technical": "Application errors and login problems",
"billing": "Invoices, payments and duplicate charges",
"support": "General customer support and newsletter preferences"
}
}
}
}' | jq .
</syntaxhighlight>

'''Output:'''

<syntaxhighlight lang="json">
{
"model": "convaiinnovations/laya",
"answers": {
"department": {
"type": "choice",
"choice": "billing",
"probabilities": {
"technical": 0.0,
"billing": 0.9999,
"support": 0.0
},
"confidence": 0.9994,
"answer_confidence": 0.9999,
"action": {
"act_probability": 1.0
}
}
},
"usage": {
"input_tokens": 61,
"output_tokens": 0,
"state_tokens": 20,
"state_tokens_dropped": 0,
"truncated": false,
"truncated_questions": [],
"prompt_tokens": 61,
"completion_tokens": 0
},
"routing": {
"model": "multilingual",
"repo": "convaiinnovations/laya/multilingual",
"reason": "explicit model='multilingual'",
"detection": null,
"workflow": null
}
}
</syntaxhighlight>

== System One Model - Usage noul ==

Use noul to check a statement, for example to detect if customer explicitly requests a refund.

'''Input:'''

<syntaxhighlight lang="bash">
curl --fail-with-body -sS \
https://llm.stoney-cloud.com/v1/systemone \
-H "apikey: $STONEY_KEY" \
-H 'Content-Type: application/json' \
--data-binary '{
"model": "convaiinnovations/laya",
"state": {
"document": "Meine Rechnung wurde zweimal belastet. Bitte erstatten Sie mir die zweite Zahlung zurück."
},
"questions": {
"refund_requested": {
"type": "noul",
"instructions": "Does the customer explicitly request a refund?",
"criteria": {
"true": "The customer explicitly asks for money to be returned.",
"false": "The customer does not ask for money to be returned."
}
}
}
}' | jq .
</syntaxhighlight>

'''Output:'''

<syntaxhighlight lang="json">
{
"model": "convaiinnovations/laya",
"answers": {
"refund_requested": {
"type": "noul",
"noul": 0.9697,
"confidence": 0.9697,
"answer_confidence": 0.9697,
"action": {
"act_probability": 1.0
}
}
},
"usage": {
"input_tokens": 65,
"output_tokens": 0,
"state_tokens": 22,
"state_tokens_dropped": 0,
"truncated": false,
"truncated_questions": [],
"prompt_tokens": 65,
"completion_tokens": 0
},
"routing": {
"model": "multilingual",
"repo": "convaiinnovations/laya/multilingual",
"reason": "explicit model='multilingual'",
"detection": null,
"workflow": null
}
}
</syntaxhighlight>

== System One Model - Usage score ==

Use score to for example estimate the urgency of a support request. Arrange criteria from lowest to highest.

'''Input:'''

<syntaxhighlight lang="bash">
curl --fail-with-body -sS \
https://llm.stoney-cloud.com/v1/systemone \
-H "apikey: $STONEY_KEY" \
-H 'Content-Type: application/json' \
--data-binary '{
"model": "convaiinnovations/laya",
"state": {
"document": "Unser gesamtes Produktionssystem ist ausgefallen. Alle Kunden sind betroffen und können keine Bestellungen aufgeben. Wir benötigen sofort Hilfe."
},
"questions": {
"urgency": {
"type": "score",
"instructions": "How urgent is this support request?",
"criteria": [
"Not urgent: informational request with no operational impact",
"Low urgency: minor inconvenience with a workaround",
"Moderate urgency: partial disruption affecting some users",
"High urgency: major disruption requiring prompt attention",
"Critical urgency: complete production outage affecting all customers"
]
}
}
}' | jq .
</syntaxhighlight>

'''Output:'''

