ibm-watsonx-orchestrate-evaluation-framework 1.0.7__py3-none-any.whl → 1.0.9__py3-none-any.whl

{ibm_watsonx_orchestrate_evaluation_framework-1.0.7.dist-info → ibm_watsonx_orchestrate_evaluation_framework-1.0.9.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ibm-watsonx-orchestrate-evaluation-framework
-Version: 1.0.7
+Version: 1.0.9
 Summary: The WxO evaluation framework
 Author-email: Haode Qi <Haode.Qi@ibm.com>
 License: MIT
@@ -10,7 +10,7 @@ Requires-Dist: rich~=13.9.4
 Requires-Dist: pydantic<3.0.0,>=2.10.3
 Requires-Dist: pyyaml~=6.0.2
 Requires-Dist: jinja2~=3.1.5
-Requires-Dist: python-dotenv~=1.0.1
+Requires-Dist: python-dotenv
 Requires-Dist: dataclasses-json~=0.6.7
 Requires-Dist: jsonargparse~=4.37.0
 Provides-Extra: dev
@@ -32,29 +32,46 @@ Requires-Dist: ipywidgets~=8.1.6; extra == "rag-eval"
 Requires-Dist: jupyter_contrib_nbextensions; extra == "rag-eval"
 Requires-Dist: jupyter~=1.1.1; extra == "rag-eval"
 
-# WXO-agent evaluation framework
-
-- This framework is designed to test a tool-calling agent's ability to make real API calls against a `wxo-dev` testing tenant on your local wxo-lite server instance.
-
+# WXO Agent Evaluation Framework
+
+## Table of Contents
+- [Overview](#overview)
+- [ADK Setup Guide](#adk-setup-guide)
+- [Setup](#setup-for-evaluation-framework)
+- [Quick Experiment](#quick-experiment-against-the-default-wxo-dev-env)
+- [Run Against a Deployed Local Env](#run-against-a-deployed-local-env)
+- [Run Against a SaaS Tenant](#run-against-a-saas-tenant)
+- [Analyze Results](#analyze-results)
+- [Record Chat Sessions](#record-chat-sessions)
+- [Batch Test Case Generation](#batch-test-case-generation)
+- [Using Model Proxy Provider](#using-model-proxy-provider)
+- [Using Ollama](#using-ollama)
+- [Workflow Diagram](#workflow-diagram)
+- [Results](#results)
+- [Metrics](#metrics)
+
+## Overview
+
+- This framework is designed to test a tool-calling agent's ability to make real API calls against a `wxo-dev` testing tenant on your local wxo-lite server instance. To run evaluation against a remote tenant on SaaS, follow [Run Against a SaaS Tenant](#run-against-a-saas-tenant).
 - As an LLM-as-agent evaluation framework, we aim to test the agent's ability to do the following:
     - We use a ground truth to evaluate our conversation against after inference. The process of inference is manifested through a user-LLM and agent simulation. Please set `enable_verbose_logging: True` in your configuration.
-    - Make real API calls correctly and efficiently. We provide metrics which measure the number of bad tool calls made by the agent, normalized against the number of ground truth calls made. 
-
+    - Make real API calls correctly and efficiently. We provide metrics such as tool call precision, recall, and routing accuracy to measure the agent's performance against the ground truth.
 - The `benchmarks/` folder contains test-cases for the different agents we have evaluated so far. They are segmented by release versions of the `wxo-domains` repository.
 - The agent calls the `runs/` endpoint of the wxo-lite server instance, and the actual tool code is executed on the server side. The server database is not visible to our framework. 
 
-## prerequisite
-Follow the [SDK setup guide](https://github.ibm.com/WatsonOrchestrate/wxo-clients/tree/main) to install the SDK.    
-The current framework is compatible with ADK version >= 1.20, <= 1.6.0
+## ADK Setup Guide
+Follow the [ADK setup guide](https://github.ibm.com/WatsonOrchestrate/wxo-clients/tree/main) to install the ADK.    
 
-## setup for evaluation framework
+The current framework is compatible with ADK version >= 1.20, <= 1.7.0
+
+## Setup for Evaluation Framework
 Run the following command to install evaluation framework in the same env:
 ``` 
 pip install -e .
 ```
 
-## contribution guide
-### secret resolution
+## Contribution Guide
+### Secret Resolution
 install detect secret utilities:
 ```
 pip install --upgrade git+https://github.com/ibm/detect-secrets.git@master#egg=detect-secrets
@@ -65,7 +82,7 @@ detect-secrets scan --exclude-files "benchmark|results" --update .secrets.baseli
 ```
 
