diff --git a/api/core/model_runtime/model_providers/gitee_ai/llm/InternVL2-8B.yaml b/api/core/model_runtime/model_providers/gitee_ai/llm/InternVL2-8B.yaml new file mode 100644 index 0000000000..d288c3dd39 --- /dev/null +++ b/api/core/model_runtime/model_providers/gitee_ai/llm/InternVL2-8B.yaml @@ -0,0 +1,93 @@ +model: InternVL2-8B +label: + en_US: InternVL2-8B +model_type: llm +features: + - vision + - agent-thought +model_properties: + mode: chat + context_size: 32000 +parameter_rules: + - name: max_tokens + use_template: max_tokens + label: + en_US: "Max Tokens" + zh_Hans: "最大Token数" + type: int + default: 512 + min: 1 + required: true + help: + en_US: "The maximum number of tokens that can be generated by the model varies depending on the model." + zh_Hans: "模型可生成的最大 token 个数，不同模型上限不同。" + + - name: temperature + use_template: temperature + label: + en_US: "Temperature" + zh_Hans: "采样温度" + type: float + default: 0.7 + min: 0.0 + max: 1.0 + precision: 1 + required: true + help: + en_US: "The randomness of the sampling temperature control output. The temperature value is within the range of [0.0, 1.0]. The higher the value, the more random and creative the output; the lower the value, the more stable it is. It is recommended to adjust either top_p or temperature parameters according to your needs to avoid adjusting both at the same time." + zh_Hans: "采样温度控制输出的随机性。温度值在 [0.0, 1.0] 范围内，值越高，输出越随机和创造性；值越低，输出越稳定。建议根据需求调整 top_p 或 temperature 参数，避免同时调整两者。" + + - name: top_p + use_template: top_p + label: + en_US: "Top P" + zh_Hans: "Top P" + type: float + default: 0.7 + min: 0.0 + max: 1.0 + precision: 1 + required: true + help: + en_US: "The value range of the sampling method is [0.0, 1.0]. The top_p value determines that the model selects tokens from the top p% of candidate words with the highest probability; when top_p is 0, this parameter is invalid. It is recommended to adjust either top_p or temperature parameters according to your needs to avoid adjusting both at the same time." + zh_Hans: "采样方法的取值范围为 [0.0,1.0]。top_p 值确定模型从概率最高的前p%的候选词中选取 tokens；当 top_p 为 0 时，此参数无效。建议根据需求调整 top_p 或 temperature 参数，避免同时调整两者。" + + - name: top_k + use_template: top_k + label: + en_US: "Top K" + zh_Hans: "Top K" + type: int + default: 50 + min: 0 + max: 100 + required: true + help: + en_US: "The value range is [0,100], which limits the model to only select from the top k words with the highest probability when choosing the next word at each step. The larger the value, the more diverse text generation will be." + zh_Hans: "取值范围为 [0,100]，限制模型在每一步选择下一个词时，只从概率最高的前 k 个词中选取。数值越大，文本生成越多样。" + + - name: frequency_penalty + use_template: frequency_penalty + label: + en_US: "Frequency Penalty" + zh_Hans: "频率惩罚" + type: float + default: 0 + min: -1.0 + max: 1.0 + precision: 1 + required: false + help: + en_US: "Used to adjust the frequency of repeated content in automatically generated text. Positive numbers reduce repetition, while negative numbers increase repetition. After setting this parameter, if a word has already appeared in the text, the model will decrease the probability of choosing that word for subsequent generation." + zh_Hans: "用于调整自动生成文本中重复内容的频率。正数减少重复，负数增加重复。设置此参数后，如果一个词在文本中已经出现过，模型在后续生成中选择该词的概率会降低。" + + - name: user + use_template: text + label: + en_US: "User" + zh_Hans: "用户" + type: string + required: false + help: + en_US: "Used to track and differentiate conversation requests from different users." + zh_Hans: "用于追踪和区分不同用户的对话请求。" diff --git a/api/core/model_runtime/model_providers/gitee_ai/llm/InternVL2.5-26B.yaml b/api/core/model_runtime/model_providers/gitee_ai/llm/InternVL2.5-26B.yaml new file mode 100644 index 0000000000..b2dee88c02 --- /dev/null +++ b/api/core/model_runtime/model_providers/gitee_ai/llm/InternVL2.5-26B.yaml @@ -0,0 +1,93 @@ +model: InternVL2.5-26B +label: + en_US: InternVL2.5-26B +model_type: llm +features: + - vision + - agent-thought +model_properties: + mode: chat + context_size: 32000 +parameter_rules: + - name: max_tokens + use_template: max_tokens + label: + en_US: "Max Tokens" + zh_Hans: "最大Token数" + type: int + default: 512 + min: 1 + required: true + help: + en_US: "The maximum number of tokens that can be generated by the model varies depending on the model." + zh_Hans: "模型可生成的最大 token 个数，不同模型上限不同。" + + - name: temperature + use_template: temperature + label: + en_US: "Temperature" + zh_Hans: "采样温度" + type: float + default: 0.7 + min: 0.0 + max: 1.0 + precision: 1 + required: true + help: + en_US: "The randomness of the sampling temperature control output. The temperature value is within the range of [0.0, 1.0]. The higher the value, the more random and creative the output; the lower the value, the more stable it is. It is recommended to adjust either top_p or temperature parameters according to your needs to avoid adjusting both at the same time." + zh_Hans: "采样温度控制输出的随机性。