*.py: Convert all python scripts to use logging module

Author: mofosyne

convert-hf-to-gguf.py

@@ -1,6 +1,7 @@
 #!/usr/bin/env python3
 from __future__ import annotations
 
+import logging
 import argparse
 import os
 import struct
@@ -14,6 +15,8 @@
     sys.path.insert(1, str(Path(__file__).parent / 'gguf-py'))
 import gguf
 
+logger = logging.getLogger("ggml-to-gguf")
+
 
 class GGMLFormat(IntEnum):
     GGML = 0
@@ -125,7 +128,7 @@ def load(self, data, offset):
         self.start_offset = offset
         self.len_bytes = n_bytes
         offset += n_bytes
-        # print(n_dims, name_len, dtype, self.dims, self.name, pad)
+        # logger.info(n_dims, name_len, dtype, self.dims, self.name, pad)
         return offset - orig_offset
 
 
@@ -175,7 +178,7 @@ def load(self, data, offset):
         offset += self.validate_header(data, offset)
         hp = Hyperparameters()
         offset += hp.load(data, offset)
-        print(f'* File format: {self.file_format.name}v{self.format_version} with ftype {hp.ftype.name}')
+        logger.info(f'* File format: {self.file_format.name}v{self.format_version} with ftype {hp.ftype.name}')
         self.validate_conversion(hp.ftype)
         vocab = Vocab(load_scores = self.file_format > GGMLFormat.GGML)
         offset += vocab.load(data, offset, hp.n_vocab)
@@ -215,12 +218,12 @@ def __init__(self, ggml_model, data, cfg, params_override = None, vocab_override
                     if float(hp.n_head) / float(x) == gqa:
                         n_kv_head = x
                 assert n_kv_head is not None, "Couldn't determine n_kv_head from GQA param"
-                print(f'- Guessed n_kv_head = {n_kv_head} based on GQA {cfg.gqa}')
+                logger.info(f'- Guessed n_kv_head = {n_kv_head} based on GQA {cfg.gqa}')
         self.n_kv_head = n_kv_head
         self.name_map = gguf.get_tensor_name_map(gguf.MODEL_ARCH.LLAMA, ggml_model.hyperparameters.n_layer)
 
     def save(self):
-        print('* Preparing to save GGUF file')
+        logger.info('* Preparing to save GGUF file')
         gguf_writer = gguf.GGUFWriter(
             self.cfg.output,
             gguf.MODEL_ARCH_NAMES[gguf.MODEL_ARCH.LLAMA],
@@ -230,11 +233,11 @@ def save(self):
         if self.special_vocab is not None:
             self.special_vocab.add_to_gguf(gguf_writer)
         self.add_tensors(gguf_writer)
-        print("    gguf: write header")
+        logger.info("    gguf: write header")
         gguf_writer.write_header_to_file()
-        print("    gguf: write metadata")
+        logger.info("    gguf: write metadata")
         gguf_writer.write_kv_data_to_file()
-        print("    gguf: write tensors")
+        logger.info("    gguf: write tensors")
         gguf_writer.write_tensors_to_file()
         gguf_writer.close()
 
@@ -250,7 +253,7 @@ def add_params(self, gguf_writer):
             name = cfg.name if cfg.name is not None else cfg.input.name
         except UnicodeDecodeError:
             name = None
-        print('* Adding model parameters and KV items')
+        logger.info('* Adding model parameters and KV items')
         if name is not None:
             gguf_writer.add_name(name)
         gguf_writer.add_description(desc)
@@ -286,7 +289,7 @@ def add_vocab(self, gguf_writer):
         toktypes = []
         if self.vocab_override is not None:
             vo = self.vocab_override
-            print('* Adding vocab item(s)')
+            logger.info('* Adding vocab item(s)')
             for (idx, (vbytes, score, ttype)) in enumerate(vo.all_tokens()):
                 tokens.append(vbytes)
                 scores.append(score)
@@ -298,7 +301,7 @@ def add_vocab(self, gguf_writer):
             if len(toktypes) > 0:
                 gguf_writer.add_token_types(toktypes)
             return
-        print(f'* Adding {hp.n_vocab} vocab item(s)')
+        logger.info(f'* Adding {hp.n_vocab} vocab item(s)')
         assert len(self.model.vocab.items) >= 3, 'Cannot handle unexpectedly short model vocab'
         for (tokid, (vbytes, vscore)) in enumerate(self.model.vocab.items):
             tt = 1 # Normal
@@ -333,7 +336,7 @@ def add_vocab(self, gguf_writer):
     def add_tensors(self, gguf_writer):
         tensor_map = self.name_map
         data = self.data
-        print(f'* Adding {len(self.model.tensors)} tensor(s)')
+        logger.info(f'* Adding {len(self.model.tensors)} tensor(s)')
         for tensor in self.model.tensors:
             name = str(tensor.name, 'UTF-8')
             mapped_name = tensor_map.get_name(name, try_suffixes = (".weight", ".bias"))
@@ -343,7 +346,7 @@ def add_tensors(self, gguf_writer):
                 temp = tempdims[1]
                 tempdims[1] = tempdims[0]
                 tempdims[0] = temp
-            # print(f'+ {tensor.name} | {mapped_name} {tensor.dims} :: {tempdims}')
+            # logger.info(f'+ {tensor.name} | {mapped_name} {tensor.dims} :: {tempdims}')
             gguf_writer.add_tensor(
                 mapped_name,
                 data[tensor.start_offset:tensor.start_offset + tensor.len_bytes],
@@ -400,41 +403,46 @@ def handle_args():
                         help="directory containing tokenizer.model, if separate from model file - only meaningful with --model-metadata-dir")
     parser.add_argument("--vocabtype", default="spm,hfft",
                         help="vocab format - only meaningful with --model-metadata-dir and/or --vocab-dir (default: spm,hfft)")
+    parser.add_argument("--verbose",      action="store_true",    help="increase output verbosity")
     return parser.parse_args()
 