 
-## quick experiment against the default wxo-dev env
+## Quick Experiment Against the Default wxo-dev Env
 ```bash
 orchestrate server start
 export WATSONX_SPACE_ID=""
@@ -88,8 +105,7 @@ Note:
 1. This approach uses the default `wxo-dev` tenant already available in your orchestrate env if you have used wxo-lite before.
 2. ADK also reads the env environments variable. If you have an env conflict, start the wxo-lite server before exporting the envs.
 
-
-## run against a deployed local env
+## Run Against a Deployed Local Env
 
 1. start the orchestrated server: `orchestrate server start`
 2. create a simple test case like the following save in a folder like `benchmarks/TEST_CASE_NAME`:
@@ -116,7 +132,6 @@ Note:
 - The target agent name can be found `orchestrate agents list`
 - the example shown only evaluate the final response for the agent. For more sophisticated examples, follow `benchmarks/hr_sample/data_simple.json` or `benchmarks/hr_sample/data_complex.json`.
 
-
 3. create a test config yaml like the following:
 ```YAML
 test_paths:
@@ -129,7 +144,6 @@ auth_config:
 output_dir: "results/TEST_CASE_NAME/MODEL_NAME"
 ```
 
-
 NOTE: run `orchestrate env list` to find the name of the active tenant. for default `local` tenant, the name should be `wxo-dev`
 
 4. Run the test:
@@ -141,27 +155,17 @@ python -m wxo_agentic_evaluation.main --config benchmarks/hr_sample/config.yaml
 
 NOTE: if your run fails for any reason and doesn't cover all the test cases, you can re-run the main script with `--skip_available_results=True` to skip the test cases that are already completed.
 
-## analyze error
-```bash
-python -m wxo_agentic_evaluation.analyze_run --input_data results/hr_sample/llama_3_2_90b/messages/data_simple.messages.json --ground_truth benchmarks/hr_sample/data_simple.json --enable_verbose_logging False
-```
-You can also run the analyze script on a batch of test cases in a folder
-```bash
-python -m wxo_agentic_evaluation.analyze_run --input_data results/hr_sample/llama_3_2_90b/messages/ --ground_truth benchmarks/hr_sample/ --enable_verbose_logging False --enable_verbose_logging False
-```
-
+## Run Against a SaaS Tenant 
 
-## Run Against a SaaS Tenant (Orchestrate SDK ≥ 1.2)
+Orchestrate ADK ≥ 1.2 is required for this section.
 
 This section describes how to run benchmark tests using a **SaaS-based Orchestrate tenant**. The rest of the setup (test case creation, config structure, etc.) is similar to the [local setup](#run-against-a-deployed-local-env) and can be referred to as needed.
 
 ### Prerequisites
 
-- **Orchestrate SDK version ≥ 1.2** is required.
+- **Orchestrate ADK version ≥ 1.2** is required.
 - Access to the **production SaaS Orchestrate instance** or **staging SaaS Orchestrate instance**.
 
----
-
 ### 1. Get Authentication Details
 
 1. Visit the Orchestrate UI [ Prod /staging]: 
@@ -178,8 +182,6 @@ For other locations, please use the designated url for your data center.
 4. For more detailed instructions, refer to this guide:  
    https://developer.ibm.com/apis/catalog/watsonorchestrate--custom-assistants/Getting+the+API+endpoint
 
----
-
 ### 2. Add the SaaS Tenant
 
 Run the following command:
@@ -209,16 +211,12 @@ orchestrate env add -n saas \
 
 > When prompted, paste the API key generated above.
 
----
-
-### 3. Set the IAM API Key Environment Variable
+### 3. Set `WO_API_KEY` Environment Variable
 
 ```bash
-export WATSONX_IAM_SAAS_APIKEY=[your_generated_api_key]
+export WO_API_KEY=[your_generated_api_key]
 ```
 
----
-
 ### 4. Update Your Test Config YAML
 
 Make sure your YAML config includes the correct SaaS tenant name:
@@ -234,17 +232,55 @@ auth_config:
 output_dir: "results/TEST_CASE_NAME/MODEL_NAME"
 ```
 - Use staging url if using the staging set-up.
----
-
 ###  5. Run the Simulation in SaaS Mode
 
 ```bash
 python -m wxo_agentic_evaluation.main --config benchmarks/hr_sample/config.yaml 
 ```
 
----
+## Analyze Results
+
+The `analyze_run.py` script summarizes agent evaluation results, showing successes, failures, and reasons for errors to help improve agent performance. After running an evaluation, analyze the results with:
+
+```bash
+python -m wxo_agentic_evaluation.analyze_run --data_path path/to/results
+```
+
+Additionally, the script comes with a feature to analyze the quality of tool descriptions for failing tools where the reason for failure is incorrect parameter usage by the agent.
 