<syntaxhighlight lang="json">
{
"model": "convaiinnovations/laya",
"answers": {
"urgency": {
"type": "score",
"score": 3.987,
"legend": {
"0": "Not urgent: informational request with no operational impact",
"1": "Low urgency: minor inconvenience with a workaround",
"2": "Moderate urgency: partial disruption affecting some users",
"3": "High urgency: major disruption requiring prompt attention",
"4": "Critical urgency: complete production outage affecting all customers"
},
"probabilities": {
"0": 0.0003,
"1": 0.0016,
"2": 0.0012,
"3": 0.0044,
"4": 0.9924
},
"confidence": 0.9673,
"answer_confidence": 0.9924,
"action": {
"act_probability": 1.0
}
}
},
"usage": {
"input_tokens": 112,
"output_tokens": 0,
"state_tokens": 31,
"state_tokens_dropped": 0,
"truncated": false,
"truncated_questions": [],
"prompt_tokens": 112,
"completion_tokens": 0
},
"routing": {
"model": "multilingual",
"repo": "convaiinnovations/laya/multilingual",
"reason": "explicit model='multilingual'",
"detection": null,
"workflow": null
}
}
</syntaxhighlight>

[[Category:AI on demand]]

AI on demand: convaiinnovations/laya

2026-10-09T12:36:54Z

Sst-djo:

AI on demand: zai-org/GLM-5.3-Flash

2026-09-11T13:48:44Z

Sst-djo: /* Calling the model */

== Calling the model ==
<syntaxhighlight lang="bash">
# Set your personal key:
STONEY_KEY=sk-...

# Set the desired model:
MODEL=zai-org/GLM-5.3-Flash

# Set your prompt:
PROMPT='Hello.'

# Set maximum amount of tokens:
MAX_TOKENS=1000

curl https://llm.stoney-cloud.com/v1/chat/completions \
--silent --fail --show-error \
--header "Authorization: Bearer $STONEY_KEY" \
--header 'Content-Type: application/json' \
--data '{
"model": "'"$MODEL"'",
"messages": [
{"role": "user", "content": "'"$PROMPT"'"}
],
"max_tokens": '"$MAX_TOKENS"'
}' \
| jq
</syntaxhighlight>

Example output:
<syntaxhighlight lang="json">
{
"id": "chatcmpl-b823aeb93ca47bab",
"object": "chat.completion",
"created": 1789132434,
"model": "zai-org/GLM-5.3-Flash",
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"content": "Hello! I'm GLM, a large language model trained by Z.ai. Nice to meet you, and I hope you're having a great day.\n\nDo you have any questions or topics of interest I can help you with? Whether it's a knowledge inquiry, creative writing, or everyday conversation, I'm happy to provide assistance.",
"refusal": null,
"annotations": null,
"audio": null,
"function_call": null,
"reasoning": "Let me consider how to respond to this greeting appropriately.\n\nThe user has initiated the conversation with a simple \"Hello,\" which calls for a warm and professional response. Since this is the beginning of our interaction, it's important to create a welcoming atmosphere that encourages open communication. I should acknowledge their greeting in a friendly way and express my readiness to assist them with any questions or concerns they might have.\n\nI need to craft a response that not only reciprocates the greeting but also clearly communicates my purpose as an AI assistant designed to be helpful, informative, and supportive. This will help set the right tone for our conversation and make the user feel comfortable sharing what they need help with.\n\nThe response should be concise yet inviting, leaving room for the user to specify their needs or questions without feeling overwhelmed by too much information at once."
},
"logprobs": null,
"finish_reason": "stop",
"stop_reason": 154827,
"token_ids": null,
"routed_experts": null
}
],
"service_tier": null,
"system_fingerprint": "vllm-0.28.1rc1.dev400+g8b6de0eb9-tp2-dp4-6f5a0747",
"usage": {
"prompt_tokens": 14,
"total_tokens": 248,
"completion_tokens": 234,
"prompt_tokens_details": null,
"completion_tokens_details": {
"reasoning_tokens": 165
}
},
"prompt_logprobs": null,
"prompt_token_ids": null,
"prompt_text": null,
"kv_transfer_params": null,
"ec_transfer_params": null,
"metrics": null
}
</syntaxhighlight>

[[Category:AI on demand]]