Enhance LLM runtime and installation script for Linux support

- Added a new function `pick_llamacpp_linux_asset_url` to select appropriate Linux release assets, prioritizing Ubuntu CUDA and Vulkan options while excluding Windows and macOS binaries.
- Updated the installation script to build `llama-server` from source when Linux CUDA binaries are unavailable, improving compatibility and flexibility.
- Revised documentation to reflect changes in asset handling and installation procedures.
- Added tests to validate the new asset selection logic, ensuring correct behavior in various scenarios.

src/gpu_rent/llm_runtime.py

@@ -357,6 +357,42 @@ def llm_remote_port(runtime: str) -> int | None:
     return None
 
 
+def pick_llamacpp_linux_asset_url(assets: list[dict[str, Any]]) -> str:
+    """Choose a Linux llama.cpp release asset URL.
+
+    Upstream ships Windows CUDA zips first; never pick win/macos/cudart-only.
+    Prefer ubuntu+cuda → linux+cuda → ubuntu vulkan x64 → ubuntu x64 CPU.
+    Mirrored in remote/install_llamacpp.sh (pick_linux_asset_url).
+    """
+    cands: list[tuple[int, str]] = []
+    for a in assets:
+        name = str(a.get("name") or "").lower()
+        url = str(a.get("browser_download_url") or "")
+        if not (url.endswith(".zip") or url.endswith(".tar.gz")):
+            continue
+        if any(x in name for x in ("win", "macos", "android", "darwin", "xcframework", "-ui.")):
+            continue
+        if "cudart" in name:
+            continue
+        score = 0
+        if "ubuntu" in name and "x64" in name and "cuda" in name:
+            score = 100
+        elif "linux" in name and "cuda" in name:
+            score = 90
+        elif "ubuntu" in name and "vulkan" in name and "x64" in name:
+            score = 50
+        elif "ubuntu" in name and "x64" in name and not any(
+            x in name for x in ("sycl", "openvino", "arm", "s390", "rocm")
+        ):
+            score = 30
+        elif "ubuntu" in name or "linux" in name:
+            score = 10
+        if score:
+            cands.append((score, url))
+    cands.sort(key=lambda t: t[0], reverse=True)
+    return cands[0][1] if cands else ""
+
+
 def append_vars_llm_runtime(vars_file: Path, runtime: str) -> None:
     from gpu_rent.varsfile import upsert_vars