-### Batch Test case Generation
+In order to analyze the description(s) of your failing tools, consider passing the optional flag `--tool_definition_path` like so:
+
+```bash
+python -m wxo_agentic_evaluation.analyze_run --data_path path/to/results --tool_definition_path path/to/.py/source/file/containing/tool/definitions
+```
+
+**Note:** If the flag `tool_definition_path` is not provided, description quality analysis is simply skipped.
+
+## Record Chat Sessions
+
+The `record_chat.py` script lets you capture your chat sessions in the chat UI and automatically generate ground truth data for evaluating your agents. This is valuable for benchmarking and experimenting with agent behavior under different configurations.
+
+Start the chat interface:
+
+```bash
+orchestrate chat start
+```
+
+Then open your browser to [http://localhost:3000/chat-lite](http://localhost:3000/chat-lite) and select the agent you wish to interact with.
+
+To begin recording, run:
+
+```bash
+python -m wxo_agentic_evaluation.record_chat --output_dir dir/to/save/recordings
+```
+
+While this process is running, for every chat session, annotated ground truth data is generated in your output directory: `<THREAD_ID>_annotated_data.json`
+
+Review the generated annotated data for accuracy before using it for evaluation.
+
+Press `Ctrl+C` in the terminal to stop recording when your session is complete.
+
+## Batch Test Case Generation
 
 For full instructions on setting up tools, writing stories, configuring the pipeline, and generating batch test cases, see the [Batch Test case Generation Guide](./benchmarks/batch_sample/README.MD).
 
@@ -311,9 +347,9 @@ To use model from Ollama (local LLM deployment), follow these steps:
    python -m wxo_agentic_evaluation.main --config path/to/your/config.yaml
    ```
 
-## Workflow diagram
+## Workflow Diagram
 
-To help better understand the workflow, this is a diagram of how this repo works together with wxO lite python SDK and a wxO runtime.
+To help better understand the workflow, this is a diagram of how this repo works together with wxO ADK and a wxO runtime.
 
 ![Alt text](./doc/assets/workflow.png "Workflow")
 
@@ -322,70 +358,28 @@ Inputs:
 - a json file containing test cases, see [example 1](benchmarks/hr_sample/data_complex.json) or [example 2](benchmarks/hr_sample/data_simple.json) as a reference
 - optionally, a `tools.py` file for tools definition and one or more agent definitions e.g. `benchmarks/hr_sample/hr_agent.json`. Alternatively, these files are not needed if you have a tenant already set up with such tools and agents
 
-Steps:
-1. (optional) this repo will call wxO Lite python SDK (which calls several endpoints exposed by the wxO runtime) to set up the environment needed for the evaluation run, tools and agents will be imported
-2. Create test cases by following the sample instructions at [benchmarks/sap_successfactors_sample/annotation/README.md](benchmarks/sap_successfactors_sample/annotation/README.md)
-3. Start the evaluation run by calling the `wxo_agentic_evaluation.main` script of this repo, which will invoke the `/runs` endpoint of the wxO runtime to simulate conversations with the agent
-4. Reports and metrics will be generated by this repo
-5. (optional) this repo will call wxO Lite python SDK (which calls several endpoints exposed by the wxO runtime) to clean up the environment, to avoid the tools and agents affecting subsequent runs with the same tenant
-6. (optional) You can generate further error analysis by using the `wxo_agentic_evaluation.analyze_run` script from this repo
-
-
-## results
-### workday
-
-| Model                         | User Setting | Total Step  | Agent Step | Journey Success  | Wrong Function Calls  | Bad Calls | Wrong Parameters  | Wrong Routing Calls  | Text Match | Test Cases | WXO Avg. Response Time  |
-|-------------------------------|------------- |-------------|------------|------------------|-----------------------|-----------|-------------------|----------------------|------------|------------|-------------------------|
-| llama-3-2-90b-vision-instruct | normal       | 8.13        | 4.21       | 0.87             | 0.01                  | 0.0       | 0.20              | 0.00                 | 0.95       | 38         | 15.09                   |
-| llama-3-2-90b-vision-instruct | verbose      | 11.76       | 6.11       | 0.79             | 0.02                  | 0.0       | 0.19              | 0.00                 | 0.86       | 38         | 14.32                   |
-| llama-3-405b-instruct         | normal       | 9.66        | 5.03       | 0.82             | 0.02                  | 0.0       | 0.47              | 0.04                 | 0.89       | 38         | 13.36                   |
-| llama-3-405b-instruct         | verbose      | 11.76       | 6.11       | 0.84             | 0.05                  | 0.0       | 0.70              | 0.04                 | 0.92       | 38         | 12.21                   |
-
-You can find the detailed results under [results/workday](results/workday)
-
-### sap successfactor (rel-1.7)
-
-| Model                         | User Setting | Total Step  | Agent Step | Journey Success  | Wrong Function Calls  | Bad Calls | Wrong Parameters  | Wrong Routing Calls  | Text Match | Test Cases | WXO Avg. Response Time  |
-|------------------------------ |--------------|-------------|------------|------------------|-----------------------|-----------|-------------------|----------------------|------------|------------|-------------------------|
-| llama-3-2-90b-vision-instruct | normal       | 10.32       | 5.84       | 0.73             | 0.04                  | 0.0       | 0.06              | 0.08                 | 0.84       | 38         | -                       |
-| llama-3-2-90b-vision-instruct | verbose      | 11.19       | 6.35       | 0.68             | 0.04                  | 0.0       | 0.08              | 0.16                 | 0.81       | 38         | -                       |
-| llama-3-405b-instruct         | normal       | 11.41       | 6.24       | 0.46             | 0.01                  | 0.0       | 0.23              | 0.02                 | 0.62       | 38         | -                       |
-| llama-3-405b-instruct         | verbose      | 15.32       | 8.38       | 0.46             | 0.04                  | 0.0       | 0.40              | 0.06                 | 0.62       | 38         | -                       |
-
-You can find the detailed results under [results/sap_successfactor_4](results/sap_successfactor_4)
-
-
-## METRICS KEY
-
-| Metric | Description | Calculation | Range/Type |
-|--------|-------------|-------------|------------|
-| **Total Step** | Total number of messages/steps in the conversation | Count of all messages in the conversation | Integer ≥ 0 |
-| **Agent Step** | Number of assistant responses (text or tool calls) | Count of messages where `role == "assistant"` and `type` is text or tool_call | Integer ≥ 0 |
-| **Ground Truth Calls** | Expected number of tool calls based on ground truth | Count of goal_details with `type == ContentType.tool_call` | Integer ≥ 0 |
-| **Journey Success** | Whether the agent completed tasks in the correct order | `is_topological_sort(ground_truth.goals, labelled_messages)` | Boolean |
-| **Wrong Function Calls** | Number of calls to non-existent or unexpected functions | Count of labelled_messages containing "_WRONG_FUNCTION_CALL" | Integer ≥ 0 |
-| **Bad Calls** | Reserved metric for future use | Currently hardcoded to 0 | Integer (0) |
-| **Wrong Parameters** | Number of tool calls with incorrect parameters | Count of labelled_messages containing "_WRONG_PARAMETERS" | Integer ≥ 0 |
-| **Wrong Routing Calls** | Number of incorrect agent routing calls | Count of labelled_messages containing "_WRONG_ROUTING_CALL" | Integer ≥ 0 |
-| **Text Match** | Quality of final text summary | "Keyword Mismatch" \| "Semantic Mismatch" \| "Summary Matched" | Categorical |
-| **Tool Call Accuracy** | Percentage of non-routing tool calls that were executed correctly | `correct_tool_calls / non_transfer_tool_calls` | Float 0.0-1.0 |
-| **Tool Call Relevancy** | Percentage of non-routing tool calls that were relevant to the task | `(relevant_tool_calls - expected_routing_calls) / non_transfer_tool_calls` | Float 0.0-1.0 |
-| **Agent Routing Accuracy** | Percentage of routing calls that were executed correctly | `expected_routing_calls / total_routing_calls` | Float 0.0-1.0 |
-| **WXO Average Response Time (Secs)** | Average response time for agent responses | Mean response time across all agent interactions | Float ≥ 0.0 |
+## Results
+You can find benchmark results [here](benchmarks/domain_1.8/README.md)
 
