Revert "Add FreeToken provider for on-device models (#91)" (#92)

This reverts commit 1b03370.

Co-authored-by: shubham3-ucb <shubham3-ucb@users.noreply.github.com>

Author: shubham3-ucb

README.md

@@ -350,23 +350,8 @@ Any [LiteLLM](https://docs.litellm.ai/)-compatible model works using `provider/m
 --model gemini/gemini-3-pro-preview                          # Gemini
 --model anthropic/claude-sonnet-4-20250514                   # Anthropic
 --model ollama/llama3 --api-base http://localhost:11434/v1   # Local (Ollama, vLLM, etc.)
---model freetoken/Qwen3.6-35B-A3B                            # On-device via FreeToken
 ```
 
-**On-device via [FreeToken](https://github.com/FlashML-org/FreeToken)** runs fully
-local, no API key. Serve a model, then use the `freetoken/` prefix; the endpoint
-(`http://127.0.0.1:1919/v1`) and key are set for you:
-
-```bash
-uv pip install "freetoken[accel]"
-ft serve --model Qwen3.6-35B-A3B          # OpenAI-compatible server on 127.0.0.1:1919
-
-uv run skydiscover-run initial_program.py evaluator.py \
-  --model freetoken/Qwen3.6-35B-A3B --search evox --iterations 100
-```
-
-Add `--api-base http://<host>:<port>/v1` if the server runs elsewhere.
-
 Multi-model pools with weighted sampling are supported in config:
 
 ```yaml

docs/content/docs/getting-started/configuration.mdx

@@ -13,23 +13,8 @@ compatible model works:
 --model gemini/gemini-3-pro-preview      # Google Gemini
 --model anthropic/claude-sonnet-4-6      # Anthropic
 --model ollama/llama3 --api-base http://localhost:11434/v1  # Local
---model freetoken/Qwen3.6-35B-A3B        # On-device via FreeToken
 ```
 
-On-device via [FreeToken](https://github.com/FlashML-org/FreeToken) runs fully local,
-no API key. Serve a model, then use the `freetoken/` prefix; the endpoint
-(`http://127.0.0.1:1919/v1`) and key are set for you:
-
-```bash
-uv pip install "freetoken[accel]"
-ft serve --model Qwen3.6-35B-A3B          # OpenAI-compatible server on 127.0.0.1:1919
-
-uv run skydiscover-run initial_program.py evaluator.py \
-  --model freetoken/Qwen3.6-35B-A3B --search evox --iterations 100
-```
-
-Add `--api-base http://<host>:<port>/v1` if the server runs elsewhere.
-
 ## Pick an algorithm
 
 | Algorithm | Flag | Best for |

skydiscover/config.py

@@ -34,14 +34,8 @@
     "huggingface": (None, ["HF_TOKEN", "HUGGINGFACE_API_KEY"]),
     "ollama": (None, []),
     "vllm": (None, []),
-    "freetoken": ("http://127.0.0.1:1919/v1", []),
 }
 
-# Local OpenAI-compatible servers need no real key, but the OpenAI client still
-# requires a non-empty string, so keyless providers get a harmless placeholder.
-_LOCAL_PROVIDERS = frozenset({"ollama", "vllm", "freetoken"})
-_LOCAL_PLACEHOLDER_KEY = "EMPTY"
-
 # Bare model-name prefixes → provider  (backwards compat for --model gpt-5, etc.)
 _BARE_PREFIX_MAP: Dict[str, str] = {
     "gpt-": "openai",
@@ -84,21 +78,16 @@ def _parse_model_spec(model_str: str) -> tuple:
     return None, model_str, None, []
 
 
-def _resolve_api_key_from_env(
-    env_vars: Optional[List[str]] = None, provider: Optional[str] = None
-) -> Optional[str]:
+def _resolve_api_key_from_env(env_vars: Optional[List[str]] = None) -> Optional[str]:
     """Return the first API key found in *env_vars*.
 
     *env_vars* typically comes from ``_parse_model_spec()``.
-    Only returns a key if it matches the provider's own env vars. Keyless local
-    providers fall back to a placeholder so the OpenAI client accepts them.
+    Only returns a key if it matches the provider's own env vars.
     """
     for var in env_vars or []:
         key = os.environ.get(var)
         if key:
             return key
-    if provider in _LOCAL_PROVIDERS:
-        return _LOCAL_PLACEHOLDER_KEY
     return None
 
 
@@ -212,7 +201,7 @@ def __post_init__(self):
                 if provider_base and not (user_set_api_base and provider == "openai"):
                     model.api_base = provider_base
                 if model.api_key is None:
-                    model.api_key = _resolve_api_key_from_env(env_vars, provider)
+                    model.api_key = _resolve_api_key_from_env(env_vars)
                 # Strip provider prefix so the API receives the bare model name
                 if "/" in model.name and provider != "openai":
                     model.name = bare_name
@@ -937,7 +926,7 @@ def apply_overrides(
                     f"Provider '{provider}' requires an explicit api_base.\n"
                     f"Example: model='{spec}', api_base='http://localhost:8000/v1'"
                 )
-            resolved_key = _resolve_api_key_from_env(env_vars, provider)
+            resolved_key = _resolve_api_key_from_env(env_vars)
             models.append(
                 LLMModelConfig(
                     name=model_name,

tests/config/test_provider_defaults.py

@@ -192,44 +192,6 @@ def test_multi_provider_each_gets_own_endpoint(self):
         assert "anthropic.com" in cfg.models[2].api_base
 
 
-# ── FreeToken / keyless local providers ────────────────────────────
-
-
-class TestFreeTokenProvider:
-    def test_parse_defaults_to_local_endpoint(self):
-        provider, name, api_base, env_vars = _parse_model_spec("freetoken/Qwen3.6-35B-A3B")
-        assert provider == "freetoken"
-        assert name == "Qwen3.6-35B-A3B"
-        assert api_base == "http://127.0.0.1:1919/v1"
-        assert env_vars == []
-
-    def test_keyless_provider_gets_placeholder_key(self):
-        assert _resolve_api_key_from_env([], "freetoken") == "EMPTY"
-        assert _resolve_api_key_from_env([], "ollama") == "EMPTY"
-        assert _resolve_api_key_from_env([], "vllm") == "EMPTY"
-
-    def test_unknown_provider_still_returns_none(self):
-        assert _resolve_api_key_from_env([], None) is None
-        assert _resolve_api_key_from_env([], "mycompany") is None
-
-    def test_config_works_with_no_env_or_flags(self):
-        """freetoken/<model> resolves endpoint, bare name, and placeholder key off-the-shelf."""
-        env = os.environ.copy()
-        env.pop("OPENAI_API_KEY", None)
-        with patch.dict(os.environ, env, clear=True):
-            cfg = LLMConfig(models=[LLMModelConfig(name="freetoken/Qwen3.6-35B-A3B")])
-        model = cfg.models[0]
-        assert model.name == "Qwen3.6-35B-A3B"
-        assert model.api_base == "http://127.0.0.1:1919/v1"
-        assert model.api_key == "EMPTY"
-
-    def test_custom_api_base_preserved(self):
-        cfg = LLMConfig(
-            models=[LLMModelConfig(name="freetoken/Qwen3.6-35B-A3B", api_base="http://gpu-box:9000/v1")],
-        )
-        assert cfg.models[0].api_base == "http://gpu-box:9000/v1"
-
-
 # ── MonitorConfig defaults ─────────────────────────────────────────