Deploy RustLean v4 GGUF

.env.example

@@ -1,5 +1,5 @@
 HF_TOKEN=hf_xxx
 # MODEL_REPO=AfkaraLP/rustlean-gguf
-# MODEL_FILE=rustlean-final.Q8_0.gguf
+# MODEL_FILE=rustlean-v4.Q8_0.gguf
 # RUSTLEAN_CTX=8192
 # N_GPU_LAYERS=-1

README.md

@@ -1,5 +1,5 @@
 ---
-title: RustLean FIM Playground
+title: RustLean v4 FIM Playground
 emoji: 🦀
 colorFrom: indigo
 colorTo: red
@@ -13,12 +13,14 @@ models:
 short_description: Cursor-aware Rust FIM code completion
 ---
 
-# RustLean FIM Playground
+# RustLean v4 FIM Playground
 
 Cursor-aware Rust code completion using
 [`AfkaraLP/rustlean-gguf`](https://huggingface.co/AfkaraLP/rustlean-gguf) — a
 Qwen2.5-Coder-1.5B model fine-tuned for **fill-in-the-middle (FIM)** on Rust,
-served via `llama-cpp-python`.
+served via `llama-cpp-python`. RustLean v4 improves private AST exact match
+from the base model's 22.5% to 35.0% while matching its deterministic
+HumanEvalPack-Rust pass rate.
 
 ## How it works
 
@@ -47,7 +49,7 @@ Optional environment overrides (Space Variables or local `.env`):
 | Variable         | Default                    | Meaning                                |
 | ---------------- | -------------------------- | -------------------------------------- |
 | `MODEL_REPO`     | `AfkaraLP/rustlean-gguf`   | HF repo id                             |
-| `MODEL_FILE`     | `rustlean-final.Q8_0.gguf` | GGUF filename                          |
+| `MODEL_FILE`     | `rustlean-v4.Q8_0.gguf`    | GGUF filename                          |
 | `RUSTLEAN_CTX`   | `8192`                     | Context window                         |
 | `N_GPU_LAYERS`   | `-1`                       | `-1` offloads all layers to the GPU    |
 

app.py

@@ -30,7 +30,7 @@ except Exception:
         return deco
 
 MODEL_REPO = os.environ.get("MODEL_REPO", "AfkaraLP/rustlean-gguf")
-MODEL_FILE = os.environ.get("MODEL_FILE", "rustlean-final.Q8_0.gguf")
+MODEL_FILE = os.environ.get("MODEL_FILE", "rustlean-v4.Q8_0.gguf")
 CTX = int(os.environ.get("RUSTLEAN_CTX", "8192"))
 N_GPU_LAYERS = int(os.environ.get("N_GPU_LAYERS", "-1"))
 
@@ -177,12 +177,14 @@ def complete(code, temperature, max_tokens, top_p):
         return "", code, "```\n" + traceback.format_exc() + "\n```"
 
 
-with gr.Blocks(title="RustLean FIM Playground") as demo:
+with gr.Blocks(title="RustLean v4 FIM Playground") as demo:
     gr.Markdown(
-        "# 🦀 RustLean — Rust FIM Completion\n"
+        "# RustLean v4: Rust FIM Completion\n"
         "Fill-in-the-middle completion from "
         "[`AfkaraLP/rustlean-gguf`](https://huggingface.co/AfkaraLP/rustlean-gguf) "
-        "(Qwen2.5-Coder-1.5B, Q8_0). Put **`<|cursor|>`** where you want the model to "
+        "(Qwen2.5-Coder-1.5B, Q8_0). The v4 release improves repository-disjoint "
+        "Rust completion while retaining the base model's compiler-tested pass rate. "
+        "Put **`<|cursor|>`** where you want the model to "
         "write code. Everything before it is the prefix, everything after is the suffix."
     )