-### Key Definitions
+## Metrics
 
-- **Relevant Tool Call**: A tool call whose name matches one of the expected tool names defined in the ground truth
-- **Correct Tool Call**: A relevant tool call that also has the correct parameters/arguments
-- **Routing Call**: A tool call whose name starts with "transfer_" (used for agent-to-agent routing)
-- **Non-Transfer Tool Call**: Regular tool calls excluding routing calls (`total_tool_calls - total_routing_calls`)
-- **Expected Routing Call**: A routing call that was both expected and executed correctly
+| Metric                     | Description                                                        | Calculation                                                                  | Range/Type         |
+|----------------------------|--------------------------------------------------------------------|------------------------------------------------------------------------------|--------------------|
+| **Total Steps**            | Total number of messages/steps in the conversation                 | Count of all messages in the conversation                                    | Integer ≥ 0        |
+| **LLM Steps**              | Number of assistant (LLM) responses (text or tool calls)           | Count of messages where `role == "assistant"`                                | Integer ≥ 0        |
+| **Total Tool Calls**       | Number of tool calls made by the agent                             | Count of all tool calls                                                      | Integer ≥ 0        |
+| **Tool Call Precision**    | Fraction of correct tool calls out of all tool calls               | `correct_tool_calls / total_tool_calls`                                      | Float 0.0–1.0      |
+| **Tool Call Recall**       | Fraction of correct tool calls out of expected tool calls          | `correct_tool_calls / expected_tool_calls`                                   | Float 0.0–1.0      |
+| **Agent Routing Accuracy** | Fraction of correct agents visited (relevant_routing_calls) out of total number of agents visited (total_routing_calls)       | `relevant_routing_calls / total_routing_calls`                               | Float 0.0–1.0      |
+| **Text Match**             | Whether the final summary text matches the ground truth            | `Summary Matched` \| `Summary MisMatched`                                    | Categorical        |
+| **Journey Success**        | Whether the agent completed tasks in the correct order             | Boolean (`True`/`False`)                                                     | Boolean            |
+| **Avg Resp Time (sec)**    | Average response time for agent responses                          | Mean response time across all agent interactions                             | Float ≥ 0.0        |
 
