Package net.thevpc.naru.ext.models.ollama

Class NaruModelProtocolOllamaNative

java.lang.Object

net.thevpc.naru.ext.models.NaruModelProtocolBase

net.thevpc.naru.ext.models.ollama.NaruModelProtocolOllamaNative

All Implemented Interfaces:

NaruModelProtocol

public class NaruModelProtocolOllamaNative
extends NaruModelProtocolBase

Nested Class Summary

Nested classes/interfaces inherited from class net.thevpc.naru.ext.models.NaruModelProtocolBase

NaruModelProtocolBase.NonRetryableWebException

Field Summary

Fields

Modifier and Type	Field	Description
static final String	THINK_ENV	Request env key carrying the think flag sent to Ollama.

Fields inherited from class net.thevpc.naru.ext.models.NaruModelProtocolBase

capabilities, chatPath, configPrefix, model, nElementReader, provider, serializer

Constructor Summary

Constructors

Constructor	Description
NaruModelProtocolOllamaNative(NaruModelProvider provider, NaruModelConfig model, String configPrefix, NaruModelCapabilities capabilities)

Method Summary

Modifier and Type	Method	Description
NaruResponse	chatStream(NaruModelRequest request, NaruTask task, NaruStreamHandler handler)
protected NaruModelRequest	preprocessRequest(NaruModelRequest mrequest, NaruTask task)	Defaults think from the model's declared capability, and lets an explicit request or ollama.think setting win.

Methods inherited from class net.thevpc.naru.ext.models.NaruModelProtocolBase

apiKey, apiKeyConfigKey, chat, chatPath, classifyStatus, commitCacheState, connectTimeout, getCapabilities, maxRetries, onResponseReceived, parseResponse, parseXmlLikeToolCall, prepareRequest, provider, providerName, readTimeout, retryPeriod, serialize, serialize, streamChat, url

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Details

THINK_ENV

public static final String THINK_ENV

Request env key carrying the think flag sent to Ollama.

Reading the request's think field is the only way to get reasoning out of a reasoning model: Ollama returns no message.thinking at all unless the request asks for it. Sending it unasked, on the other hand, is worse than useless -- Ollama rejects a model that does not support thinking, so a default of true would break every non-reasoning model.

See Also:

• Constant Field Values

Constructor Details

NaruModelProtocolOllamaNative

public NaruModelProtocolOllamaNative(NaruModelProvider provider,
 NaruModelConfig model,
 String configPrefix,
 NaruModelCapabilities capabilities)

Method Details

preprocessRequest

protected NaruModelRequest preprocessRequest(NaruModelRequest mrequest,
 NaruTask task)

Defaults think from the model's declared capability, and lets an explicit request or ollama.think setting win.

The capability is the answer, not a guess: Ollama reports a reasoning model in /api/show, and that is the only thing that distinguishes a model which can be asked to think from one that will refuse the request.

Overrides:

preprocessRequest in class NaruModelProtocolBase

chatStream

public NaruResponse chatStream(NaruModelRequest request,
 NaruTask task,
 NaruStreamHandler handler)