…tion

The goals of this PR was to allow developers to use LLMs across their development and production environments just by specifying a short model family name and without having lots of conditional statements handling differences in environments.

E.g. calling llms("llama3.2") should obtain a reference to the "llama3.2:3b" model when working with ollama local inference and a reference to "meta-llama/Llama-3.2-3B-Instruct" when using Hugging Face Hub serverless inference.

Differences in naming of models, and the fact that each model is actually a collection of dozen of versions that differ in number of parameters (distillation) and quantization techniques made the naming resolution a tricky task.

Here are the new capabilities of the llms() API:

1. Tower now recognizes ~170 names of model families as of August 2025.

2. Users can specify a model family e.g. llms("deepseek-r1") in both local and Tower cloud environments, and Tower will resolve the model family to a particular model that is available for inference:
- in local environment, Tower will find the model that is installed and, if there are multiple installed, pick the one with the largest number of parameters
- in Tower cloud environments, Tower will take the first model returned by HF search, making sure that this model is servable by the Inference Service, if specified by users

In addition to using model family names, Users can also specify a particular model both in local and Tower cloud environments:
locally: llms("deepseek-r1:14b") or llms("llama3.2:latest")
serverlessly: llms("deepseek-ai/DeepSeek-R1-0528") or llms("meta-llama/Llama-3.2-3B-Instruct")

Expected use:

A developer wants to use use a model of the "llama3.2" family first in development and then in production. They would add this code to their Tower app:

```
model_name=os.getenv("MODEL_NAME")
llm = llms(model_name)
```

They would set up their environments as follows:

"dev" environment:

Tower Secrets:

MODEL_NAME = "llama3.2"
(any model of this family installed locally will do)

TOWER_INFERENCE_ROUTER=ollama

"prod" environment:

Tower Secrets:

MODEL_NAME = meta-llama/Llama-3.2-3B-Instruct
(use a particular model)

TOWER_INFERENCE_ROUTER=hugging_face_hub
TOWER_INFERENCE_ROUTER_API_KEY=hf_123456789
TOWER_INFERENCE_SERVICE=<together|...>

…l-llm-feature-that-retrieves-best

1. added HF Hub and Ollama as dependencies to pyproject.toml (not a brad comment)
2. bumped HF version to 0.34.3
3. updated pytest.ini.template with settings to find source code
4. in tower context and elsewhere, renamed inference_service to inference_provider
5. in llms, addressed bunch of comments from brad
6. removed test_env.py
7. addressed lots of brad comments re test_llms.py
8. updated uv.lock