Features, bugs, research, ... that we are not actively working on

[zimmski]

TODO sort and sort out:

• Models
  • Better prompting with templates: Setting a mandatory document start Better prompting with templates: Setting a mandatory document start #29
  • Retry with feedback and retry without feedback. Retry with feedback and retry without feedback. #30
  • Implement "chain of thought" tasks Implement "chain of thought" tasks #31
  • Deal with dependencies requested by LLMs Deal with dependencies requested by LLMs #174
  • Evaluation run for all "good open weight models" with all available quantizations and different GPUs Evaluation run for all "good open weight models" with all available quantizations and different GPUs #209
  • Exclude openrouter models auto and flavor-of-the-week automatically in the provider
  • Rethink retry logic for LLM Providers Rethink retry logic for LLM Providers #305
  • Openrouter Provider preferences Openrouter Provider preferences #286
  • Include more models (We have the main problem that there are multiple models coming out every day. We should not wait for a "new version" of the eval, we should test these models right away and compare them. big problem: how do we promote findings?)
    • Nous-Hermes-2-SOLAR-10.7B; Also "Tree of Thoughts" approach might be interesting as a task
  • Maybe use https://huggingface.co/inference-endpoints/dedicated
• Metrics & Reporting
  • Evaluation folder with date cannot be created on windows Evaluation folder with date cannot be created on windows #151
  • extend the report command such that it takes result csv's and automatically
    • does the summing and aggregation (if we still want that to be a separate step)
    • finds the maximum scores for that evaluation run
    • once we have the leaderboard, we basically want to configure the repository such that we just add a model to some config somewhere and the GitHub actions run automatically and benchmark this model
    • or in a similar fashion, we just do a new release and the GitHub actions run automatically and benchmark everything for the new version
  • Automatically updated leader board for this repository: Do an up-to-date leaderboard/dashboard for current models current evaluation #26
    • Take a look at current leaderboards and evals to know what could be interesting Current popular code leaderboards are [LiveCodeBench](https://huggingface.co/spaces/livecodebench/leaderboard), the [BigCode models leaderboard](https://huggingface.co/spaces/bigcode/bigcode-models-leaderboard), [CyberSecEval](https://huggingface.co/spaces/facebook/CyberSecEval) and [CanAICode](https://huggingface.co/spaces/mike-ravkine/can-ai-code-results)
  • Scoring
    • Infer if a model produced "too much" code Infer if a model produced "too much" code #44
    • Introduce an AST-differ that also gives metrics Introduce an AST-differ that also gives metrics #80
    • Add linters where each error is a metric Add linters where each error is a metric #81
    • Include metrics about the models for comparing models Include metrics about the models for comparing models #82
    • Coverage for Java is tracked for lines, while Go is tracked for ranges Coverage for Java is tracked for lines, while Go is tracked for ranges #193
    • Weight "executed code" more prominently Weight "executed code" more prominently #233
    • AST differ Introduce an AST-differ that also gives metrics #80
    • Linters Add linters where each error is a metric #81
    • Automatically infer "Extra code" Infer if a model produced "too much" code #44
    • Figure out the "perfect" coverage score so we can display percentage of coverage reached
    • Make coverage metric fair
      • "Looking through logs... Java consistently has more code than Go for the same tasks, which yields more coverage. So a model that solves all Java tasks but no Go is automatically higher ranked than the opposite." -> only Symflower coverage will make this fair
    • distinguish between latency (time-to-first-token) and throughput (tokens generated per second)
    • Failing tests should receive a score penalty
  • Metrics
    • Track query tokens and save them to CSV Track query tokens and save them to CSV #347
    • Non-benchmark metrics (cost, weights open, ...) Include metrics about the models for comparing models #82
      • Save the descriptons of the models as well: https://openrouter.ai/api/v1/models The reason is that these can change over time, and we need to know after a while what they where. e.g right now i would like to know if mistral-7b-instruct for the last evaluation was v0.1. or not
    • Query REAL costs of all the testing of a model: the reason this is interesting is that some models have HUGE outputs, and since more output means more costs, this should be addressed in the score.
