outman/dify
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: optimize ollama model default parameters (#2894)

Browse Source
This commit is contained in:
takatost 2024-03-19 18:34:23 +08:00 committed by GitHub
parent 53d428907b
commit fbbba6db92
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
api/core/model_runtime/model_providers/ollama/llm/llm.py
View File

@ -449,7 +449,7 @@ class OllamaLargeLanguageModel(LargeLanguageModel):
help=I18nObject(en_US="The temperature of the model. " help=I18nObject(en_US="The temperature of the model. "
"Increasing the temperature will make the model answer " "Increasing the temperature will make the model answer "
"more creatively. (Default: 0.8)"), "more creatively. (Default: 0.8)"),
default=0.8, default=0.1,
min=0, min=0,
max=2 max=2
), ),
@ -472,7 +472,6 @@ class OllamaLargeLanguageModel(LargeLanguageModel):
help=I18nObject(en_US="Reduces the probability of generating nonsense. " help=I18nObject(en_US="Reduces the probability of generating nonsense. "
"A higher value (e.g. 100) will give more diverse answers, " "A higher value (e.g. 100) will give more diverse answers, "
"while a lower value (e.g. 10) will be more conservative. (Default: 40)"), "while a lower value (e.g. 10) will be more conservative. (Default: 40)"),
default=40,
min=1, min=1,
max=100 max=100
), ),
@ -483,7 +482,6 @@ class OllamaLargeLanguageModel(LargeLanguageModel):
help=I18nObject(en_US="Sets how strongly to penalize repetitions. " help=I18nObject(en_US="Sets how strongly to penalize repetitions. "
"A higher value (e.g., 1.5) will penalize repetitions more strongly, " "A higher value (e.g., 1.5) will penalize repetitions more strongly, "
"while a lower value (e.g., 0.9) will be more lenient. (Default: 1.1)"), "while a lower value (e.g., 0.9) will be more lenient. (Default: 1.1)"),
default=1.1,
min=-2, min=-2,
max=2 max=2
), ),
@ -494,7 +492,7 @@ class OllamaLargeLanguageModel(LargeLanguageModel):
type=ParameterType.INT, type=ParameterType.INT,
help=I18nObject(en_US="Maximum number of tokens to predict when generating text. " help=I18nObject(en_US="Maximum number of tokens to predict when generating text. "
"(Default: 128, -1 = infinite generation, -2 = fill context)"), "(Default: 128, -1 = infinite generation, -2 = fill context)"),
default=128, default=512 if int(credentials.get('max_tokens', 4096)) >= 768 else 128,
min=-2, min=-2,
max=int(credentials.get('max_tokens', 4096)), max=int(credentials.get('max_tokens', 4096)),
), ),
@ -504,7 +502,6 @@ class OllamaLargeLanguageModel(LargeLanguageModel):
type=ParameterType.INT, type=ParameterType.INT,
help=I18nObject(en_US="Enable Mirostat sampling for controlling perplexity. " help=I18nObject(en_US="Enable Mirostat sampling for controlling perplexity. "
"(default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)"), "(default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)"),
default=0,
min=0, min=0,
max=2 max=2
), ),
@ -516,7 +513,6 @@ class OllamaLargeLanguageModel(LargeLanguageModel):
"the generated text. A lower learning rate will result in slower adjustments, " "the generated text. A lower learning rate will result in slower adjustments, "
"while a higher learning rate will make the algorithm more responsive. " "while a higher learning rate will make the algorithm more responsive. "
"(Default: 0.1)"), "(Default: 0.1)"),
default=0.1,
precision=1 precision=1
), ),
ParameterRule( ParameterRule(
@ -525,7 +521,6 @@ class OllamaLargeLanguageModel(LargeLanguageModel):
type=ParameterType.FLOAT, type=ParameterType.FLOAT,
help=I18nObject(en_US="Controls the balance between coherence and diversity of the output. " help=I18nObject(en_US="Controls the balance between coherence and diversity of the output. "
"A lower value will result in more focused and coherent text. (Default: 5.0)"), "A lower value will result in more focused and coherent text. (Default: 5.0)"),
default=5.0,
precision=1 precision=1
), ),
ParameterRule( ParameterRule(
@ -543,7 +538,6 @@ class OllamaLargeLanguageModel(LargeLanguageModel):
type=ParameterType.INT, type=ParameterType.INT,
help=I18nObject(en_US="The number of layers to send to the GPU(s). " help=I18nObject(en_US="The number of layers to send to the GPU(s). "
"On macOS it defaults to 1 to enable metal support, 0 to disable."), "On macOS it defaults to 1 to enable metal support, 0 to disable."),
default=1,
min=0, min=0,
max=1 max=1
), ),
@ -563,7 +557,6 @@ class OllamaLargeLanguageModel(LargeLanguageModel):
type=ParameterType.INT, type=ParameterType.INT,
help=I18nObject(en_US="Sets how far back for the model to look back to prevent repetition. " help=I18nObject(en_US="Sets how far back for the model to look back to prevent repetition. "
"(Default: 64, 0 = disabled, -1 = num_ctx)"), "(Default: 64, 0 = disabled, -1 = num_ctx)"),
default=64,
min=-1 min=-1
), ),
ParameterRule( ParameterRule(
@ -573,7 +566,6 @@ class OllamaLargeLanguageModel(LargeLanguageModel):
help=I18nObject(en_US="Tail free sampling is used to reduce the impact of less probable tokens " help=I18nObject(en_US="Tail free sampling is used to reduce the impact of less probable tokens "
"from the output. A higher value (e.g., 2.0) will reduce the impact more, " "from the output. A higher value (e.g., 2.0) will reduce the impact more, "
"while a value of 1.0 disables this setting. (default: 1)"), "while a value of 1.0 disables this setting. (default: 1)"),
default=1,
precision=1 precision=1
), ),
ParameterRule( ParameterRule(
@ -583,7 +575,6 @@ class OllamaLargeLanguageModel(LargeLanguageModel):
help=I18nObject(en_US="Sets the random number seed to use for generation. Setting this to " help=I18nObject(en_US="Sets the random number seed to use for generation. Setting this to "
"a specific number will make the model generate the same text for " "a specific number will make the model generate the same text for "
"the same prompt. (Default: 0)"), "the same prompt. (Default: 0)"),
default=0
), ),
ParameterRule( ParameterRule(
name='format', name='format',