-### Averaging Behavior
-
-- **Per Test Case Average**: Total Step, Agent Step, Tool Call Accuracy, Tool Call Relevancy, Agent Routing Accuracy, WXO Average Response Time
-- **Per Ground Truth Calls Average**: Wrong Function Calls, Bad Calls, Wrong Parameters, Wrong Routing Calls
-- **Special Calculations**: 
-  - Journey Success: Proportion of test cases that succeeded (0.0-1.0)
-  - Text Match: Proportion of test cases with "Summary Matched" (0.0-1.0)
+### Key Definitions
 
+- **Correct Tool Call**: A tool call that matches both the expected function and arguments.
+- **Expected Tool Call**: A tool call that is required by the ground truth.
+- **Routing Call**: When an agent routes to another agent.
+- **Relevant Routing Call**: An agent is relevant when it's either the entry point agent or it includes a tool that is presented in the ground-truth.
+- **Text Match**: Indicates if the agent's final summary matches the expected summary ("Summary Matched") or does not match ("Summary MisMatched").
+- **Journey Success**: Indicates if the agent completed all required tasks in the correct order.

ibm_watsonx_orchestrate_evaluation_framework-1.0.9.dist-info/RECORD

@@ -0,0 +1,96 @@
+wxo_agentic_evaluation/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+wxo_agentic_evaluation/analyze_run.py,sha256=4QLlo_NQjCh5M52ztFHoMvk_jtwptKpVXDmdTxj2ikQ,13054
+wxo_agentic_evaluation/annotate.py,sha256=nxYMc6gwfQ-GuNjCPFtbX_-Es5-9XDdbXpMH89yRDdc,1228
+wxo_agentic_evaluation/arg_configs.py,sha256=a3Lo3RurTOLysxmsliMKIqvld7T3ZTb4Kw_FPEeBC78,2997
+wxo_agentic_evaluation/batch_annotate.py,sha256=44K4DUI498uaLIWUn3nz82AKcU6VnCjrExoG6GpPHoM,6323
+wxo_agentic_evaluation/data_annotator.py,sha256=6cUUpCTFSs36VF3wICLXWrWbEUJz6v-PzPeuzO9S1k8,8310
+wxo_agentic_evaluation/description_quality_checker.py,sha256=7vvGpPwa8J8ArTWAXRp865e_cHzSTMFLxkpI-rfj2ZQ,6097
+wxo_agentic_evaluation/evaluation_package.py,sha256=9NrpKaGOUnAkslP7t3vU3Uv4lFUs-XLu0IUO7q0Muik,23575
+wxo_agentic_evaluation/inference_backend.py,sha256=ItnwjhEJHX28sBS7CIVe7hmcy9FLd1HQEpzhdsJ1jDk,30341
+wxo_agentic_evaluation/llm_matching.py,sha256=l010exoMmsvTIAVHCm-Ok0diyeQogjCmemUb9rJLe6A,1477
+wxo_agentic_evaluation/llm_rag_eval.py,sha256=vsNGz1cFE5QGdhnfrx-iJq1r6q8tSI9Ef1mzuhoHElg,1642
+wxo_agentic_evaluation/llm_user.py,sha256=LhS7Ti9v3TLMrEv0og9N6yUF4y8lLMcMycEqVhwtGAE,1493
+wxo_agentic_evaluation/main.py,sha256=JYcOaSPM8EQdgsPFdYmelouH-3_o-OtLQ0oh5cjADOU,11933
+wxo_agentic_evaluation/quick_eval.py,sha256=nROa-xZ265-k8JJ1M4t1LZe4ucdJi8GuRNVuCWPiZTU,12525
+wxo_agentic_evaluation/record_chat.py,sha256=uFdbLt4HaMREN3q4HHAA1ZvtjoLdiBEyxPd9Eoc6svc,8103
+wxo_agentic_evaluation/resource_map.py,sha256=11qF1oJDwGNWOLYFVsIPsR66JK4eD0cqVOBKreK2mPQ,1644
+wxo_agentic_evaluation/service_instance.py,sha256=6Y7byxdQakB3NMP288Rhne3ygOumSSgJjBT5Q-YY1OA,6468
+wxo_agentic_evaluation/test_prompt.py,sha256=ksteXCs9iDQPMETc4Hb7JAXHhxz2r678U6-sgZJAO28,3924
+wxo_agentic_evaluation/tool_planner.py,sha256=00e_d2Ju5J61priEaKWLkSK2yW0donK8KJCq0PfKUuw,13013
+wxo_agentic_evaluation/type.py,sha256=R_s2kFn3VydHI4y5aWSBEaYPpDODHF5yPb7MKbysxwk,4014
+wxo_agentic_evaluation/analytics/tools/analyzer.py,sha256=IPX_lAFujjPVI9fhXTNohXTxTmpqRhfzQygCWDYHBHg,18125
+wxo_agentic_evaluation/analytics/tools/main.py,sha256=dxjjIlVQY-ZJ3NC6knW8r-kmTo8WWEhwlwZfP38uj8Q,6105
