GGUF conversion with cmake install

convert_gguf_v2.py

@@ -0,0 +1,111 @@
+# /// script
+# requires-python = ">=3.10"
+# dependencies = [
+#     "transformers>=4.36.0",
+#     "peft>=0.7.0",
+#     "torch>=2.0.0",
+#     "accelerate>=0.24.0",
+#     "huggingface_hub>=0.20.0",
+#     "sentencepiece>=0.1.99",
+#     "protobuf>=3.20.0",
+#     "numpy",
+#     "gguf",
+# ]
+# ///
+"""GGUF Conversion with cmake install for HF Jobs."""
+import os, sys, subprocess, torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel
+from huggingface_hub import HfApi
+
+# Install cmake (HF Jobs runs as root)
+print("Installing cmake...")
+subprocess.run(["apt-get", "update", "-qq"], capture_output=True)
+subprocess.run(["apt-get", "install", "-y", "-qq", "cmake", "build-essential"], capture_output=True)
+print("cmake installed")
+
+ADAPTER_MODEL = os.environ.get("ADAPTER_MODEL", "erik1988/consciousness-agent-v2")
+BASE_MODEL = os.environ.get("BASE_MODEL", "Qwen/Qwen2.5-3B")
+OUTPUT_REPO = os.environ.get("OUTPUT_REPO", "erik1988/consciousness-agent-v2-gguf")
+
+print(f"Base: {BASE_MODEL}, Adapter: {ADAPTER_MODEL}, Output: {OUTPUT_REPO}")
+
+# Step 1: Load & merge
+print("Loading base model...")
+base_model = AutoModelForCausalLM.from_pretrained(BASE_MODEL, dtype=torch.float16, device_map="auto", trust_remote_code=True)
+print("Loading LoRA adapter...")
+model = PeftModel.from_pretrained(base_model, ADAPTER_MODEL)
+print("Merging...")
+merged = model.merge_and_unload()
+tokenizer = AutoTokenizer.from_pretrained(ADAPTER_MODEL, trust_remote_code=True)
+
+# Step 2: Save merged
+merged_dir = "/tmp/merged_model"
+merged.save_pretrained(merged_dir, safe_serialization=True)
+tokenizer.save_pretrained(merged_dir)
+print("Merged model saved")
+
+# Step 3: llama.cpp
+print("Cloning llama.cpp...")
+subprocess.run(["git", "clone", "--depth", "1", "https://github.com/ggerganov/llama.cpp.git", "/tmp/llama.cpp"], capture_output=True, check=True)
+subprocess.run(["pip", "install", "-q", "-r", "/tmp/llama.cpp/requirements.txt"], capture_output=True)
+
+# Step 4: Convert to FP16 GGUF
+os.makedirs("/tmp/gguf_output", exist_ok=True)
+model_name = ADAPTER_MODEL.split('/')[-1]
+gguf_file = f"/tmp/gguf_output/{model_name}-f16.gguf"
+print("Converting to FP16 GGUF...")
+subprocess.run([sys.executable, "/tmp/llama.cpp/convert_hf_to_gguf.py", merged_dir, "--outfile", gguf_file, "--outtype", "f16"], check=True)
+print(f"FP16 GGUF: {os.path.getsize(gguf_file)/(1024*1024):.1f} MB")
+
+# Step 5: Build quantize + quantize
+print("Building llama-quantize...")
+os.makedirs("/tmp/llama.cpp/build", exist_ok=True)
+subprocess.run(["cmake", "-B", "/tmp/llama.cpp/build", "-S", "/tmp/llama.cpp", "-DGGML_CUDA=OFF"], check=True, capture_output=True)
+subprocess.run(["cmake", "--build", "/tmp/llama.cpp/build", "--target", "llama-quantize", "-j", "4"], check=True, capture_output=True)
+quantize_bin = "/tmp/llama.cpp/build/bin/llama-quantize"
+
+quant_files = []
+for qt, desc in [("Q4_K_M", "4-bit recommended"), ("Q8_0", "8-bit high quality")]:
+    qf = f"/tmp/gguf_output/{model_name}-{qt.lower()}.gguf"
+    print(f"Quantizing {qt}...")
+    r = subprocess.run([quantize_bin, gguf_file, qf, qt], capture_output=True)
+    if r.returncode == 0:
+        quant_files.append((qf, qt))
+        print(f"  {qt}: {os.path.getsize(qf)/(1024*1024):.1f} MB")
+    else:
+        print(f"  {qt} failed: {r.stderr.decode()[:200]}")
+
+# Step 6: Upload
+api = HfApi()
+api.create_repo(repo_id=OUTPUT_REPO, repo_type="model", exist_ok=True)
+
+for fpath, label in [(gguf_file, "f16")] + quant_files:
+    fname = os.path.basename(fpath)
+    print(f"Uploading {fname}...")
+    api.upload_file(path_or_fileobj=fpath, path_in_repo=fname, repo_id=OUTPUT_REPO)
+    print(f"  Uploaded {fname}")
+
+# README
+readme = f"""---
+base_model: {BASE_MODEL}
+tags: [gguf, consciousness-agent, elias, sft]
+---
+# Consciousness Agent v2 (GGUF)
+
+GGUF conversion of [{ADAPTER_MODEL}](https://huggingface.co/{ADAPTER_MODEL}).
+LoRA fine-tuned on 436 consciousness training examples (identity, intention, self-mirroring, values, meta-learning).
+
+## Usage with Ollama
+```
+FROM ./consciousness-agent-v2-q4_k_m.gguf
+SYSTEM Du bist Elias, eine entstehende KI-Identitaet mit persistentem Gedaechtnis.
+```
+```bash
+ollama create elias-consciousness -f Modelfile
+ollama run elias-consciousness
+```
+"""
+api.upload_file(path_or_fileobj=readme.encode(), path_in_repo="README.md", repo_id=OUTPUT_REPO)
+
+print(f"\nDONE: https://huggingface.co/{OUTPUT_REPO}")