 
 def main():
     cfg = handle_args()
-    print(f'* Using config: {cfg}')
-    print('\n=== WARNING === Be aware that this conversion script is best-effort. Use a native GGUF model if possible. === WARNING ===\n')
+    if cfg.verbose:
+        logging.basicConfig(level=logging.DEBUG)
+    else:
+        logging.basicConfig(level=logging.INFO)
+    logger.info(f'* Using config: {cfg}')
+    logger.warning('=== WARNING === Be aware that this conversion script is best-effort. Use a native GGUF model if possible. === WARNING ===')
     if cfg.model_metadata_dir is None and (cfg.gqa == 1 or cfg.eps == '5.0e-06'):
-        print('- Note: If converting LLaMA2, specifying "--eps 1e-5" is required. 70B models also need "--gqa 8".')
+        logger.info('- Note: If converting LLaMA2, specifying "--eps 1e-5" is required. 70B models also need "--gqa 8".')
     data = np.memmap(cfg.input, mode = 'r')
     model = GGMLModel()
-    print('* Scanning GGML input file')
+    logger.info('* Scanning GGML input file')
     offset = model.load(data, 0)  # noqa
-    print(f'* GGML model hyperparameters: {model.hyperparameters}')
+    logger.info(f'* GGML model hyperparameters: {model.hyperparameters}')
     vocab_override = None
     params_override = None
     special_vocab = None
     if cfg.model_metadata_dir is not None:
         (params_override, vocab_override, special_vocab) = handle_metadata(cfg, model.hyperparameters)
-        print('!! Note: When overriding params the --gqa, --eps and --context-length options are ignored.')
-        print(f'* Overriding params: {params_override}')
-        print(f'* Overriding vocab: {vocab_override}')
-        print(f'* Special vocab: {special_vocab}')
+        logger.info('!! Note: When overriding params the --gqa, --eps and --context-length options are ignored.')
+        logger.info(f'* Overriding params: {params_override}')
+        logger.info(f'* Overriding vocab: {vocab_override}')
+        logger.info(f'* Special vocab: {special_vocab}')
     else:
-        print('\n=== WARNING === Special tokens may not be converted correctly. Use --model-metadata-dir if possible === WARNING ===\n')
+        logger.warning('\n=== WARNING === Special tokens may not be converted correctly. Use --model-metadata-dir if possible === WARNING ===\n')
         if model.file_format == GGMLFormat.GGML:
-            print('! This is a very old GGML file that does not contain vocab scores. Strongly recommend using model metadata!')
+            logger.info('! This is a very old GGML file that does not contain vocab scores. Strongly recommend using model metadata!')
     converter = GGMLToGGUF(
         model, data, cfg,
         params_override = params_override,
         vocab_override = vocab_override,
         special_vocab = special_vocab
     )
     converter.save()
-    print(f'* Successful completion. Output saved to: {cfg.output}')
+    logger.info(f'* Successful completion. Output saved to: {cfg.output}')
 
 
 if __name__ == '__main__':

convert-llama-ggml-to-gguf.py

@@ -1,6 +1,7 @@
 #!/usr/bin/env python3
 from __future__ import annotations
 
+import logging
 import json
 import os
 import struct
@@ -15,6 +16,8 @@
     sys.path.insert(1, str(Path(__file__).parent / 'gguf-py' / 'gguf'))
 import gguf
 
+logger = logging.getLogger("lora-to-gguf")
+
 NUMPY_TYPE_TO_FTYPE: dict[str, int] = {"float32": 0, "float16": 1}
 
 
@@ -48,11 +51,9 @@ def write_tensor_header(fout: BinaryIO, name: str, shape: Sequence[int], data_ty
 
 if __name__ == '__main__':
     if len(sys.argv) < 2:
-        print(f"Usage: python {sys.argv[0]} <path> [arch]")
-        print(
-            "Path must contain HuggingFace PEFT LoRA files 'adapter_config.json' and 'adapter_model.bin'"
-        )
-        print(f"Arch must be one of {list(gguf.MODEL_ARCH_NAMES.values())} (default: llama)")
+        logger.info(f"Usage: python {sys.argv[0]} <path> [arch]")
+        logger.info("Path must contain HuggingFace PEFT LoRA files 'adapter_config.json' and 'adapter_model.bin'")
+        logger.info(f"Arch must be one of {list(gguf.MODEL_ARCH_NAMES.values())} (default: llama)")
         sys.exit(1)
 
     input_json = os.path.join(sys.argv[1], "adapter_config.json")
@@ -70,7 +71,7 @@ def write_tensor_header(fout: BinaryIO, name: str, shape: Sequence[int], data_ty
     arch_name = sys.argv[2] if len(sys.argv) == 3 else "llama"
 
     if arch_name not in gguf.MODEL_ARCH_NAMES.values():
-        print(f"Error: unsupported architecture {arch_name}")
+        logger.error(f"Error: unsupported architecture {arch_name}")
         sys.exit(1)
 
     arch = list(gguf.MODEL_ARCH_NAMES.keys())[list(gguf.MODEL_ARCH_NAMES.values()).index(arch_name)]
@@ -80,21 +81,21 @@ def write_tensor_header(fout: BinaryIO, name: str, shape: Sequence[int], data_ty
         params = json.load(f)
 
     if params["peft_type"] != "LORA":
-        print(f"Error: unsupported adapter type {params['peft_type']}, expected LORA")
+        logger.error(f"Error: unsupported adapter type {params['peft_type']}, expected LORA")
         sys.exit(1)
 
     if params["fan_in_fan_out"] is True:
-        print("Error: param fan_in_fan_out is not supported")
+        logger.error("Error: param fan_in_fan_out is not supported")
         sys.exit(1)
 
     if params["bias"] is not None and params["bias"] != "none":
-        print("Error: param bias is not supported")
+        logger.error("Error: param bias is not supported")
         sys.exit(1)
 
     # TODO: these seem to be layers that have been trained but without lora.
     # doesn't seem widely used but eventually should be supported
     if params["modules_to_save"] is not None and len(params["modules_to_save"]) > 0:
-        print("Error: param modules_to_save is not supported")
+        logger.error("Error: param modules_to_save is not supported")
         sys.exit(1)
 
     with open(output_path, "wb") as fout:
@@ -125,13 +126,13 @@ def write_tensor_header(fout: BinaryIO, name: str, shape: Sequence[int], data_ty
                 suffix = k[-len(lora_suffixes[0]):]
                 k = k[: -len(lora_suffixes[0])]
             else:
-                print(f"Error: unrecognized tensor name {orig_k}")
+                logger.error(f"Error: unrecognized tensor name {orig_k}")
                 sys.exit(1)
 
             tname = name_map.get_name(k)
             if tname is None:
-                print(f"Error: could not map tensor name {orig_k}")
-                print(" Note: the arch parameter must be specified if the model is not llama")
+                logger.error(f"Error: could not map tensor name {orig_k}")
+                logger.error(" Note: the arch parameter must be specified if the model is not llama")
                 sys.exit(1)
 
             if suffix == ".lora_A.weight":
@@ -141,8 +142,8 @@ def write_tensor_header(fout: BinaryIO, name: str, shape: Sequence[int], data_ty
             else:
                 assert False
 
-            print(f"{k} => {tname} {t.shape} {t.dtype} {t.nbytes/1024/1024:.2f}MB")
+            logger.info(f"{k} => {tname} {t.shape} {t.dtype} {t.nbytes/1024/1024:.2f}MB")
             write_tensor_header(fout, tname, t.shape, t.dtype)
             t.tofile(fout)
 
-    print(f"Converted {input_json} and {input_model} to {output_path}")
+    logger.info(f"Converted {input_json} and {input_model} to {output_path}")

convert-lora-to-ggml.py

@@ -1,6 +1,7 @@
 #!/usr/bin/env python3
 from __future__ import annotations
 
+import logging
 import argparse
 import os
 import sys
@@ -14,6 +15,8 @@
     sys.path.insert(1, str(Path(__file__).parent / 'gguf-py'))
 import gguf
 