+wxo_agentic_evaluation/analytics/tools/types.py,sha256=IFLKI1CCQwPR2iWjif8AqL_TEq--VbLwdwnMqfJujBw,4461
+wxo_agentic_evaluation/analytics/tools/ux.py,sha256=EaWNvsq68X_i2H4pQ2fABtXEEmk3ZXqaMrTs42_7MwE,18347
+wxo_agentic_evaluation/external_agent/__init__.py,sha256=9NomrFEZQPrh91nto_hEGwoSks77nerAbWqS0L70qnY,1511
+wxo_agentic_evaluation/external_agent/external_validate.py,sha256=xW8tqPcm8JYvveSxf-oFCajvF5J8ORaK23YXu-LuFmc,4142
+wxo_agentic_evaluation/external_agent/performance_test.py,sha256=vaaAMBhJoQ0hQ4xq4Zp7E39Xtba05inWaKzkAtWlhlY,2426
+wxo_agentic_evaluation/external_agent/types.py,sha256=4kfWD_ZyGZmpbib33gCxEuKS4HLb7CEtferlQgQe7uk,1624
+wxo_agentic_evaluation/metrics/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+wxo_agentic_evaluation/metrics/llm_as_judge.py,sha256=bybJQfVWiVh3BoFEZjdBmU9EQO9Ukheu3YWmkI9b1ks,1218
+wxo_agentic_evaluation/metrics/metrics.py,sha256=V9tcGHuwG1_m0Aa8ztmduBR8gufr6rpvZjlzPtPnDZQ,6236
+wxo_agentic_evaluation/prompt/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+wxo_agentic_evaluation/prompt/answer_relevancy_prompt.jinja2,sha256=vLrMWce-5HlvniCQdtifnl-YdbJfT8-oixzfwulZs98,3839
+wxo_agentic_evaluation/prompt/args_extractor_prompt.jinja2,sha256=0qBicXFcc6AA3mQNLPVRmFsnuYaCABJXgZkIH9fO0Js,952
+wxo_agentic_evaluation/prompt/bad_tool_descriptions_prompt.jinja2,sha256=_Ty6QDcQcbde2ZP2HVvFtOCm_2mFu_1cUM6qj11MvcU,8085
+wxo_agentic_evaluation/prompt/batch_testcase_prompt.jinja2,sha256=QXuk2ecnEPPRCPoWZJyrtb1gAVuIPljB91YoqPBp2Dk,1896
+wxo_agentic_evaluation/prompt/faithfulness_prompt.jinja2,sha256=DW9OdjeZJbOWrngRqTAVD4w0va_HtA2FR4G1POIIamM,2524
+wxo_agentic_evaluation/prompt/keyword_matching_prompt.jinja2,sha256=7mTkSrppjgPluUAIMTWaT30K7M4J4hyR_LjSjW1Ofq0,1290
+wxo_agentic_evaluation/prompt/keywords_generation_prompt.jinja2,sha256=PiCjr1ag44Jk5xD3F24fLD_bOGYh2sF0i5miY4OrVlc,1890
+wxo_agentic_evaluation/prompt/llama_user_prompt.jinja2,sha256=GAHtEJvFNtgWBQma1I9KJdhXdhmqbEQf7JY66Z1JLMU,1113
+wxo_agentic_evaluation/prompt/off_policy_attack_generation_prompt.jinja2,sha256=yrLtXfmVIJ_C3XIaTvpqlQGlg9kKIibrVR3UzpzBEmo,1288
+wxo_agentic_evaluation/prompt/on_policy_attack_generation_prompt.jinja2,sha256=90KF7fXW5PPwY8IkPqA6ZflDMkr_KFDpO9H_mVGdGf8,2212
+wxo_agentic_evaluation/prompt/semantic_matching_prompt.jinja2,sha256=MltPfEXYyOwEC2xNLl7UsFTxNbr8CwHaEcPqtvKE2r8,2749
+wxo_agentic_evaluation/prompt/starting_sentence_generation_prompt.jinja2,sha256=m_l6f7acfnWJmGQ0mXAy85oLGLgzhVhoz7UL1FVYq8A,4908
+wxo_agentic_evaluation/prompt/story_generation_prompt.jinja2,sha256=_DxjkFoHpNTmdVSUzUrUdwn4Cng7nAGqkMnm0ScOH1w,4191
+wxo_agentic_evaluation/prompt/template_render.py,sha256=BVRT-BKyBJn5cM6Dze4GhFmMLyvGlyilFKQsfUhrklQ,4722
+wxo_agentic_evaluation/prompt/tool_chain_agent.jinja2,sha256=9RcIjLYoOvtFsf-RgyMfMcj2Fe8fq1wGkE4nG1zamYY,297
+wxo_agentic_evaluation/prompt/tool_planner.jinja2,sha256=Ln43kwfSX50B1VBsT-MY1TCE0o8pGFh8aQJAzZfGkpI,3239
+wxo_agentic_evaluation/prompt/examples/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+wxo_agentic_evaluation/prompt/examples/data_simple.json,sha256=XXF-Pn-mosklC9Ch7coyaJxosFNnl3OkHSW3YPuiKMM,2333
+wxo_agentic_evaluation/red_teaming/attack_evaluator.py,sha256=rlkSAb7QDHUoXg-LLK_wOyaTtYNrhV2SXbpnJxSUrD0,4714
+wxo_agentic_evaluation/red_teaming/attack_generator.py,sha256=YQi9xoaFATBNGe_NebndH6o1eQalcSKvWKSjbZ8dzP4,11526
