Upload convert_survival_gguf.py with huggingface_hub

convert_survival_gguf.py

@@ -1,6 +1,5 @@
-
 # /// script
-# dependencies = ["peft", "transformers", "torch", "huggingface_hub", "sentencepiece"]
+# dependencies = ["peft", "transformers", "torch", "huggingface_hub", "sentencepiece", "cmake"]
 # ///
 
 import os
@@ -36,17 +35,15 @@ tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL_ID)
 tokenizer.save_pretrained(MERGED_DIR)
 
 print("Cloning llama.cpp...")
+if os.path.exists("llama.cpp"):
+    subprocess.run(["rm", "-rf", "llama.cpp"])
 subprocess.run(["git", "clone", "https://github.com/ggerganov/llama.cpp"], check=True)
 
 print("Installing llama.cpp requirements...")
 subprocess.run(["pip", "install", "-r", "llama.cpp/requirements.txt"], check=True)
 
-print("Converting to GGUF (Q4_K_M)...")
-# Note: Newer llama.cpp uses convert_hf_to_gguf.py
-# We perform quantization in two steps: convert to fp16 gguf, then quantize
-# Or if convert script supports outtype...
-
-# Step 1: Convert to FP16 GGUF
+print("Converting to GGUF (FP16)...")
+# Convert to FP16 GGUF using the python script
 subprocess.run([
     "python", "llama.cpp/convert_hf_to_gguf.py",
     MERGED_DIR,
@@ -54,10 +51,22 @@ subprocess.run([
     "--outtype", "f16"
 ], check=True)
 
-# Step 2: Quantize to Q4_K_M
-subprocess.run(["make", "-C", "llama.cpp", "llama-quantize"], check=True)
+print("Building llama-quantize with CMake...")
+# Create build directory
+os.makedirs("llama.cpp/build", exist_ok=True)
+
+# Run cmake configuration
+subprocess.run(["cmake", "-B", "llama.cpp/build", "-S", "llama.cpp"], check=True)
+
+# Build the project
+subprocess.run(["cmake", "--build", "llama.cpp/build", "--config", "Release", "-j"], check=True)
+
+print("Quantizing to Q4_K_M...")
+# The binary is usually in llama.cpp/build/bin/llama-quantize
+quantize_bin = "llama.cpp/build/bin/llama-quantize"
+
 subprocess.run([
-    "./llama.cpp/llama-quantize",
+    quantize_bin,
     "merged_fp16.gguf",
     GGUF_FILE,
     "Q4_K_M"
@@ -75,4 +84,4 @@ api.upload_file(
     repo_type="model"
 )
 
-print("Done! GGUF available at:", f"https://huggingface.co/{OUTPUT_REPO}")
+print("Done! GGUF available at:", f"https://huggingface.co/{OUTPUT_REPO}")