chore: wip

[zeeshanlakhani]

No description provided.

[codecov]

Codecov Report

Attention: Patch coverage is 0.82645% with 120 lines in your changes are missing coverage. Please review.

Project coverage is 69.19%. Comparing base (2d0cfe9) to head (bf786db).
Report is 1 commits behind head on main.

❗ Current head bf786db differs from pull request most recent head 528c69b. Consider uploading reports for the commit 528c69b to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #640      +/-   ##
==========================================
- Coverage   69.85%   69.19%   -0.67%     
==========================================
  Files          98       98              
  Lines       13394    13515     +121     
==========================================
- Hits         9357     9352       -5     
- Misses       4037     4163     +126     

Files	Coverage Δ
homestar-wasm/src/wasmtime/world.rs	87.94% <100.00%> (+0.05%)	⬆️
homestar-functions/subtract/src/bindings.rs	0.00% <0.00%> (ø)
homestar-wasm/src/wasmtime/host/helpers.rs	9.86% <0.00%> (-31.80%)	⬇️

... and 2 files with indirect coverage changes

[bgins]

Suggested change

To more applicability encode the [MapReduce][map-reduce] example from the [Crowdforge][crowdforge] paper, I implemented a `prompt_chain` Wasm/WASI function registered on the Host that takes in a system prompt (e.g. "You are journalist writing about cities."), an input (e.g. an ongoing article), a map step prompt with a `{{text}}` placeholder that is filled in, a reduce step, which folds over (combines) the generated text(s) from the map step, and then the optional LLaMA model stored as a [`gguf`][gguf]. If the optional model path is not provided, the Host will fall back to the default `Meta-Llama-3-8B-Instruct.Q4_0.gguf` model.

To more applicably encode the [MapReduce][map-reduce] example from the [Crowdforge][crowdforge] paper, I implemented a `prompt_chain` Wasm/WASI function registered on the Host that takes in a system prompt (e.g. "You are journalist writing about cities."), an input (e.g. an ongoing article), a map step prompt with a `{{text}}` placeholder that is filled in, a reduce step, which folds over (combines) the generated text(s) from the map step, and then the optional LLaMA model stored as a [`gguf`][gguf]. If the optional model path is not provided, the Host will fall back to the default `Meta-Llama-3-8B-Instruct.Q4_0.gguf` model.