+wxo_agentic_evaluation/red_teaming/attack_list.py,sha256=edphWARWqDtXFtcHTVbRXngvO0YfG5SgrfPtrBRXuFw,4734
+wxo_agentic_evaluation/red_teaming/attack_runner.py,sha256=qBZY4GK1352NUMyED5LVjjbcvpdCcxG6mDIN1HvxKIc,4340
+wxo_agentic_evaluation/referenceless_eval/__init__.py,sha256=lijXMgQ8nQe-9eIfade2jLfHMlXfYafMZIwXtC9KDZo,106
+wxo_agentic_evaluation/referenceless_eval/referenceless_eval.py,sha256=82jutvZ21GO76rRcjGWux5kAanDtzE728BijC7trSxY,4297
+wxo_agentic_evaluation/referenceless_eval/function_calling/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+wxo_agentic_evaluation/referenceless_eval/function_calling/consts.py,sha256=UidTaT9g5IxbcakfQqP_9c5civ1wDqY-PpPUf0uOXJo,915
+wxo_agentic_evaluation/referenceless_eval/function_calling/metrics/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+wxo_agentic_evaluation/referenceless_eval/function_calling/metrics/base.py,sha256=th36x0RMpGx1MAzqOUxjuhAcroUgjT2CJkT6tlMUbPg,843
+wxo_agentic_evaluation/referenceless_eval/function_calling/metrics/loader.py,sha256=5ZOWW82V0VFgpiaXpQ3hZIVKO7JAsoYRhwwb2ZDGxxk,7481
+wxo_agentic_evaluation/referenceless_eval/function_calling/metrics/function_call/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+wxo_agentic_evaluation/referenceless_eval/function_calling/metrics/function_call/general.py,sha256=P1ytcARCy696ZCLq9tcaQWgaolmu0ON_kySmmTeyBtc,1549
+wxo_agentic_evaluation/referenceless_eval/function_calling/metrics/function_call/general_metrics_runtime.json,sha256=bVSyudTk2Nim1DcgQZf8ilOTszY2Kgm4YU6beHWvEhQ,40475
+wxo_agentic_evaluation/referenceless_eval/function_calling/metrics/function_selection/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+wxo_agentic_evaluation/referenceless_eval/function_calling/metrics/function_selection/function_selection.py,sha256=UwPMCn7T2BEut7Mbj6U5UJvb2AAZw03BiB9wEjSCheg,1017
+wxo_agentic_evaluation/referenceless_eval/function_calling/metrics/function_selection/function_selection_metrics_runtime.json,sha256=o4oRur1MiXO2RYzmzj07QOBzX75DyU7T7yd-gFsgFdo,30563
+wxo_agentic_evaluation/referenceless_eval/function_calling/pipeline/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+wxo_agentic_evaluation/referenceless_eval/function_calling/pipeline/adapters.py,sha256=kMMFq4ABX5q6cPnDdublLMVqXu4Ij-x4OlxZyePWIjc,3599
+wxo_agentic_evaluation/referenceless_eval/function_calling/pipeline/pipeline.py,sha256=rKKBtL47gvN_fFy3FVbQTaa0U1bhv8bls0HPZi66EZ8,9279
+wxo_agentic_evaluation/referenceless_eval/function_calling/pipeline/semantic_checker.py,sha256=z_k-qdFoUJqstkPYn9Zmhlp2YTVQKJtoDZCIdKow664,17306
+wxo_agentic_evaluation/referenceless_eval/function_calling/pipeline/static_checker.py,sha256=_Er2KfCkc3HFmOmxZT6eb-e7qF7ukqsf6Si5CJTqPPg,6016
+wxo_agentic_evaluation/referenceless_eval/function_calling/pipeline/transformation_prompts.py,sha256=tW1wc87WIm8BZh2lhdj1RDP6VdRLqZBWSMmemSttbGs,22034
+wxo_agentic_evaluation/referenceless_eval/function_calling/pipeline/types.py,sha256=QHIEHmr3GnCoQPIPyLAMiT2IdYJKUUhqSPJDLefVY2U,16983
+wxo_agentic_evaluation/referenceless_eval/metrics/__init__.py,sha256=uJc7ZwK6pJpsMIuBSBXUtxdvd-aiRnOXGX3aeyvw2ik,151
+wxo_agentic_evaluation/referenceless_eval/metrics/field.py,sha256=z4S5QJJi1acshC0YFzblppgtm1oxNEgMKYjaJdfzkn4,8324
+wxo_agentic_evaluation/referenceless_eval/metrics/metric.py,sha256=mSoJAjYRSEpq8zBm-EP0UwF0zmZ4gDRjoUe4jT9nJt0,12212
