- Training the Tokenizer - 03-06-2025
- Commented in [crewAIInc/crewAI] on 2025-06-03.
AI Summary: @Vidit-Ostwal has suggested that after attempting to use a single task bound with two tools, the agent is behaving unpredictably. They have expressed a desire to review logs to understand the situation better, asking for the relevant information to be shared. This highlights the need for troubleshooting to identify the underlying issues affecting the agent's performance when integrating multiple tools within a single task setup.
- Commented in [crewAIInc/crewAI] on 2025-06-03.
AI Summary: @Vidit-Ostwal has suggested that after binding an agent with only two tools, it functions as expected. However, in practical scenarios, a minimum of ten tools per agent may be required for scalability. The current setup may hinder optimal tool invocation and efficiency. Further exploration is necessary to determine if crewai agents are designed for multiple tool calls in a single invocation. Assistance in optimizing or addressing this limitation would be appreciated, and the necessity for a test task using multiple tools has been recommended.
- Commented in [crewAIInc/crewAI] on 2025-06-03.
AI Summary: @Vidit-Ostwal has suggested that binding the agent with only two tools has led to it functioning as expected. However, in practical applications, a minimum of ten tools per agent may be necessary. To optimize this setup, the suggestion seeks input on potential solutions or strategies. It emphasizes the importance of having an adequate number of tools in order for workflows to remain scalable and efficient in real-world scenarios. Additional details can be provided if needed.
- Commented in [crewAIInc/crewAI] on 2025-06-03.
AI Summary: @Vidit-Ostwal has suggested testing with a reduced number of tools to determine if the same behavior is observed. This approach may help isolate the issue and provide clearer insights into the problem being investigated. Simplifying the number of variables in the testing environment could lead to more accurate results and understanding. It is a practical step aimed at troubleshooting the current situation effectively and could facilitate resolving any discrepancies that have been noticed.
- Commented in [crewAIInc/crewAI] on 2025-06-03.
AI Summary: @Vidit-Ostwal has suggested clarification regarding the roles of knowledge_source and embedder in the context of both agent and crew levels. It is important to understand how these components interact and function at the different levels to ensure effective implementation. Detailed definitions and examples of how each element operates within the respective levels would enhance comprehension and application. Overall, a clearer distinction and explanation of these roles would benefit the collaborative effort.
No issues raised recently.
- Opened a PR in [ariG23498/gemma3-object-detection]: Added get_tokenizer_with_new_tokens_func (2025-06-02).
AI Summary: @Vidit-Ostwal has suggested implementing a dynamic function that accepts configuration directly from a configuration file. This function would enhance the tokenizer by incorporating tokens based on the provided configuration. The goal is to streamline the process of adding tokens, making it more efficient for users to customize the tokenizer according to specific needs. This approach aims to improve flexibility and usability by allowing direct adjustments through the configuration file, ultimately enhancing the overall functionality of the tokenizer.
- Starred tokenbender/avataRL on 2025-06-01.
No repositories forked recently.