Respect tokenizer.ggml.add_bos_token value when tokenizing (#4040)

* gguf-py: gguf-dump: Respect --no-tensor flag in JSON mode.

* Respect add_bos_token GGUF metadata value

* gguf-py: Try to fix SpecialVocab giving up too easily for the Nth time

3 files changed, 24 insertions, 18 deletions

diff --git a/gguf-py/gguf/vocab.py b/gguf-py/gguf/vocab.py
index 71192a92..b9f50a0a 100644
--- a/gguf-py/gguf/vocab.py
+++ b/gguf-py/gguf/vocab.py
@@ -117,17 +117,18 @@ class SpecialVocab:
 
     def _try_load_from_tokenizer_json(self, path: Path) -> bool:
         tokenizer_file = path / 'tokenizer.json'
-        if not tokenizer_file.is_file():
-            return False
-        with open(tokenizer_file, encoding = 'utf-8') as f:
-            tokenizer = json.load(f)
-        if self.load_merges:
-            merges = tokenizer.get('model', {}).get('merges')
-            if isinstance(merges, list) and merges and isinstance(merges[0], str):
-                self.merges = merges
+        if tokenizer_file.is_file():
+            with open(tokenizer_file, encoding = 'utf-8') as f:
+                tokenizer = json.load(f)
+            if self.load_merges:
+                merges = tokenizer.get('model', {}).get('merges')
+                if isinstance(merges, list) and merges and isinstance(merges[0], str):
+                    self.merges = merges
+            added_tokens = tokenizer.get('added_tokens', {})
+        else:
+            added_tokens = {}
         tokenizer_config_file = path / 'tokenizer_config.json'
-        added_tokens = tokenizer.get('added_tokens')
-        if added_tokens is None or not tokenizer_config_file.is_file():
+        if not tokenizer_config_file.is_file():
             return True
         with open(tokenizer_config_file, encoding = 'utf-8') as f:
             tokenizer_config = json.load(f)
@@ -135,6 +136,10 @@ class SpecialVocab:
             add_entry = tokenizer_config.get(f'add_{typ}_token')
             if isinstance(add_entry, bool):
                 self.add_special_token[typ] = add_entry
+            if not added_tokens:
+                # We will need this to get the content for the token, so if it's empty
+                # may as well just give up.
+                continue
             entry = tokenizer_config.get(f'{typ}_token')
             if isinstance(entry, str):
                 tc_content = entry
diff --git a/gguf-py/pyproject.toml b/gguf-py/pyproject.toml
index af777c3e..6e3f9e85 100644
--- a/gguf-py/pyproject.toml
+++ b/gguf-py/pyproject.toml
@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "gguf"
-version = "0.5.2"
+version = "0.5.3"
 description = "Read and write ML models in GGUF for GGML"
 authors = ["GGML <ggml@ggml.ai>"]
 packages = [
diff --git a/gguf-py/scripts/gguf-dump.py b/gguf-py/scripts/gguf-dump.py
index 5141873d..dbf89150 100755
--- a/gguf-py/scripts/gguf-dump.py
+++ b/gguf-py/scripts/gguf-dump.py
@@ -86,13 +86,14 @@ def dump_metadata_json(reader: GGUFReader, args: argparse.Namespace) -> None:
             curr["value"] = str(bytes(field.parts[-1]), encoding="utf-8")
         else:
             curr["value"] = field.parts[-1].tolist()[0]
-    for idx, tensor in enumerate(reader.tensors):
-        tensors[tensor.name] = {
-            "index": idx,
-            "shape": tensor.shape.tolist(),
-            "type": tensor.tensor_type.name,
-            "offset": tensor.field.offset,
-        }
+    if not args.no_tensors:
+        for idx, tensor in enumerate(reader.tensors):
+            tensors[tensor.name] = {
+                "index": idx,
+                "shape": tensor.shape.tolist(),
+                "type": tensor.tensor_type.name,
+                "offset": tensor.field.offset,
+            }
     json.dump(result, sys.stdout)