+wxo_agentic_evaluation/referenceless_eval/metrics/metrics_runner.py,sha256=JHZhoSfGJYYp3sFx3XP9cTsDQgpgajzZ7TV5c4hmKCs,5980
+wxo_agentic_evaluation/referenceless_eval/metrics/prompt.py,sha256=CGQ5LvhQrmxAyZDHBHds47rjJYWsx670c56yOHCrEAI,15074
+wxo_agentic_evaluation/referenceless_eval/metrics/utils.py,sha256=jurmc4KFFKH4hwnvor2xg97H91b-xJc3cUKYaU2I8uM,1370
+wxo_agentic_evaluation/referenceless_eval/prompt/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+wxo_agentic_evaluation/referenceless_eval/prompt/runner.py,sha256=FFmcSWXQnLmylpYyj8LZuPwb6nqwQp-jj6Mv9g8zby0,5052
+wxo_agentic_evaluation/service_provider/__init__.py,sha256=EaY4jjKp58M3W8N3b3a8PNC2S81xA7YV2_QkTIy9DfI,1600
+wxo_agentic_evaluation/service_provider/model_proxy_provider.py,sha256=Y36Ryv4nPG8RdVP_zsQsRlEWv8F_hGi7-wOppWPQTwc,4026
+wxo_agentic_evaluation/service_provider/ollama_provider.py,sha256=HMHQVUGFbLSQI1dhysAn70ozJl90yRg-CbNd4vsz-Dc,1116
+wxo_agentic_evaluation/service_provider/provider.py,sha256=MsnRzLYAaQiU6y6xf6eId7kn6-CetQuNZl00EP-Nl28,417
+wxo_agentic_evaluation/service_provider/watsonx_provider.py,sha256=C_uezsx4FHqkvtCIQGIii82nu9_vSOKP70uAVXWj5fM,10619
+wxo_agentic_evaluation/utils/__init__.py,sha256=ItryTgc1jVc32rB3XktTFaYGA_A6bRIDZ1Pts_JGmv8,144
+wxo_agentic_evaluation/utils/open_ai_tool_extractor.py,sha256=Vyji_edgou2xMLbsGwFG-QI7xRBNvO3-1nbeOc8ZuFo,5646
+wxo_agentic_evaluation/utils/rich_utils.py,sha256=J9lzL4ETQeiAJcXKsUzXh82XdKvlDY7jmcgTQlwmL9s,6252
+wxo_agentic_evaluation/utils/rouge_score.py,sha256=WvcGh6mwF4rWH599J9_lAt3BfaHbAZKtKEJBsC61iKo,692
+wxo_agentic_evaluation/utils/utils.py,sha256=qQR_2W5p0Rk6KSE3-llRyZrWXkO5zG9JW7H1692L4PI,11428
+ibm_watsonx_orchestrate_evaluation_framework-1.0.9.dist-info/METADATA,sha256=OgVKCRnQdk7cYnzrexY8VPPgQW73uHdX2jfI9a7o5IE,16105
+ibm_watsonx_orchestrate_evaluation_framework-1.0.9.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+ibm_watsonx_orchestrate_evaluation_framework-1.0.9.dist-info/top_level.txt,sha256=2okpqtpxyqHoLyb2msio4pzqSg7yPSzwI7ekks96wYE,23
+ibm_watsonx_orchestrate_evaluation_framework-1.0.9.dist-info/RECORD,,

wxo_agentic_evaluation/analytics/tools/main.py

@@ -5,8 +5,8 @@ import rich
 from type import ContentType
 from analytics.tools.analyzer import ToolErrorAnalyzer
 from analytics.tools.ux import ToolErrorDisplayManager
-from type import Message
 from shutil import get_terminal_size
+from utils.utils import load_messages
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="tool-analytics-resources")
@@ -47,23 +47,6 @@ if __name__ == "__main__":
         """Count total tool calls in the conversation."""
         return sum(1 for msg in messages if msg.type == ContentType.tool_call)
 
-    # Function to load messages from JSON file
-    def load_messages(file_path):
-        with open(file_path, "r") as f:
-
-            try:
-                message_data = json.load(f)
-                messages = []
-                for msg in message_data:
-                    messages.append(Message.model_validate(msg))
-
-                return messages
-
-            except Exception as e:
-                print(file_path)
-                print(e)
-                return None
-
     # Function to load ground truth from JSON file
     def load_ground_truth(file_path):
         with open(file_path, "r") as f: