base on LLM Transparency Tool (LLM-TT), an open-source interactive toolkit for analyzing internal workings of Transformer-based language models. *Check out demo at* https://huggingface.co/spaces/facebook/llm-transparency-tool-demo <h1> <img width="500" alt="LLM Transparency Tool" src="https://github.com/facebookresearch/llm-transparency-tool/assets/1367529/795233be-5ef7-4523-8282-67486cf2e15f"> </h1> <img width="832" alt="screenshot" src="https://github.com/facebookresearch/llm-transparency-tool/assets/1367529/78f6f9e2-fe76-4ded-bb78-a57f64f4ac3a"> ## Key functionality * Choose your model, choose or add your prompt, run the inference. * Browse contribution graph. * Select the token to build the graph from. * Tune the contribution threshold. * Select representation of any token after any block. * For the representation, see its projection to the output vocabulary, see which tokens were promoted/suppressed but the previous block. * The following things are clickable: * Edges. That shows more info about the contributing attention head. * Heads when an edge is selected. You can see what this head is promoting/suppressing. * FFN blocks (little squares on the graph). * Neurons when an FFN block is selected. ## Installation ### Dockerized running ```bash # From the repository root directory docker build -t llm_transparency_tool . docker run --rm -p 7860:7860 llm_transparency_tool ``` ### Local Installation ```bash # download git clone [email protected]:facebookresearch/llm-transparency-tool.git cd llm-transparency-tool # install the necessary packages conda env create --name llmtt -f env.yaml # install the `llm_transparency_tool` package pip install -e . # now, we need to build the frontend # don't worry, even `yarn` comes preinstalled by `env.yaml` cd llm_transparency_tool/components/frontend yarn install yarn build ``` ### Launch ```bash streamlit run llm_transparency_tool/server/app.py -- config/local.json ``` ## Adding support for your LLM Initially, the tool allows you to select from just a handful of models. Here are the options you can try for using your model in the tool, from least to most effort. ### The model is already supported by TransformerLens Full list of models is [here](https://github.com/neelnanda-io/TransformerLens/blob/0825c5eb4196e7ad72d28bcf4e615306b3897490/transformer_lens/loading_from_pretrained.py#L18). In this case, the model can be added to the configuration json file. ### Tuned version of a model supported by TransformerLens Add the official name of the model to the config along with the location to read the weights from. ### The model is not supported by TransformerLens In this case the UI wouldn't know how to create proper hooks for the model. You'd need to implement your version of [TransparentLlm](./llm_transparency_tool/models/transparent_llm.py#L28) class and alter the Streamlit app to use your implementation. ## Citation If you use the LLM Transparency Tool for your research, please consider citing: ```bibtex @article{tufanov2024lm, title={LM Transparency Tool: Interactive Tool for Analyzing Transformer Language Models}, author={Igor Tufanov and Karen Hambardzumyan and Javier Ferrando and Elena Voita}, year={2024}, journal={Arxiv}, url={https://arxiv.org/abs/2404.07004} } @article{ferrando2024information, title={Information Flow Routes: Automatically Interpreting Language Models at Scale}, author={Javier Ferrando and Elena Voita}, year={2024}, journal={Arxiv}, url={https://arxiv.org/abs/2403.00824} } ```` ## License This code is made available under a [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/) license, as found in the LICENSE file. However you may have other legal obligations that govern your use of other content, such as the terms of service for third-party models. ", Assign "at most 3 tags" to the expected json: {"id":"9415","tags":[]} "only from the tags list I provide: [{"id":77,"name":"3d"},{"id":89,"name":"agent"},{"id":17,"name":"ai"},{"id":54,"name":"algorithm"},{"id":24,"name":"api"},{"id":44,"name":"authentication"},{"id":3,"name":"aws"},{"id":27,"name":"backend"},{"id":60,"name":"benchmark"},{"id":72,"name":"best-practices"},{"id":39,"name":"bitcoin"},{"id":37,"name":"blockchain"},{"id":1,"name":"blog"},{"id":45,"name":"bundler"},{"id":58,"name":"cache"},{"id":21,"name":"chat"},{"id":49,"name":"cicd"},{"id":4,"name":"cli"},{"id":64,"name":"cloud-native"},{"id":48,"name":"cms"},{"id":61,"name":"compiler"},{"id":68,"name":"containerization"},{"id":92,"name":"crm"},{"id":34,"name":"data"},{"id":47,"name":"database"},{"id":8,"name":"declarative-gui "},{"id":9,"name":"deploy-tool"},{"id":53,"name":"desktop-app"},{"id":6,"name":"dev-exp-lib"},{"id":59,"name":"dev-tool"},{"id":13,"name":"ecommerce"},{"id":26,"name":"editor"},{"id":66,"name":"emulator"},{"id":62,"name":"filesystem"},{"id":80,"name":"finance"},{"id":15,"name":"firmware"},{"id":73,"name":"for-fun"},{"id":2,"name":"framework"},{"id":11,"name":"frontend"},{"id":22,"name":"game"},{"id":81,"name":"game-engine "},{"id":23,"name":"graphql"},{"id":84,"name":"gui"},{"id":91,"name":"http"},{"id":5,"name":"http-client"},{"id":51,"name":"iac"},{"id":30,"name":"ide"},{"id":78,"name":"iot"},{"id":40,"name":"json"},{"id":83,"name":"julian"},{"id":38,"name":"k8s"},{"id":31,"name":"language"},{"id":10,"name":"learning-resource"},{"id":33,"name":"lib"},{"id":41,"name":"linter"},{"id":28,"name":"lms"},{"id":16,"name":"logging"},{"id":76,"name":"low-code"},{"id":90,"name":"message-queue"},{"id":42,"name":"mobile-app"},{"id":18,"name":"monitoring"},{"id":36,"name":"networking"},{"id":7,"name":"node-version"},{"id":55,"name":"nosql"},{"id":57,"name":"observability"},{"id":46,"name":"orm"},{"id":52,"name":"os"},{"id":14,"name":"parser"},{"id":74,"name":"react"},{"id":82,"name":"real-time"},{"id":56,"name":"robot"},{"id":65,"name":"runtime"},{"id":32,"name":"sdk"},{"id":71,"name":"search"},{"id":63,"name":"secrets"},{"id":25,"name":"security"},{"id":85,"name":"server"},{"id":86,"name":"serverless"},{"id":70,"name":"storage"},{"id":75,"name":"system-design"},{"id":79,"name":"terminal"},{"id":29,"name":"testing"},{"id":12,"name":"ui"},{"id":50,"name":"ux"},{"id":88,"name":"video"},{"id":20,"name":"web-app"},{"id":35,"name":"web-server"},{"id":43,"name":"webassembly"},{"id":69,"name":"workflow"},{"id":87,"name":"yaml"}]" returns me the "expected json"