Model Parameters

ModelFileParameters is an enum of generation parameters that Ollama exposes via its Modelfile. These parameters are serialized under the options key in the toApi() output (which is what Ollama's /v1 endpoint expects).

use ByJG\LlmApiObjects\Enum\ModelFileParameters;

$chat->setParameter(ModelFileParameters::temperature, 0.7);

setParameter() validates the value before accepting it. Passing an invalid value throws \InvalidArgumentException.

warning

These parameters are Ollama-specific. They are sent under the options key, which is not a standard OpenAI API field. If you use this with the OpenAI API directly, the extra options key will be ignored or may cause an error.

Parameter Reference

Enum case	API key	Type	Valid values	Default	Description
mirostat	mirostat	int	0, 1, 2	0	Enable Mirostat sampling. 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0
mirostatEta	mirostat_eta	float	≥ 0	0.1	Mirostat learning rate — how quickly the algorithm responds to feedback
mirostatTau	mirostat_tau	float	≥ 0	5.0	Controls the balance between coherence and diversity
numCtx	num_ctx	int	> 0	2048	Context window size in tokens
repeatLastN	repeat_last_n	int	≥ -1	64	How far back to look to prevent repetition. 0 = disabled, -1 = num_ctx
repeatPenalty	repeat_penalty	float	> 0	1.1	Strength of repetition penalty. Higher = penalizes more
temperature	temperature	float	> 0	0.8	Creativity of responses. Higher = more creative
seed	seed	int	≥ 0	0	Random seed for reproducible generation
stop	stop	string	any string	—	Stop sequence. Generation halts when this pattern is encountered
tfsZ	tfs_z	float	≥ 1	1	Tail free sampling. Higher = reduces impact of low-probability tokens. 1.0 = disabled
numPredict	num_predict	int	≥ -1	-1	Max tokens to generate. -1 = infinite
topK	top_k	int	> 0	40	Limits token selection to the top K tokens. Higher = more diverse
topP	top_p	float	> 0, ≤ 1	0.9	Nucleus sampling threshold. Works with top_k
minP	min_p	float	> 0, ≤ 1	—	Minimum token probability relative to the most likely token

Example

use ByJG\LlmApiObjects\Enum\ModelFileParameters;
use ByJG\LlmApiObjects\Model\Chat;
use ByJG\LlmApiObjects\Model\Message;
use ByJG\LlmApiObjects\Enum\Role;

$chat = new Chat(
    model: 'llama3',
    messages: [
        new Message(role: Role::user, message: 'Write a short poem about the sea.'),
    ],
);

$chat->setParameter(ModelFileParameters::temperature, 0.9);
$chat->setParameter(ModelFileParameters::numCtx, 4096);
$chat->setParameter(ModelFileParameters::seed, 42);
$chat->setParameter(ModelFileParameters::numPredict, 200);

print_r($chat->toApi());
// 'options' => ['temperature' => 0.9, 'num_ctx' => 4096, 'seed' => 42, 'num_predict' => 200]

Validation

Each parameter has strict validation. Invalid values throw \InvalidArgumentException:

// OK
$chat->setParameter(ModelFileParameters::mirostat, 1);

// Throws InvalidArgumentException — mirostat only accepts 0, 1, or 2
$chat->setParameter(ModelFileParameters::mirostat, 3);

// Throws InvalidArgumentException — temperature must be > 0
$chat->setParameter(ModelFileParameters::temperature, -0.5);