Enhance GPU probing and performance tuning in provisioning

- Introduced GPU probing functionality to gather and store GPU specifications in `/mnt/swarm_data/.gpu-rent-gpu.json`, aiding in performance tuning.
- Updated `install_ollama.sh` and `install_llamacpp.sh` to utilize GPU information for configuring optimal runtime parameters.
- Enhanced `provision.py` to include GPU probing and performance tuning logic, ensuring better resource allocation for LLM operations.
- Improved documentation in `decisions.md`, `llm.md`, and `swarmui.md` to reflect changes in GPU handling and performance tuning processes.
- Added new tests to validate the GPU probing and model resolution logic, ensuring robustness in handling various GPU configurations.

src/gpu_rent/provision.py

@@ -35,6 +35,50 @@ def _pkg_text(name: str) -> str:
     return files("gpu_rent.remote").joinpath(name).read_text(encoding="utf-8")
 
 
+def probe_gpu(cfg: Config, host: str, log: Log) -> dict:
+    """Write /mnt/swarm_data/.gpu-rent-gpu.json; return parsed dict."""
+    out = run_python(
+        cfg,
+        host,
+        _pkg_text("gpu_probe.py"),
+        remote_path="/tmp/gpu-rent-gpu_probe.py",
+        timeout=60,
+        log=log,
+    )
+    # Last JSON line from script stdout
+    data: dict = {}
+    for line in reversed(out.splitlines()):
+        line = line.strip()
+        if line.startswith("{"):
+            try:
+                data = json.loads(line)
+                break
+            except json.JSONDecodeError:
+                continue
+    if data.get("ok"):
+        log(
+            f"GPU: {data.get('name')} "
+            f"{data.get('vram_mib')} MiB cap={data.get('compute_cap')} "
+            f"tier={data.get('tier')}"
+        )
+    else:
+        log(f"GPU probe: {data.get('error') or 'нет данных'}")
+    return data
+
+
+def tune_swarm_perf(cfg: Config, host: str, log: Log) -> bool:
+    """Install sage/triton into Comfy venv + ExtraArgs. Returns True if SwarmUI restart needed."""
+    out = run_python(
+        cfg,
+        host,
+        _pkg_text("tune_swarm_perf.py"),
+        remote_path="/tmp/gpu-rent-tune_swarm_perf.py",
+        timeout=900,
+        log=log,
+    )
+    return "RESTART_SWARMUI=1" in out
+
+
 def seed_extensions(cfg: Config, host: str, log: Log, *, update: bool = True) -> bool:
     from gpu_rent.llm_runtime import normalize_runtime
 
@@ -372,6 +416,12 @@ def provision_vm(
     push_tree(cfg, host, cfg.local_workflows_dir, f"{DATA}/CustomWorkflows", log, models=False)
     if cfg.pull_output:
         pull_tree(cfg, host, f"{DATA}/Output", cfg.local_output_dir, log)
+
+    try:
+        probe_gpu(cfg, host, log)
+    except Exception as exc:
+        log(f"GPU probe: {exc}")
+
     ensure_swarmui_running(cfg, host, log, restart=restart)
 
     from gpu_rent.state import load_state, save_state