From bc69dfad086cbe89c1441f859b9790b465a00cb4 Mon Sep 17 00:00:00 2001
From: Daniel Han <danielhanchen@gmail.com>
Date: Wed, 1 Jul 2026 00:49:23 -0700
Subject: [PATCH] MLX CI: find llama-cli where save_pretrained_gguf actually
 installs it (#6777)

* MLX CI: find llama-cli where save_pretrained_gguf actually installs it

The GGUF reload step hardcoded the CWD-relative paths llama.cpp/llama-cli and
llama.cpp/build/bin/llama-cli, but save_pretrained_gguf builds and installs llama.cpp
under unsloth_zoo's LLAMA_CPP_DEFAULT_DIR ($UNSLOTH_LLAMA_CPP_PATH, else
~/.unsloth/llama.cpp), so the reload could not find the binary and failed the Mac M1
job with "llama-cli not found". _find_llama_cli now searches that install directory
(and honors the env override) before falling back to the old CWD layout, with a
recursive glob as a last resort. The search is a strict superset of the previous
paths, so it cannot regress a layout that already worked.

* MLX CI: return an absolute llama-cli path from the locator

Resolve the located binary to an absolute path. If UNSLOTH_LLAMA_CPP_PATH is a
relative directory (e.g. "."), Path(".") / "llama-cli" normalizes to the bare name
"llama-cli", and subprocess.run treats a separator-less argument as a PATH lookup
rather than a file to execute, raising FileNotFoundError. resolve() makes the returned
path absolute so it always runs the intended binary.

* MLX CI: give llama-cli EOF on stdin so GGUF reload cannot hang

With the binary now found, the GGUF reload actually invokes llama-cli and it timed
out after 300s generating 24 tokens on a 270m model, which is a stdin block rather
than slow generation: subprocess.run captured stdout/stderr but left stdin inherited,
so -no-cnv still left llama-cli waiting for interactive input. Pass
stdin=subprocess.DEVNULL so it receives an immediate EOF and runs the single prompt to
completion.

---------

Co-authored-by: danielhanchen <michaelhan2050@gmail.com>
---
 tests/studio/run_real_mlx_smoke.py | 51 ++++++++++++++++++++++++++----
 1 file changed, 45 insertions(+), 6 deletions(-)

diff --git a/tests/studio/run_real_mlx_smoke.py b/tests/studio/run_real_mlx_smoke.py
index ea8b8621a9..0843a2b447 100644
--- a/tests/studio/run_real_mlx_smoke.py
+++ b/tests/studio/run_real_mlx_smoke.py
@@ -515,14 +515,50 @@ def cmd_reload(args) -> int:
     return 0
 
 
+def _find_llama_cli() -> Path | None:
+    """Locate the llama-cli binary save_pretrained_gguf built.
+
+    save_pretrained_gguf installs llama.cpp under unsloth_zoo's LLAMA_CPP_DEFAULT_DIR
+    ($UNSLOTH_LLAMA_CPP_PATH or ~/.unsloth/llama.cpp), not the working directory, so
+    search there first and keep the CWD-relative layout as a fallback.
+    """
+    bases: list[Path] = []
+    env_dir = os.environ.get("UNSLOTH_LLAMA_CPP_PATH")
+    if env_dir:
+        bases.append(Path(env_dir))
+    try:
+        from unsloth_zoo.llama_cpp import LLAMA_CPP_DEFAULT_DIR
+        bases.append(Path(LLAMA_CPP_DEFAULT_DIR))
+    except Exception:
+        bases.append(Path.home() / ".unsloth" / "llama.cpp")
+    bases.append(Path("llama.cpp"))
+
+    seen: set[Path] = set()
+    for base in bases:
+        if base in seen:
+            continue
+        seen.add(base)
+        for rel in ("llama-cli", "build/bin/llama-cli"):
+            cand = base / rel
+            if cand.is_file() and os.access(cand, os.X_OK):
+                # Absolute: a separator-less relative path would send subprocess
+                # to a PATH lookup instead of running the file.
+                return cand.resolve()
+        # Last resort: the binary may sit under an unexpected build subdir.
+        if base.is_dir():
+            for cand in sorted(base.glob("**/llama-cli")):
+                if cand.is_file() and os.access(cand, os.X_OK):
+                    return cand.resolve()
+    return None
+
+
 def _reload_gguf(save_dir: Path, metrics: dict) -> int:
-    candidates = [
-        Path("llama.cpp/llama-cli"),
-        Path("llama.cpp/build/bin/llama-cli"),
-    ]
-    llama_cli = next((c for c in candidates if c.exists()), None)
+    llama_cli = _find_llama_cli()
     if llama_cli is None:
-        raise SystemExit(f"llama-cli not found; checked {candidates}")
+        raise SystemExit(
+            "llama-cli not found under $UNSLOTH_LLAMA_CPP_PATH, "
+            "~/.unsloth/llama.cpp, or ./llama.cpp"
+        )
 
     gguf_files = sorted(save_dir.glob("*.gguf"))
     if not gguf_files:
@@ -549,6 +585,9 @@ def _reload_gguf(save_dir: Path, metrics: dict) -> int:
             capture_output = True,
             text = True,
             timeout = 300,
+            # Hand llama-cli an immediate EOF; without it -no-cnv can still leave the
+            # process blocked reading stdin, which times out instead of generating.
+            stdin = subprocess.DEVNULL,
         )
 
     metrics["llama_cli_returncode"] = proc.returncode