  • Reporting
    • Do an up-to-date leaderboard/dashboard for current models current evaluation Do an up-to-date leaderboard/dashboard for current models current evaluation #26
    • Bar charts should have have their value on the bar. The axis values do not work that well
    • Pick an example or several examples per category: goal is to find interesting results automatically, because it will get harder and harder to go manually through results.
    • Total-scores vs costs scatterplot. Result is upper-left-corner sweat spot: cheap and good results.
    • Scoring, Categorization, Bar Charts split by language.
    • Piechart of whole evaluations costs: for each LLM show how much it costs. Result is to see which LLMs are costing the most to run the eval.
    • deep-dive content
      • What are results that align with expectations? what are results against expectations? E.g. are there small LLMs that are better than big ones?
      • Are there big LLMs that totally fail?
      • Are there small LLMs that are surprisingly good?
      • What about LLMs where the commonity doesn't know that much yet: e.g. Snowflake, DBRX, ...
    • Order models by open.weight, allows commercial-use, closed, and price(!) and size: e.g. https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1 is great because open-weight, and Apache2 so commerical-use allowed. Should be better rated than GPT4
    • Categorize by parameters/experts https://www.reddit.com/r/LocalLLaMA/comments/1cdivc8/comment/l1davhv/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button
    • Compare input/output/request/... costs https://twitter.com/oleghuman/status/1786296672785420744
• Logging
  • Remove absolute paths completely e.g. in stack traces too.
  • Log request and response in their own files, so both can be used 1:1 (character for character) directly for debugging them: Keep individual coverage files and LLM query/responses #204
• Tooling & Installation
  • CI and tools
    • InstallToolsPath is not used for test execution (make test) InstallToolsPath is not used for test execution (make test) #93
    • Test for pulling Ollama model is flaky Test for pulling Ollama model is flaky #135
    • Flaky CI because of corrupted Z3 installation Flaky CI because of corrupted Z3 installation #107
    • Follow up: Ollama Support Follow up: Ollama Support #100
    • ollama_llama_server and other background processes we start must be killed on CTRL+C ollama_llama_server and other background processes we start must be killed on CTRL+C #164
    • Enable Ruby tests in Windows CI Enable Ruby tests in Windows CI #334
    • Move Dependency installation of Docker into multistage builds Move Dependency installation of Docker into multistage builds #319
  • Rescore existing models / eval with fixes e.g. when we do a better code repair tool, the LLM answer did not change, so we should rescore right away with new version of tool over a whole result of an eval.
  • Automatic tool installation with fixed version
    • Go
    • Java
  • Ensure that non-critical CLI input validation (such as unavailable models) does not panic
  • Ollama support
    • Install and test Ollama on MacOS
    • Install and test Ollama on Windows
  • Allow to forward CLI commands to be evaluated: Ollama provider #27 (comment)
  • Refactor Model and Provider to be in the same package Preload Ollama models before inference and unload afterwards #121 (comment)
• Evaluation
  • Interactive result comparison Interactive result comparison #208
  • Benchmark quantized models, because they need less memory
    • Evaluation run for all "good open weight models" with all available quantizations and different GPUs #209
      • https://twitter.com/HaihaoShen/status/1789178048308543688
  • Do an evaluation with different temperatures
  • Java
    • Let the Java test case for No test files actually identify and error that there are no test files (needs to be implemented in symflower test)
  • LLM
    • Improve LLM prompt
      • Take a look at https://x.com/dottxtai/status/1798443290913853770
      • Add an app-name to the requests so people know we are the eval https://openrouter.ai/docs#quick-start shows that other openapi-packages implement custom headers, but the one Go package we are using does not implement that. So do a PR to contribute.