温度值在 [0.0, 1.0] 范围内，值越高，输出越随机和创造性；值越低，输出越稳定。建议根据需求调整 top_p 或 temperature 参数，避免同时调整两者。" + + - name: top_p + use_template: top_p + label: + en_US: "Top P" + zh_Hans: "Top P" + type: float + default: 0.7 + min: 0.0 + max: 1.0 + precision: 1 + required: true + help: + en_US: "The value range of the sampling method is [0.0, 1.0]. The top_p value determines that the model selects tokens from the top p% of candidate words with the highest probability; when top_p is 0, this parameter is invalid. It is recommended to adjust either top_p or temperature parameters according to your needs to avoid adjusting both at the same time." + zh_Hans: "采样方法的取值范围为 [0.0,1.0]。top_p 值确定模型从概率最高的前p%的候选词中选取 tokens；当 top_p 为 0 时，此参数无效。建议根据需求调整 top_p 或 temperature 参数，避免同时调整两者。" + + - name: top_k + use_template: top_k + label: + en_US: "Top K" + zh_Hans: "Top K" + type: int + default: 50 + min: 0 + max: 100 + required: true + help: + en_US: "The value range is [0,100], which limits the model to only select from the top k words with the highest probability when choosing the next word at each step. The larger the value, the more diverse text generation will be." + zh_Hans: "取值范围为 [0,100]，限制模型在每一步选择下一个词时，只从概率最高的前 k 个词中选取。数值越大，文本生成越多样。" + + - name: frequency_penalty + use_template: frequency_penalty + label: + en_US: "Frequency Penalty" + zh_Hans: "频率惩罚" + type: float + default: 0 + min: -1.0 + max: 1.0 + precision: 1 + required: false + help: + en_US: "Used to adjust the frequency of repeated content in automatically generated text. Positive numbers reduce repetition, while negative numbers increase repetition. After setting this parameter, if a word has already appeared in the text, the model will decrease the probability of choosing that word for subsequent generation." + zh_Hans: "用于调整自动生成文本中重复内容的频率。正数减少重复，负数增加重复。设置此参数后，如果一个词在文本中已经出现过，模型在后续生成中选择该词的概率会降低。" + + - name: user + use_template: text + label: + en_US: "User" + zh_Hans: "用户" + type: string + required: false + help: + en_US: "Used to track and differentiate conversation requests from different users." + zh_Hans: "用于追踪和区分不同用户的对话请求。" diff --git a/api/core/model_runtime/model_providers/gitee_ai/llm/_position.yaml b/api/core/model_runtime/model_providers/gitee_ai/llm/_position.yaml index 13c31ad02b..c942cda3b2 100644 --- a/api/core/model_runtime/model_providers/gitee_ai/llm/_position.yaml +++ b/api/core/model_runtime/model_providers/gitee_ai/llm/_position.yaml @@ -6,3 +6,5 @@ - deepseek-coder-33B-instruct-chat - deepseek-coder-33B-instruct-completions - codegeex4-all-9b +- InternVL2.5-26B +- InternVL2-8B diff --git a/api/core/model_runtime/model_providers/gitee_ai/llm/llm.py b/api/core/model_runtime/model_providers/gitee_ai/llm/llm.py index 0c253a4a0a..68aaad2e3f 100644 --- a/api/core/model_runtime/model_providers/gitee_ai/llm/llm.py +++ b/api/core/model_runtime/model_providers/gitee_ai/llm/llm.py @@ -29,18 +29,26 @@ class GiteeAILargeLanguageModel(OAIAPICompatLargeLanguageModel): user: Optional[str] = None, ) -> Union[LLMResult, Generator]: self._add_custom_parameters(credentials, model, model_parameters) - return super()._invoke(model, credentials, prompt_messages, model_parameters, tools, stop, stream, user) + return super()._invoke( + GiteeAILargeLanguageModel.MODEL_TO_IDENTITY.get(model, model), + credentials, + prompt_messages, + model_parameters, + tools, + stop, + stream, + user, + ) def validate_credentials(self, model: str, credentials: dict) -> None: - self._add_custom_parameters(credentials, None) - super().validate_credentials(model, credentials) + self._add_custom_parameters(credentials, model, None) + super().validate_credentials(GiteeAILargeLanguageModel.MODEL_TO_IDENTITY.get(model, model), credentials) - def _add_custom_parameters(self, credentials: dict, model: Optional[str]) -> None: + def _add_custom_parameters(self, credentials: dict, model: Optional[str], model_parameters: dict) -> None: if model is None: model = "Qwen2-72B-Instruct" - model_identity = GiteeAILargeLanguageModel.MODEL_TO_IDENTITY.get(model, model) - credentials["endpoint_url"] = f"https://ai.gitee.com/api/serverless/{model_identity}/" + credentials["endpoint_url"] = "https://ai.gitee.com/v1" if model.endswith("completions"): credentials["mode"] = LLMMode.COMPLETION.value else: