Walter Nelson

Low-rank ADVI

36

#### Submission Checklist - [x] Run unit tests: `./runTests.py src/test/unit` - [x] Run cpplint: `make cpplint` - [x] Declare copyright holder and open-source license: see below - [ ] Flesh...

#### Summary: Continues a discussion with @avehtari from [here](https://discourse.mc-stan.org/t/intermediate-between-mean-field-and-full-rank-advi/8506). Distilled down: full-rank ADVI is constrained by memory. The mean-field approximation can be problematic for certain models. A sensible intermediate (a...

### Summary Include a way to vary `prefix_inspections` in modulefile generation for specific packages. ### Rationale Per https://github.com/spack/spack/issues/27346#issuecomment-1435695502, including `LD_LIBRARY_PATH` in all generated modulefiles is a bad idea in many...

Is it possible to get `select` logprobs like in versions <0.1.0?

2

**Is your feature request related to a problem? Please describe.** In old (handlebar) versions one could do something like: ```python program = guidance("The quick brown fox jumps over the lazy...

Allow the LLM to dictate key order when generating JSON

2

**Is your feature request related to a problem? Please describe.** `_gen_json_object` dictates key order. **Describe the solution you'd like** Instead, allow the LLM to dictate it, reducing perplexity when the...

Should `oneOf` in JSON schemas be supported?

4

**Is your feature request related to a problem? Please describe.** `oneOf` in JSON schemas is not currently supported: https://json-schema.org/understanding-json-schema/reference/combining **Describe the solution you'd like** Either: * Explicitly state that `oneOf`...

Support callbacks/tool use in JSON generation

3

**Is your feature request related to a problem? Please describe.** When generating a complex schema, sometimes I want to interleave generation, templating, and tool use. One way to support this...

Support huge JSON schemas

3

**Is your feature request related to a problem? Please describe.** Huge JSON schemas fail when generating the template. **Describe the solution you'd like** Ideally, the whole schema should not be...

`select` produces different results than `gen`, even though the maximum likelihood answer should be the same (tokenization/token healing issue?)

18

**The bug** I have a minimal reproducible example where I would expect `select` and `gen` to produce similar results, but they don't. My experimentation suggests maybe a tokenization or token...