        • We need to fork or use another package Roadmap for v0.5.0 #79 (comment)
      • Add markers for system and user e.g. Ollama provider #27 (comment)
      • Think about a standardized way of print outputs e.g. JSON https://twitter.com/WuMinghao_nlp/status/1789094583290507626
  • Prepare language and evaluation logic for multiple files:
    • Use symflower symbols to receive files
  • Evaluation tasks
    • Evaluation task: TDD Evaluation task: TDD #194
    • Assess failing tests Assess failing tests #235
    • Add evaluation task for "querying the relative test file path of a relative implementation file path" e.g. "What is the test relative file path for some/implementation/file.go" ... it is "some/implementation/file_test.go" for most cases.
    • Add evaluation task for code refactoring: two function with the same code -> extract into a helper function
    • Add evaluation task for implementing and fixing bugs using TDD
  • Check determinism of models e.g. execute each plain repository X-times, and then check if they are stable.
  • Code repair
    • 0-shot, 1-shot, ...
      • With LLM repair
      • With tool repair
  • Do test file paths through
    • symflower symbols
    • Task for models
  • Move towards generated cases so models cannot integrate fixed cases to always have 100% score
  • Think about adding more trainings data generation features: This will also help with dynamic cases
    • Heard that Snowflake Arctic is very open with how they gathered trainings data... so we see what LLM creators think and want of trainings data
• Documentation
  • Clean up and extend README
    • Better examples for contributions
    • Overhaul explanation of "why" we need evaluation, i.e. why is it good to evaluate for an empty function that does nothing.
    • Extend "how to extend the benchmark" section with instructions on how to add new tasks + languages, so we can even use LLMs to add new stuff
  • Write down a playbook for evaluations, e.g. one thing that should happen is that we let the benchmark play 5 times and then sum up points, but ... the runs should have at least one hour break in between to not run into cached responses.
• Content
  • Benchmark that showcases base-models vs their fine-tuned coding model e.g. in v0.5.0 we see that Codestral, codellama, ... are worse
  • Snowflake against Databricks would be a nice comparison since they align company-wise and are new
  • Write Tutorial for using Ollama
  • YouTube video for using Ollama
  • Blog post about the different suffixed of models e.g. "chat" and "instruct" and eval them somehow. Idead from https://www.reddit.com/r/LocalLLaMA/comments/1bz5oyx/comment/kyrfap4/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button
  • Blog post about HumanEval
  • Blog post about training a small LLM directly on HumanEval
  • Blog post about "non-determinism of LLMs" https://community.openai.com/t/a-question-on-determinism/8185 good starting point, and how we can make them at least more stable.
  • Blogpost idea: misleading comments, weird coding style... how much does it take to confuse the most powerful AI? @ahumenberger
    • Maybe not only comments. What about obfuscated code, e.g. function and variables names are just random strings?
• Research
  • Take a look at https://twitter.com/SMT_Solvers/status/1783540994304066006
  • Take a look at all OpenRouter's features of the API e.g. https://openrouter.ai/docs#parameters
    • https://www.reddit.com/r/LocalLLaMA/comments/1cihrdt/comment/l29o97q/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button mentioned that repetition_penalty disabled helps with performance and better results for coding.
    • Requested new models for the eval
      https://www.reddit.com/r/LocalLLaMA/comments/1cihrdt/comment/l2d4im0/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button
  • Look at what Paul Gauthier is doing with benchmarking of Aider (and Aider too) https://twitter.com/paulgauthier/status/1787827703158386921?s=46 seems like a perfect match for what we want to do as tasks
  • Look at MLX which maybe helps with execution for us https://twitter.com/loganthorneloe/status/1787845883519775120
  • Take a look at xLSTM https://twitter.com/HochreiterSepp/status/1788072466675335185
  • Take a look at eval https://twitter.com/JiaweiLiu_/status/1783959954321252697
  • Take a look at evaluation framework https://twitter.com/hamelhusain/status/1788936691576975382?s=61
  • Dig through https://arxiv.org/pdf/2405.14782 thanks to https://x.com/clefourrier/status/1793913394871062970
  • Take a look at https://x.com/dottxtai/status/1798443290913853770
• Think about a commercial effort of the eval, that we can balance some of the costs that goes into maintaining this eval