Refactor LLM configuration to remove llamacpp support

- Removed references to llamacpp from configuration files, scripts, and documentation, streamlining the LLM setup process to focus solely on Ollama.
- Updated environment variables and paths to eliminate llamacpp-related entries, ensuring clarity in the configuration.
- Adjusted CLI commands and help messages to reflect the removal of llamacpp, enhancing user experience and reducing confusion.
- Revised documentation to provide clear guidance on using Ollama exclusively, including updates to setup instructions and runtime options.

src/gpu_rent/provision.py

@@ -31,19 +31,6 @@ DATA = "/mnt/swarm_data"
 
 # Forwarded to remote install_*.sh (from .env / gpu-rent.vars → os.environ).
 _OLLAMA_INSTALL_ENV = ("OLLAMA_VERSION", "OLLAMA_SHA256")
-_LLAMACPP_INSTALL_ENV = (
-    "LLAMACPP_TAG",
-    "LLAMACPP_ASSET_URL",
-    "LLAMACPP_SHA256",
-    "LLAMACPP_BUILD_CUDA",
-    "LLAMACPP_BACKEND",
-    "LLAMACPP_FORCE_REINSTALL",
-    "LLAMACPP_NGL",
-    "LLAMACPP_CTX",
-    "LLAMACPP_HOST",
-    "LLAMACPP_PORT",
-    "LLAMACPP_EXTRA_ARGS",
-)
 
 
 def _remote_llm_env(cfg: Config, *keys: str) -> dict[str, str]:
@@ -450,7 +437,7 @@ def provision_llm(cfg: Config, host: str, log: Log) -> None:
                 check=False,
             )
 
-    # Always drop the other runtime so VRAM is not held by a leftover unit.
+    # Drop LLM units that should not hold VRAM for this runtime.
     if runtime == "none":
         log("LLM: none — останавливаю gpu-rent-ollama / gpu-rent-llamacpp если были")
         _stop_units("gpu-rent-ollama", "gpu-rent-llamacpp")
@@ -490,71 +477,8 @@ def provision_llm(cfg: Config, host: str, log: Log) -> None:
                 timeout=7200,
                 log=log,
             )
-    elif runtime == "llamacpp":
-        _stop_units("gpu-rent-ollama")
-        from gpu_rent.llm_runtime import (
-            gguf_filename_from_url,
-            parse_llamacpp_models,
-        )
-
-        entries = parse_llamacpp_models(cfg.llamacpp_models_manifest)
-        defaults = [e for e in entries if e.default]
-        if defaults:
-            log(
-                "llama.cpp preferred: "
-                f"{defaults[0].filename or gguf_filename_from_url(defaults[0].url)}"
-            )
-        if entries:
-            jobs = []
-            for e in entries:
-                jobs.append(
-                    {
-                        "url": e.url,
-                        "filename": e.filename or gguf_filename_from_url(e.url),
-                    }
-                )
-                if e.mmproj_url:
-                    jobs.append(
-                        {
-                            "url": e.mmproj_url,
-                            "filename": gguf_filename_from_url(e.mmproj_url),
-                        }
-                    )
-            put_text(
-                cfg, host, "/tmp/gpu-rent-llamacpp-models.json", json.dumps(jobs, indent=2)
-            )
-            hf = (cfg.hf_token or "").strip()
-            if hf:
-                put_text(cfg, host, "/tmp/gpu-rent-hf.token", hf + "\n", mode=0o600)
-            else:
-                log(
-                    "⚠ HF_TOKEN не задан — gated GGUF (abliterated и др.) часто дают 401. "
-                    "Добавь HF_TOKEN=hf_… в .env → https://huggingface.co/settings/tokens"
-                )
-            log(f"llama.cpp: скачиваю {len(jobs)} GGUF из манифеста")
-            run_python(
-                cfg,
-                host,
-                _pkg_text("llamacpp_fetch.py"),
-                remote_path="/tmp/gpu-rent-llamacpp_fetch.py",
-                timeout=7200,
-                log=log,
-            )
-        else:
-            log("llamacpp-models.yaml пуст — GGUF skip (положи вручную)")
-        log("LLM: ставим/запускаем llama.cpp server")
-        import os
-
-        # Vulkan finishes in seconds; CUDA compile needs up to ~15–20 min.
-        run_script_sudo(
-            cfg,
-            host,
-            _pkg_text("install_llamacpp.sh"),
-            remote_path="/tmp/gpu-rent-install_llamacpp.sh",
-            timeout=3600,
-            env=_remote_llm_env(cfg, *_LLAMACPP_INSTALL_ENV),
-            log=log,
-        )
+    else:
+        raise CloudError(f"неизвестный LLM_RUNTIME={runtime!r}")
     st = load_state()
     st.notes = dict(st.notes or {})
     st.notes["llm_runtime"] = runtime
@@ -614,7 +538,7 @@ def provision_vm(
     rt = normalize_runtime(cfg.llm_runtime)
     if not swarm and rt == "none":
         raise CloudError(
-            "llm-only: нужен LLM_RUNTIME=ollama|llamacpp (или --ollama / --llamacpp)"
+            "llm-only: нужен LLM_RUNTIME=ollama (или --ollama / --llm ollama)"
         )
 
     # Arm ASAP so mid-provision failures still leave auto-stop on the VM.
@@ -692,6 +616,4 @@ def provision_vm(
         log("SwarmUI слушает 127.0.0.1:7801 — gpu-rent tunnel")
     if rt == "ollama":
         log(f"Ollama API → localhost:{cfg.ollama_local_port} (туннель)")
-    elif rt == "llamacpp":
-        log(f"llama.cpp → localhost:{cfg.llamacpp_local_port} (туннель)")
     log("Hold killer: gpu-rent hold   |  Стоп GPU: gpu-rent stop")