Upload COLAB_MERGE_FULL_GGUF.md with huggingface_hub

COLAB_MERGE_FULL_GGUF.md

@@ -0,0 +1,125 @@
+# Create the full standalone Gemopus v1 E2B GGUF in Colab
+
+This creates:
+
+```text
+Gemopus-v1-e2b-Q4_K_M-merged.gguf
+```
+
+That file is the full standalone LM Studio model:
+
+```text
+Gemma 4 E2B Q4_K_M base + Gemopus checkpoint-400 LoRA
+```
+
+## Colab Cell
+
+Run this in Google Colab. Use a runtime with enough disk space. GPU is not required for the merge, but a T4 runtime is fine.
+
+Before running, replace:
+
+```text
+PASTE_YOUR_HF_TOKEN_HERE
+```
+
+with a Hugging Face write token.
+
+```python
+import os
+import subprocess
+from pathlib import Path
+
+HF_TOKEN = "PASTE_YOUR_HF_TOKEN_HERE"
+REPO_ID = "Hashhasapi/Gemopus-v1-e2b"
+BASE_REPO = "lmstudio-community/gemma-4-E2B-it-GGUF"
+BASE_FILE = "gemma-4-E2B-it-Q4_K_M.gguf"
+LORA_FILE = "Gemopus-v1-e2b-lora.gguf"
+MERGED_FILE = "Gemopus-v1-e2b-Q4_K_M-merged.gguf"
+
+os.environ["HF_TOKEN"] = HF_TOKEN
+
+def run(cmd, cwd=None):
+    print("\n$", " ".join(cmd) if isinstance(cmd, list) else cmd)
+    subprocess.run(cmd, cwd=cwd, shell=isinstance(cmd, str), check=True)
+
+run("apt-get update")
+run("apt-get install -y --no-install-recommends git cmake build-essential ca-certificates python3-pip")
+run("python3 -m pip install -U huggingface_hub")
+
+Path("/content/base").mkdir(exist_ok=True)
+Path("/content/lora").mkdir(exist_ok=True)
+
+run([
+    "hf", "download", BASE_REPO, BASE_FILE,
+    "--local-dir", "/content/base",
+    "--token", HF_TOKEN,
+])
+
+run([
+    "hf", "download", REPO_ID, LORA_FILE,
+    "--local-dir", "/content/lora",
+    "--token", HF_TOKEN,
+])
+
+if not Path("/content/llama.cpp").exists():
+    run(["git", "clone", "--depth", "1", "https://github.com/ggml-org/llama.cpp", "/content/llama.cpp"])
+
+run([
+    "cmake",
+    "-S", "/content/llama.cpp",
+    "-B", "/content/llama.cpp/build",
+    "-DGGML_CUDA=OFF",
+    "-DLLAMA_CURL=OFF",
+    "-DLLAMA_BUILD_TESTS=OFF",
+    "-DLLAMA_BUILD_EXAMPLES=OFF",
+    "-DLLAMA_BUILD_TOOLS=ON",
+])
+
+run([
+    "cmake",
+    "--build", "/content/llama.cpp/build",
+    "--config", "Release",
+    "-j", str(os.cpu_count() or 2),
+    "--target", "llama-export-lora",
+])
+
+export_bin = Path("/content/llama.cpp/build/bin/llama-export-lora")
+if not export_bin.exists():
+    matches = list(Path("/content/llama.cpp/build").rglob("llama-export-lora*"))
+    if not matches:
+        raise FileNotFoundError("Could not find llama-export-lora after build.")
+    export_bin = matches[0]
+
+run([
+    str(export_bin),
+    "-m", f"/content/base/{BASE_FILE}",
+    "--lora", f"/content/lora/{LORA_FILE}",
+    "-o", f"/content/{MERGED_FILE}",
+    "-t", str(os.cpu_count() or 2),
+])
+
+merged = Path(f"/content/{MERGED_FILE}")
+print("Merged file size:", merged.stat().st_size / (1024**3), "GiB")
+
+run([
+    "hf", "upload", REPO_ID,
+    str(merged),
+    MERGED_FILE,
+    "--repo-type", "model",
+    "--token", HF_TOKEN,
+    "--commit-message", "Add merged standalone Gemopus v1 E2B GGUF",
+])
+
+print("Done:")
+print(f"https://huggingface.co/{REPO_ID}/blob/main/{MERGED_FILE}")
+```
+
+## After Upload
+
+In LM Studio, download/load:
+
+```text
+Gemopus-v1-e2b-Q4_K_M-merged.gguf
+```
+
+That one should load directly without a LoRA adapter.