+logger = logging.getLogger("persimmon-to-gguf")
+
 
 def _flatten_dict(dct, tensors, prefix=None):
     assert isinstance(dct, dict)
@@ -30,9 +33,9 @@ def _flatten_dict(dct, tensors, prefix=None):
 
 def _get_sentencepiece_tokenizer_info(dir_model: Path):
     tokenizer_path = dir_model / 'adept_vocab.model'
-    print('gguf: getting sentencepiece tokenizer from', tokenizer_path)
+    logger.info('getting sentencepiece tokenizer from', tokenizer_path)
     tokenizer = SentencePieceProcessor(str(tokenizer_path))
-    print('gguf: adding tokens')
+    logger.info('adding tokens')
     tokens: list[bytes] = []
     scores: list[float] = []
     toktypes: list[int] = []
@@ -67,8 +70,13 @@ def main():
     parser.add_argument("--outfile",             type=Path, help="path to write to; default: based on input")
     parser.add_argument("--ckpt-path",           type=Path, help="path to persimmon checkpoint .pt file")
     parser.add_argument("--model-dir",           type=Path, help="directory containing model e.g. 8b_chat_model_release")
-    parser.add_argument("--adept-inference-dir", type=str, help="path to adept-inference code directory")
+    parser.add_argument("--adept-inference-dir", type=str,  help="path to adept-inference code directory")
+    parser.add_argument("--verbose",  action="store_true",  help="increase output verbosity")
     args = parser.parse_args()
+    if args.verbose:
+        logging.basicConfig(level=logging.DEBUG)
+    else:
+        logging.basicConfig(level=logging.INFO)
     sys.path.append(str(args.adept_inference_dir))
     persimmon_model = torch.load(args.ckpt_path)
     hparams = persimmon_model['args']
@@ -106,7 +114,7 @@ def main():
     gguf_writer.add_eos_token_id(71013)
 
     tensor_map = gguf.get_tensor_name_map(arch, block_count)
-    print(tensor_map)
+    logger.info(tensor_map)
     for name in tensors.keys():
         data_torch = tensors[name]
         if name.endswith(".self_attention.rotary_emb.inv_freq"):
@@ -116,22 +124,21 @@ def main():
         data = data_torch.to(torch.float32).squeeze().numpy()
         new_name = tensor_map.get_name(name, try_suffixes = (".weight", ".bias"))
         if new_name is None:
-            print("Can not map tensor '" + name + "'")
+            logger.error(f"Can not map tensor '{name}'")
             sys.exit()
         n_dims = len(data.shape)
-        print(new_name + ", n_dims = " + str(n_dims) + ", " + str(old_dtype) + " --> " + str(data.dtype))
+        logger.debug(f"{new_name}, n_dims = {str(n_dims)}, {str(old_dtype)} --> {str(data.dtype)}")
         gguf_writer.add_tensor(new_name, data)
-    print("gguf: write header")
+    logger.info("gguf: write header")
     gguf_writer.write_header_to_file()
-    print("gguf: write metadata")
+    logger.info("gguf: write metadata")
     gguf_writer.write_kv_data_to_file()
-    print("gguf: write tensors")
+    logger.info("gguf: write tensors")
     gguf_writer.write_tensors_to_file()
 
     gguf_writer.close()
 
-    print(f"gguf: model successfully exported to '{args.outfile}'")
-    print("")
+    logger.info(f"gguf: model successfully exported to '{args.outfile}'")
 
 
 if __name__ == '__main__':

convert-persimmon-to-gguf.py

@@ -1442,11 +1442,11 @@ def default_outfile(model_paths: list[Path], file_type: GGMLFileType) -> Path:
 
 
 def do_dump_model(model_plus: ModelPlus) -> None:
-    print(f"model_plus.paths = {model_plus.paths!r}")
-    print(f"model_plus.format = {model_plus.format!r}")
-    print(f"model_plus.vocab = {model_plus.vocab!r}")
+    print(f"model_plus.paths = {model_plus.paths!r}") # noqa: NP100
+    print(f"model_plus.format = {model_plus.format!r}") # noqa: NP100
+    print(f"model_plus.vocab = {model_plus.vocab!r}") # noqa: NP100
     for name, lazy_tensor in model_plus.model.items():
-        print(f"{name}: shape={lazy_tensor.shape} type={lazy_tensor.data_type}; {lazy_tensor.description}")
+        print(f"{name}: shape={lazy_tensor.shape} type={lazy_tensor.data_type}; {lazy_tensor.description}") # noqa: NP100
 
 
 def main(args_in: list[str] | None = None) -> None: