New python based entry point for containers

Signed-off-by: Jiri Podivin <[email protected]>

Author: jpodivin

.devops/full.Dockerfile

@@ -16,4 +16,4 @@ COPY . .
 
 RUN make
 
-ENTRYPOINT ["/app/.devops/tools.sh"]
+ENTRYPOINT ["/app/.devops/tools.py"]

.devops/tools.py

@@ -0,0 +1,47 @@
+#!/bin/env python3
+
+import argparse
+import os
+import subprocess as sp
+import glob
+
+parser = argparse.ArgumentParser()
+group = parser.add_mutually_exclusive_group()
+group.add_argument("--run", "-r", action='store_true', help="Run a model previously converted into ggml")
+group.add_argument("--convert", "-c", action='store_true', help="Convert a llama model into ggml")
+group.add_argument("--quantize", "-q", action='store_true', help="Optimize with quantization process ggml")
+group.add_argument("--all-in-one", "-a", action='store_true', help="Execute --convert & --quantize")
+parser.add_argument("model", type=str, help="Directory containing model file, or model file itself (*.pth, *.pt, *.bin)")
+parser.add_argument("--quant-method", "-Q", type=str, help="Chosen quantization method.", default="q4_0")
+
+known_args, unknown_args = parser.parse_known_args()
+
+model_path = known_args.model
+converted_models = glob.glob(os.path.join(model_path, 'ggml-model-*.bin'))
+quantized_models = glob.glob(os.path.join(model_path, f'ggml-model-{known_args.quant_method}.bin'))
+
+if known_args.convert:
+    sp.run(['python3', './convert.py', model_path] + unknown_args, check=True)
+
+if known_args.run:
+    sp.run(['./main', '-m', model_path] + unknown_args, check=True)
+
+if known_args.quantize:
+    if not converted_models:
+        print(f"No models ready for quantization found in {model_path}")
+        exit(1)
+    sp.run(['./quantize', converted_models[0], known_args.quant_method] + unknown_args, check=True)
+
+if known_args.all_in_one:
+    if not converted_models:
+        sp.run(['python3', './convert.py', model_path], check=True)
+        converted_models = glob.glob(os.path.join(model_path, 'ggml-model-*.bin'))
+    else:
+        print(
+            f"Converted models found {converted_models}! No need to convert.")
+    if not quantized_models:
+        sp.run(['./quantize', converted_models[0], known_args.quant_method] + unknown_args, check=True)
+    else:
+        print(
+            f"Quantized models found {quantized_models}! No need to quantize.")
+exit()