Update Sol architecture implementation

modeling_sol_lite.py

@@ -199,7 +199,7 @@ class ReleasedSolLiteControl(nn.Module):
         return F.linear(self.norm(x), self.embedding.weight)
 
 
-class SolLiteForCausalLM(nn.Module):
+class SolForCausalLM(nn.Module):
     """XSA + selective recurrence + loop conditioning + EngramLite."""
 
     def __init__(self, cfg: SolLiteConfig = SolLiteConfig()):
@@ -257,7 +257,7 @@ def variant_config(name: str) -> SolLiteConfig:
 def build_variant(name: str) -> nn.Module:
     if name == "released_control":
         return ReleasedSolLiteControl()
-    return SolLiteForCausalLM(variant_config(name))
+    return SolForCausalLM(variant_config(name))
 
 
 def architecture_audit() -> dict[str, int | float | bool]:
@@ -290,7 +290,7 @@ def load_model(model_dir: str | Path, device: str | torch.device = "cpu"):
     raw = json.loads((model_dir / "config.json").read_text())
     fields = SolLiteConfig.__dataclass_fields__
     cfg = SolLiteConfig(**{key: value for key, value in raw.items() if key in fields})
-    model = SolLiteForCausalLM(cfg)
+    model = SolForCausalLM(cfg)
     model.load_state_dict(load_file(model_dir / "model.safetensors"), strict=True)
     model.to(device).eval()
     tokenizer = AutoTokenizer.from_pretrained(model_dir)
@@ -299,7 +299,7 @@ def load_model(model_dir: str | Path, device: str | torch.device = "cpu"):
 
 @torch.inference_mode()
 def generate(
-    model: SolLiteForCausalLM,
+    model: SolForCausalLM,
     tokenizer,
     prompt: str,
     max_new_tokens: int = 64,