Generalize Support for LiteLLM Models #265 (#272)

* model_info (Model -> ConfiguredModel in constants) - 265

* predictor function for unknown spec

* update full list of API keys

* add gemini3 and gpt5.2 to constants

* return models based on opt obj when models is None

* reorganize functions in model info/helper

* add tests and update model references and imports

* move validation from config to query processor

* add json file for model score/latency and update predictor function

* update model references and imports

* update dependencies and related test cases

* update Model to have both string and enum

* model_info -> model_helper

* update model usage in query config

* rollback import changes for CuratedModel -> Model

* ModelProvider class

* update all switch cases to ModelProvider when applicable

* reverted CuratedModel changes

* add test cases

* add additional test cases

* fix formatting issues

* add prompt caching stats for #262

* restructure Model class

* fix Model enum issue

* add sorting logic to model class

* use singular json file for info fetching

* expand model list and updates curated_model_info file

* restructure model info fetching, update Model class and test cases

* script to update pz_models_information and update get_optimal_models

* is_deepseek_model

* add audio cache read/creation

* remove claude sonnet 3.5 (retired)

* add deepseek-chat

* add .json files to pyproject.toml so that  is packaged too

* revert uvicorn dependency

* some small tweaks

* passing tests

---------

Co-authored-by: joycequ <joycequ@mit.edu>
Co-authored-by: Matthew Russo <mdrusso@mit.edu>

Author: 3 people

.gitignore

@@ -53,6 +53,7 @@ testdata/enron-eval/*.txt
 pyrightconfig.json
 
 myenv/
+pz-env/
 
 # abacus-research data
 abacus-research/cuad-data/*

abacus-research/helper-scripts/mmqa-baseline.py

@@ -7,7 +7,7 @@
 import numpy as np
 from openai import OpenAI
 
-from palimpzest.constants import MODEL_CARDS, Cardinality, Model
+from palimpzest.constants import Cardinality, Model
 from palimpzest.query.generators.generators import get_json_from_answer
 
 
@@ -109,8 +109,9 @@ def f1(preds: list | None, targets: list):
         completion = client.chat.completions.create(**payload)
 
         # compute total cost
-        usd_per_input_token = MODEL_CARDS[model_name]["usd_per_input_token"]
-        usd_per_output_token = MODEL_CARDS[model_name]["usd_per_output_token"]
+        model = Model(model_name)
+        usd_per_input_token = model.get_usd_per_input_token()
+        usd_per_output_token = model.get_usd_per_output_token()
         input_tokens = completion.usage.prompt_tokens
         output_tokens = completion.usage.completion_tokens
         total_cost += input_tokens * usd_per_input_token + output_tokens * usd_per_output_token

pyproject.toml

@@ -14,7 +14,7 @@ dependencies = [
     "chromadb>=1.0.15",
     "colorama>=0.4.6",
     "datasets>=4.0.0",
-    "fastapi~=0.115.0",
+    "fastapi>=0.115.0",
     "gradio>=5.26.0",
     "litellm>=1.76.1",
     "numpy==2.0.2",
@@ -63,8 +63,7 @@ where = ["src"]
 namespaces = false
 
 [tool.setuptools.package-data]
-"*" = ["*.txt", "*.rst", "*.md"]
-
+"*" = ["*.txt", "*.rst", "*.md", "*.json"]
 
 [tool.pytest.ini_options]
 testpaths = ["tests/pytest"]